Customer Service Training Manual
IFTA Staff Training Part One
11th -12th October 2006
CUSTOMER SERVICE BASICS 3
INTRODUCTION TO CUSTOMER SERVICE 3
CUSTOMER SERVICE IN THE 21ST CENTURY 3
THE THREE KEY ELEMENTS 3
EXPAND YOUR DEFINITION OF SERVICE 3
WHO ARE YOUR CUSTOMERS? 3
DEVELOP A CUSTOMER FRIENDLY APPROACH 4
WHAT CUSTOMER SERVICE MEANS 4
CUSTOMER SERVICE QUALITIES 5
PROFESSIONAL QUALITIES IN CUSTOMER SERVICE 5
GOOD INFORMATION IS OFTEN GOOD SERVICE 6
SIMPLE ACTIONS HUGE RETURNS 7
CONVERSATIONS OVER THE TELEPHONE 8
IT’S NOT WHAT YOU SAY, IT’S HOW YOU SAY IT 8
TELEPHONE ETIQUETTE 8
ANSWERING THE TELEPHONE 8
TROUBLESHOOTING 9
RESPOND TO YOUR BUSINESS EMAIL QUICKLY! 10
SO WHAT IS PROMPT ANSWERING YOUR BUSINESS EMAIL? 10
FIRST IMPRESSIONS – YOU ONLY GET ONE 10
MAKING A GOOD FIRST IMPRESSION 10
TEN MAJOR DO’S AND DON’TS OF CUSTOMER SERVICE 11
HELPFUL REMINDERS FOR POLITE AND FRIENDLY RESPONSES 13
PRACTICE WHAT YOU PREACH – DEALING WITH THE CUSTOMER 14
COMMUNICATING WITH THE UNSATISFIED CUSTOMER 14
SOLVING THE CUSTOMER’S PROBLEMS 15
FOLLOW -UP WITH THE CUSTOMER 15
IT PAYS TO PLEASE 15
INITIATIVE 16
CUSTOMER SERVICE TRAITS TO COPY 16
BENEFITS OF GOOD CUSTOMER SERVICE 16
WHAT DO IFTA CUSTOMERS THINK AT THE MOMENT? 17
IMPROVING THE CUSTOMER SERVICE THAT IFTA PROVIDES 27
PACKING SPECIMENS 27
PACKING BUTTERFLIES 27
PACKING BEETLES 27
PACKING AN ORDER FOR A CUSTOMER 28
CITES 29
2
Customer Service Basics
Introduction to Customer Service
“There is only one boss, and whether a person shines shoes for a living or heads up the biggest corporation in the world, the boss remains the same. It is the customer! The customer is the person who pays everyone’s salary and who decides whether a business is going to succeed or fail. In fact, the customer can fire everybody in the company from the chairman (CEO) on down, and he can do it simply by spending his money somewhere else.
Literally everything we do, every concept perceived, every technology developed and associate employed, is directed with this one objective clearly in mind – pleasing the customer.”
Sam M. Walton, CEO Wal-Mart
Credo from Sam Walton the owner and CEO of Wal-Mart – an international chain of department stores and the most successful company in retailing in the world.
Customer Service in the 21st Century
Ask any CEO of a company, president of a bank, manager of an office, minister or staff person and they will tell you HOW IMPORTANT the customer is to their operations and success. In meeting after meeting, heads of industry, the service sector, utilities, and government try to convince the audience how much they believe in customer service.
“It is our mission, it is our number one priority, it is our goal, it is why we are in business, etc...,” often prove to be mere epitaphs. Unfortunately, these same “customer friendly” executives go back to their offices, de-employ office staff, fail to initiate a customer service improvement plan and send memos out saying customer complaints are unjustified and overblown.
The Three Key Elements
Expand Your Definition of Service
How you define service shapes every interaction you have with your customers. Limited definitions of service based on an exchange of monies for goods or service misses the overall point of customer service. “Service” should provide the customer with more than a product or action taken on his/her behalf. It should provide satisfaction. In essence, the customer should walk away pleased at the result of the transaction – not just content but actually happy. A happy customer will continue to be a buying customer and a returning customer.
Who are Your Customers?
Customers, buyers and clients want to pay a fair price for quality service or products, and feel satisfied they have paid for a service/product and received what they have paid for in return. They also want someone to take care of them. They need someone to understand their needs and help answer them. They need someone to hold their hands and walk them through a process. Customer service starts with the ability to listen to the customer and find out through polite questioning what he/she needs or wants.
Customer service and contact with a client mean that the customer will be heard and his/her problems will not go unanswered or ignored. It also means getting to know your client, his/her likes-dislikes, ideas, background, etc.
The other most important aspect to do is to listen to what the customer is saying. If people do not understand what is motivating the customer, they will not be successful in handling them. Do research on customers, their habits, and what they want and expect.
Most customer service is defined by how a company or organization treats “external customers,” but there is “internal customer service” as well. While this manual mainly addresses “external customers,” expanding your definition of customer service to include coworkers will lead toward even greater success. Remember, the internal customer chain is just like the external, we are all customers both inside and outside the company or organization. As a Wall Street Journal article succinctly put it, “Poorly Treated Employees Treat Customers Just as Poorly.”
Develop a Customer Friendly Approach
One commonality among all companies or organizations that provide good service is the development of a system and attitude promoting customer friendly service. By “customer friendly” we mean viewing the customer as the most important part of your job. The cliché, “The customer is always right” is derived from this customer friendly environment.
Two critical qualities to the “Customer Friendly Approach”:
* Communications
* Relationships
The two main tasks of successful customer relations are to communicate and develop relationships. They don’t take a huge effort, but don’t happen instantaneously either. Positive dialogue/communication with your customers and developing ongoing relationships wit h your customers are perhaps the two most important qualities to strive for in customer service.
What Customer Service Means
As mentioned earlier, customer service means providing a quality product or service that satisfies the needs/wants of a customer and keeps them coming back. Good customer service means much more – it means continued success, increased profits, higher job satisfaction, improved company or organization morale, better teamwork, and market expansion of services/products.
Think about it places where you enjoy doing business – stores, petrol stations, suppliers, banks, etc. Why, aside from the actual product or service they provide, do you like doing business with them? You probably find them courteous, timely, friendly, flexible, interested, and a series of other exemplary qualities. They not only satisfy your needs and help you in your endeavors but make you feel positive and satisfied. You come to rely on their level of service to meet your needs and wants.
On the other hand, let’s review a business you dislike patronizing maybe even hate utilizing but in some cases do so out of necessity. Maybe it is the Police when you need a new driver’s license or maybe it is the local store that carries a product you need but who offers lousy
service when you purchase. In both of these cases we are willing to hypothesize that the customer experience is marred by long lines, gruff service, inefficient processing, impolite and unfriendly clerks or salespeople, lack of flexibility, and no empathy for your customer plight. In these cases you feel abused, unsatisfied, and taken advantage of – in essence, your experience is wholly negative.
Unfortunately, in the cases we outlined above there is no competition for the services/products offered or you would gladly not consider using either the Ministry of Transport or the rude department store. This is the advantage of a monopoly on a good or service because in a competitive marketplace, the unsatisfied customer shops elsewhere.
Remember, good customer service results in consumer satisfaction and return customers and growth in business. Poor customer service, except for monopolistic strongholds, generally
results in consumer dissatisfaction, lack of returning customers and dwindling business.
Customer Service Qualities
Customer Service = Accountability + Delivery
Professional Qualities in Customer Service
Professionals who constantly deal with customers (inside and outside the company) need to strive for certain qualities to help them answer customer needs.
The professional qualities of customer service to be emphasized always relate to what the customer wants. After years of polling and market research, it turns out customers are constantly internalizing their customer service experience. What this means is they are grading your customer service during each transaction but you rarely know it. While there are a multitude of customer needs, six basics needs stand out:
* Friendliness – the most basic and associated with courtesy and politeness.
* Empathy – the customer needs to know that the service provider appreciates their wants and circumstances.
* Fairness – the customer wants to feel they receive adequate attention and reasonable answers.
* Control – the customer wants to feel his/her wants and input has influence on the outcome.
* Information – customers want to know about products and services but in a pertinent and time-sensitive manner.
It is also very important for customer service employees to have information about their product or service. Service providers who answer, “I don’t know” or “It is not my department” are automatically demeaned and demoted in the mind of the customer. These employees can end up feeling hostile as well as unequipped. Customers want information, and they disrespect and distrust the person who is supposed to have information but does not.
Good Information is Often Good Service
Employees need to be empowered to satisfy customers. Employees will give bad service to customers if they themselves receive bad service and little feedback from their managers and
supervisors. Remember: external customer service starts with internal customer service.
Simple Actions Huge Returns
* Customers will spend up to 10% more for the same product with better service.
* When customers receive good service they tell 10-12 people on average.
* When customers receive poor service they tell upwards of 20 people.
* There is an 82% chance customers will repurchase from a company where they were satisfied.
* There is a 91% chance that poor service will dissuade a customer from ever going back to a company.
It is often not what you articulate but how it is presented. What you wear and how you express yourself has a lot to do with how what you say is received.
Have you ever noticed how a person who is dressed-up, even in older or out-of-style clothing, always commands more authority and respect? The impression they make and what they have to say is enhanced by their personal presentation, facial and hand gestures, as well as the substance of what they have to say. As it turns out, substance is only part of the equation of being persuasive and influencing perception.
On one level this seems unfair and superficial because what a person says and how they behave should be more important than if they are well groomed, smiling and dressed-up. Yet visual perception plays a vital role in human impressions and reactions. For reasons psychologists do not always understand, nature and learned behavior have taught humans to perceive neat, smiling, well-presented individuals in a more commanding manner.
It is clear that just looking good will not produce the desired level of customer satisfaction.
* Smiling – there is nothing like a smile and pleasant face to greet a customer, especially if he/she has a complaint. A smile and polite conversation can immediately disarm a disgruntled customer. Facial expression sets a positive tone before you even begin speaking. A relaxed or pleasant facial expression is the ideal most of the time.
* Eye contact – always look into your customer’s eyes. Directly address customers.
* How you look – personal grooming has a big impact on your customers. Dirty hands, messy hair and poor dress can mean the loss of an otherwise happy customer. When interacting with customers, dress neatly and in a professional manner so as to command respect and to let customers know you take seriously your position.
* Shaking hands – when shaking hands with a customer a firm and professional handshake is expected. This part of the greeting is now common among both men and women in a professional environment.
* Be attentive - when listening to a customer, slightly lean towards your customer and nod your head ever so slightly to indicate you are listening.
* Tone of voice – always convey friendliness and amicability. Do not raise your voice in frustration or anger no matter how difficult or tiresome a customer may behave.
* Hand gestures - use hand movements to emphasize what you say (even on the phone) and to emphasize your feelings.
* Personal space – this is the distance that feels comfortable between you and another person. If another person approaches you and invades your personal space, you automatically move back without thought. You are uncomfortable. Leave adequate distance between you and your customer. Adequate space is important to making customers feel secure and unthreatened.
* Posture – slumping in a chair or leaning against a wall while interacting with a customer are sure signs you are not interested in the customer. Your pose or posture should express attention, friendliness, and openness. Lean forward, face the customer and nod to let them know you are interested.
* Observation - notice how your customer behaves and what he/she reacts positively to while you are providing service.
Remember, the little, interpersonal actions noted above mean a great deal in the area of customer relations. They can change customer perceptions and ultimately affect the success of your customer relations efforts.
Conversations Over the Telephone
It’s Not What You Say, It’s How You Say It
The moment you pick up a telephone, body language and visual perceptions disappear and your tone of voice becomes dominant.
Almost the entire message you project to the customer over the phone is derived from tone of voice and attitude.
For example:
* A flat tone of voice says to the customer, “I don’t like my job and would rather be elsewhere.”
* Slow pitch and presentation say, “I am sad and lonely – do not bother me.”
* A high pitch, rapid voice says, “I am enthusiastic and excited!’
* A loud voice says, “I’m angry and aggressive.”
Telephone Etiquette
Telephone etiquette, unlike more varying body language, can be uniform and is not culturally based. The telephone is often the first or last place a customer comes in contact with an organization or company. Being telephone friendly is one of the least expensive and cost-effective ways to deliver better customer service.
Answering the Telephone
How a company answers the phone can tell the whole story of how they treat customers and employees. The correct phrase said in the right order in a positive tone leaves a good impression and starts the customer-client relationship off on the right foot.
Pick up the phone in three rings. More than three rings signals chaos in your office or inattentiveness on the part of your company or organization.
Greet the caller, e.g. “hello”, “good morning”. Good manners shows you respect the caller.
Give your name, e.g., “Hi, my name is Emma”. This is a courtesy that serves to personalize the customer service experience as well as allowing the customer to hold you accountable for your level of service. He/she now has a point of reference and someone to contact when he/she calls back.
Ask the customer if or how you can help. Asking to help tells the customer you are there to serve his/her needs and to solve his/her problems. This also leaves the customer with a positive impression.
Put it altogether and you have a good example:
“Good morning, thanks for calling the Insect Farming and Trading Agency, my name is Emma, how may I help you?”
The greeting is key, it sets the tone and style of the whole interaction. Troubleshooting
Some things which may upset a customer are simply unavoidable. Here are some tips on how to best handle these situations.
“Putting a Customer on Hold”
Ask the customer if you can put them on hold; wait for them to say “yes” or “no” and then
explain it will only be for a short period of time.
Explain to customers why you are putting them on hold.
Thank customers for holding.
“Transferring a Call”
Ask the customer if they mind being transferred; wait for them to say “yes” or “no” and
explain why they are being transferred and to whom.
“Taking a Message”
Explain your co-workers absence in a positive light but do not be too specific. Explain that your co-worker is in a meeting, conference, briefing, or training. Do not say he or she is gravely ill, is too hung over to come to work, never called in today, can’t be found, that you do not know where he or she is, or that he or she “was just here”.
Give a reasonable estimate of when the co-worker will return.
Offer to help the caller, take a message or transfer to another staff member.
If a co-worker is on holiday and will not return to the office for some time, it is permissible to say that he or she is on holiday. However, avoid details such as, “Raymond is at the beach and I am sure he is having a great time.” While such details may seem innocuous and even humorous, they give the wrong impression to those seeking service.
“Ending the Call”
This is the final step in good telephone etiquette. A good customer service representative ends the call on a positive note, repeating any actions agreed to be taken and what is going to be done to help or serve the customer.
Respond to your business email quickly!
Answering your business email promptly should be a priority for all business. Not only is email an important communication line with your customers, it is often used by them to gauge that your trustworthy.
If a customer sends you an e-mail with a simple question, and you take forever to answer it, what does that say about the rest of your operation? It's one of the tell-tale signs customers use to seperate men from boys. And we all want to play with the big guys, don't we?
Talking about the big businesses, surveys show that the Top-500 fail miserably at answering their business email. Jupiter Communications reported that 42% took more than 5 days to answer a simple question. In the world of Internet, that might as well have been forever. If a customer has to wait that long for an answer, most likely she will have taken their business elsewhere. 35% of companies don't even bother to answer at all. I guess, they just don't like customers ;-) Forrester Research is reporting figures that are similar.
So what is prompt answering your business email?
Business email should be answered within 24 hours max. No exceptions. At that rate, your doing a lot better than a lot of other businesses.
If you really want your customer service to shine, you should consider answering your business email twice a day with a 12 hour interval.
It is even better to check out your direct competition by sending them an e-mail as if you are a potential customer. Send them more than one on several days. Especially check out mondays, fridays and weekends. Track the time it's taking them to answer, and implement a procedure to beat them at the business email game.
OK, I understand that for small businesses, resources are limited. But your stream of business email is most likely to be a lot less than for big guns. And if you check and answer e-mail regularly, numbers of e-mails to answer are usually very easy to handle.
First Impressions – You Only Get One
Making a Good First Impression
Every salesperson in every business knows the importance of making a positive first impression. Sales people know their success and livelihood will depend on how theirpotential customer perceives them in the first 30 seconds of interaction. Good salespeople develop an almost instantaneous rapport with potential customers. Customers like them, follow their advice and then buy their product.
The reality is that we prefer doing business with those we like and trust. Impressions are the
key to developing trust and confidence in the customer.
As the old saying goes, “You will never get a second chance to make a first impression.” This is why the first impression is extremely important and can set the tone for all future transactions.
Here are some ways of creating positive impressions, some of which have already been discussed:
* Thoughtfulness in meeting the customer’s needs
* Personal responsibility for a customer
* Quick problem solving for customer
* Offering immediate assistance
* Friendliness
* Using customer’s name in a conversation
* Pleasant voice tone
* Polite and courteous manners
* Neatness
* A genuine smile
Here are some factors that create a negative impression:
* Making the customer wait
* Not answering the phone promptly
* Not saying “please” and/or “thank you”
* Speaking loudly or condescendingly to customers or colleagues
* Making faces, frowning, acting distant, not smiling
* Looking disheveled or like you do not care about your appearance
* A poor handshake
* Focusing on another task while addressing or servicing a customer.
Remember, impressions stay with those you meet, especially customers, and once registered; negative impressions are difficult to overcome.
Ten Major Do’s and Don’ts of Customer Service
Every day customer service representatives face situations when what they say makes or breaks a service interaction. Below are ten phrases that should never be used because they frustrate and anger customers.
* “No.”
* “I don’t know.”
* “That’s not my job./That’s not my department.”
* “You are right – that is bad”
* “Calm down.”
* “I’m busy right now.”
* “Call me back.”
* “That’s not my fault.”
* “You need to talk to my supervisor.”
* “You want it by when?”
No: Everyone hates the word “no”. It is de -motivating, discouraging, and disinteresting. You will hear this word throughout your life as a customer and as a service provider. “No” is tantamount to “bad service.” “No” is easy, cheap, unproductive and negative – it means failure. Unfortunately, “no” is the word we most often hear when a new idea, request or concept is introduced. Admittedly, there are times when you will have to say “no,” but focus on what you can do for the customer (accentuate the positive) and not the negatives of the situation. Better to say “What I can do is...” and demonstrate that you care and want to provide quality service despite your current limitations.
I don’t know: Good service means never saying, “I don’t know.” When a customer hears “I don’t know,” they hear, “I don’t feel like finding the information you need.” Better to say, “I’ll find out” or “Let me look into this and get back to you ASAP.”
That’s not my job./That’s not my department: When a customer asks you to do something that you do not know how to do or do not have the authority to do, become a catalyst by leading the customer to the person or department who can help him/her solve the problem. Better to say, “Let me transfer to the person who ca n immediately help you will this problem.”
You’re right – that is bad: Many inexperienced customer service representatives think by sympathizing with the customer’s plight, he/she will win over the customer rather than actually doing something to solve the customer’s problem. If a customer expresses annoyance or frustration, do not make it worse by commiserating with him/her. Empathize with the customer but seek to solve the problem.
Likewise, it does not do your company or organization any good to criticize co-workers or other departments within the company or to the customers. All interested parties end up looking unprofessional and inept. Rather try your best to accommodate the customer. Do not promise anything you cannot deliver but do try to serve the customer well. Better to say, “I understand your frustration, let’s see how we can solve this problem.”
Calm down: When customers are upset or angry let them vent (within reason) and they will eventually calm down. Telling them to “calm down” is belittling, and often serves only to infuriate them further. Better to say, “I’m sorry.” This is one of the ideal phrases for customer service – it helps to placate the angriest of customers and allows you to begin the process of solving a customer complaint or request and “meet him/her half way.”
Apologizing does not mean you agree with the customer but it is a means to empathize and move beyond the emotion of the moment and negative impact.
I’m busy right now: It is not easy to juggle customers. You are often helping one customer when another calls or visits your service area. Asking a customer to be patient or politely asking them to wait is very different than putting them off and saying you are too busy to help. Leaving them standing there or on hold are two of the mortal sins of customer service.
“Being too busy” is tantamount to saying that you do not care and they are not important. Let the customer know they are important and you are aware of their presence. Better to say, “I’ll be with you in one moment” or “Please hold and I’ll be right with you.”
Call me back: This expression conveys little interest on the part of the customer relation’s employee for the needs and wants of the customer. You should always call the customer back because you want their business and are responsive to their requests. Being proactive is part of good customer service.
That’s not my fault: If an angry customer accuses you of creating a problem, rightly or wrongly, the natural reaction is to defend oneself. However, this is not the best course of action. The customer has a problem that needs to be solved. By resisting the need to defend yourself, and focusing on the needs of the customer, you can resolve the problem faster and with less stress and confrontation. Better to say, “Let’s see what we can do about this problem.”
You Need to Talk to My Supervisor: This cliché of bad customer service has angered and frustrated customers decades. Customers often ask for things outside the scope of your work or authority – maybe even outside the services/products provided by your company. While passing off these requests to your manager is a tempting option, it is better if you attempt to solve the problem yourself or directly go to the supervisor yourself and get a solution. You become a service hero for the customer and the supervisor. Better to say, “Let me find that out for you.”
You Want it by When?: Customers often make unrealistic demands, especially when it comes to time. Your first reaction may be annoyance and you may want to make a snide or sarcastic comment. However, the best approach is to hold off on displaying a negative attitude and making a poor impression. Better to say, “I will call you right back after I find out if that is feasible.”
Helpful Reminders for Polite and Friendly Responses
Wrong Approach
Polite and Friendly
Alternative
“I don’t know.”
“I’ll find out.”
“No.”
“What I can do is...”
“That’s not my job.”
“Let me find the right person who can help you with ...”
“You’re right – this is bad.”
“I understand your frustrations.”
“That’s not my fault.”
“Let’s see what we can do about this.”
“You want it by when?”
“I’ll try my best.”
“Calm down.”
“I’m sorry.”
“I’m busy right now.”
“I’ll be with you in just a
moment.”
“Call me back.”
“I will call you back, what is your telephone ber.”
Practice what you preach – Dealing with the customer Communicating with the Unsatisfied Customer
How many times have you as a customer run into the problem of excuses. There is a problem and the sales person, technician or customer service representative is making lame excuses, namely:
* It is the fault of the computer.
* It is the fault of the other sales clerk.
* It is the fault of the chief of the department.
* It is the fault of the system.
* It is the fault of the Government.
* It is just the way it is.
Sometimes it feels as if nothing is anybody’s fault or is in anybody’s department. This is poor customer service. Good customer service means acc ountability, responsibility and taking action to satisfy the customer.
Having discussed the importance of knowing how the customer feels and WHAT NOT TO SAY, let’s address the notion of how to communicate with an unsatisfied customer.
If your customer is unsatisfied (for just or unjust reasons), you will have to use some of the many techniques of the customer service professional to win their support and continued loyalty. When coming into contact with a customer, communicating with him/her, or analyzing problems, do not forget to use the following methods or qualities of the customer service professional:
Listen: It is of primary importance when dealing with an unsatisfied or complaining customer to listen attentively to his/her complaint, gripe, frustration or grievance. Be patient, attentive, and friendly.
Express you are sorry:
‘We are sorry for this mistake/problem.”
“We are terribly sorry for this inconvenience.”
“How can we work to solve this problem together?”
“I can imagine how frustrated you are.”
Do not argue and do not interrupt: This will only worsen the situation, especially if the
customer is angry. Let him speak before you try to discuss with him what has happened.
Do not lose your self-control: If you stay relaxed, customers will calm down.
Point out facts: Listen carefully – and write everything down. Do not make any comments until the customer is finished talking.
Admit the problem: If you can suggest a solution, do it. If not tell the customer what actions you will take and what actions will follow. Never make the mistake of promising something you are not able to do.
Involve the customer in problem solving: Suggest the customer alternative solutions, if
they exist. Customers appreciate the opportunity to choose the ways of problem solving.
Follow-up: Make sure that the promised measures are taken. If you do not fulfill what was promised and ignore the customer’s complaint, the problem will grow. Next time it will be more difficult to solve.
Give the customer a “way back”: Sometimes customers are wrong. You should let them leave with dignity, without feeling embarrassed.
Do not question the customer’s correctness: From the very beginning you should believe that the customer may be right. Always be open minded toward the customer’s opinion, make them feel they deserve to be listened to.
Solving the Customer’s Problems
When you listen to the customer’s complaint you take responsibility to solve the problem.
* Listen without interruption and with full attention.
* Behave without aggression, and without arguing.
* Do not extend excuses for the problem, and thank the customer for drawing their attention to it and helping solve it.
* Express sympathy and full understanding.
Customer service problem solving involves:
* Ask necessary questions to get more complete information and completed picture of a situation
* Find out exactly what the customer needs you to do for them
* Explain first what you can do, and then gently add what you cannot do
* Discuss in detail all opinions, and then decide what needs to be done
* Undertake immediately what was discussed
* Check the result to make sure the customer is completely satisfied
Follow -Up with the Customer
It pays to please
We like companies that treat us well, and some people will even pay more to obtain this.
Here are some recent statistics that prove the point:
* People spend up to 10 percent more for the same product with better service
* When people receive good service, on average, they tell 11 people
* When people receive poor service, on average, they tell up to 20 people
* There is an 80 percent chance that customers will repurchase from a company if their complaint is handled quickly and pleasantly
* If the service is really poor, 90 percent of customers won’t come back
It is extremely important to make sure that all customer service measures that were discussed or promised are in fact taken. It is not enough for the customer to experience a satisfactory telephone or face-to-face interaction. If nothing comes of the contact they will be even more frustrated and unhappy. Make sure you do whatever you have promised in a timely manner.
Initiative
Initiative is the difference between adequate customer service and customer service that wins you a customer for life.
Everyday examples of exceptional customer service:
* Taxi driver who opens the door for you or waits at night for you to safely get into your destination.
* Computer technician who does computer work and then calls back a week later to make sure your IT is functioning well.
* Car salesperson that calls a month after you buy a car to make sure it is running well.
* The petrol station attendant who washes your window or checks your oil.
* The electric company who calls and checks to make sure your service is working well and apologizes for any “brown outs” or “black outs.”
None of these customer service people HAD to make this extra effort or go to this trouble. These “goodwill initiatives” are beyond the call of duty and make the customer beyond satisfied. They make the customer remember the transaction or occasion.
Customer service traits to copy
* Be on time, open on time, deliver on time
* Follow through and deliver your promises
* Go the extra kilometer for customers
* Offer you customer options
* Express empathy to upset customers
* Treat customers as the MOST important part of your job
* Treat co-workers as if they are customers
* Give customers your name and contact details
Benefits of good customer service
Beneficiary Benefit
Higher income (more sales, repeat business, referred business)
Recognition
Personal satisfaction & fulfillment
Less stress
Higher self-awareness and self-control
Greater authenticity
Happier life at work
Happier life outside work
More repeat business
More referred business
Fewer returns
Better reputation
Higher morale, happier employees
Organizations Lower employee turnover
Fewer complaints
Higher productivity
Better work environment
Higher inventory turnover
Higher profits
What do IFTA customers think at the moment?
Here are some excerpts from the insectnet.com website where many IFTA customers talk about the customer service that they have received from you. Some are good but many show that you could be doing better.
Nipam Apr-04-03, 10:51 PM (PST)
Member since Jan-7-03
20 posts
"IFTA (New Guinea) - Good dealer report"
Hello,
I know a while back there was some discussion about the Insect Farming and Trading Agency in Papua New Guinea. I just completed a purchase from them and overall I was very happy with the purchase. A few specimens showed a bit more wear than I would have liked, and the shipping from New Guinea can be a bit slow.
On the whole, however, the specimens were excellent (especially the Ornithoptera), they all came with good collection data, and the prices were very good. David Whitaker at IFTA was especially helpful and rapid with his email communication.
I certainly plan on ordering from them again in the future and recommend them highly and without reservation.
Cheers,
Nipam
Nipam Apr-06-03, 08:38 AM (PST)
Member since Jan-7-03
20 posts
1. "RE: IFTA (New Guinea) - Good dealer report"
In response to message #0
Just a brief addition to my initial comments:
I've finished mounting more specimens and am really impressed with the quality of what was sent by IFTA. One or two specimens did show some wear, but most all the rest are in pristine condition. The birdwings are all perfect and have some of the most spectacular color that I have ever seen in these species (especially the O. caelestis and the O. priamus urvillanus)!
Also, regarding placing an order. Don't be put off by the fact that you will need to make a money transfer through your bank (IFTA doesn't currently use PayPal or take credit cards). IFTA provides very detailed information on how to set up the bank transfer. I just handed their email to the teller at my bank and they were able to get all the information they needed from it to set up the transfer in a few minutes. IFTA ships by DHL and this makes it easy for you to track the progress of the shipment. Mine stayed at LAX for a couple days, but theere was no problems with clearance with Customs and Fish and Wildlife as IFTA had taken care of all the needed CITES permits.
Once again thanks go to David Whittaker and IFTA for some great
specimens.
17
Nipam
entom May-25-05, 07:27 AM (PST)
Member since May-25-05
11 posts
"IFTA PNG"
Hello everybody,
do you know anything about IFTA from Bulolo PNG?
last year i made an order and agreed with David Whitaker to process with it in December 2004,
so that all the species i wanted were available.
i received last email in November 2004, and since then they do not reply to any emails.
if anyone has any news,please just let me know.
best regards
karol(poland) entom@op.pl
3. "RE: IFTA PNG"
In response to message #2
18
collector Jan-07-03, 05:36 PM (PST)
Charter Member
24 posts
"insect farm Papua New Guinea"
Is the "Insect Farming and Trading Agency of Papua New Guinea" a reliable supplier?
I would appreciate to hear from anyone who has ordered from them.
Thank You
prelson
Charter Member
14 posts
1. "RE: insect farm Papua New Guinea"
In response to message #0
The single transaction I had with them was somewhat slow, but everything that they promised to send they did and the condition was as indicated.
They did forget to attach the CITES permit original to the outside of the package, which made for a bit of a hassle on the "import side", so you might want to remind them of that.
I'm not sure that my bank had ever done a wire transfer to Papua New Guinea before, but it worked fine.
Peter
jasj Aug-08-06, 08:38 AM (PST)
Member since Jul-17-06
2 posts
"Is IFTA reliable?"
Hi
I made an order to IFTA in March. A few days later I recieved the invoice with the message not to pay until the CITES permits was OK. After about two months I asked for a status on my order. I sent the message several times before response. The status was that the administration who gives the permits was delaying the prosess. Over the last months I have made many requests without any response. Should I forget the whole thing and find the specimens elsewhere?
Jan
vytas Aug-09-06, 02:04 PM (PST)
Member since Oct-31-05
18 posts
1.
In
"RE: Is IFTA reliable?"
response to message #0
Hi Jan,
I tried to make trade with IFTA about a year ago. It took about 8 months of "communication" till
finally I stopped this nonsense. My recommendation - don't waste your time!
wish you luck
Vytautas Visinskas
ML Aug-09-06, 06:46 PM (PST)
Charter Member
151 posts
2. "RE: Is IFTA reliable?"
In response to message #1
Waste of time.
Michel Lauzon/Montreal, Canada
Michel.Lauzon@Gmail.com
I drank milk with Elvis last night.
Man did not land on the moon.
Bush is not a lying SOB.
cetoniinae Aug-09-06, 06:46 PM (PST)
Member since Feb-4-05
75 posts
3. "RE: Is IFTA reliable?"
In response to message #1
Sadly IFTA does not seem to respond any longer. The organization had a good start with a good ecological ethic, but I think it is poorly manned and underfunded nowadays. I have picked up some amazing specimens from them in the past but only when an American or European was at the helm and responses were more forthcoming - and even then, I had better luck when I phoned them.... I have no idea who is working with them now, but after three years of trying to get a response from them, I too have given up...
22
vytas Jan-10-06, 10:23 PM (PST)
Member since Oct-31-05
12 posts
"Something more about IFTA"
My adventure with IFTA from Papua New Guinea started in february 2005 when I made payment
by W.U. Now it is more than 10 month passed and today I took these money back. They did not
find time to take this payment. I do not understand this kind of business and do not recommend
to make any deals with IFTA to nobody.
Best regards to all
Vytautas Visinskas
Lithuania
papilio28570 Jan-25-06, 00:36 AM (PST)
Member since Nov-1-05
195 posts
1.
In
"RE: Something more about IFTA"
response to message #0
I had searched their web site and wondered how they could be in business selling so cheaply at the prices they had listed. I suspected scam right away.
Bob Cavanaugh
Agrias Jan-26-06, 01:11 PM (PST)
Member since Jan-20-06
3 posts
2. "RE: Something more about IFTA"
In response to message #1
Just to let you know I had a transaction with IFTA approximately 2 years ago that took a while (about 5 months) but it worked out fine. The specimens were packaged well and A1.
Sincerely,
Maurice Bottos
bgarthe Jan-27-06, 01:11 PM (PST)
Member since May-8-03
415 posts
3. "RE: Something more about IFTA"
In response to message #1
Yep be careful!!!!
1. A couple of years ago I received (as part of a group order) stuff that was simply BAD. Specimens were not nearly as promised. Did they replace or re-send?? No. I lost out on some $ and will not try it again
2. I actually tried to set up an order with several staff people there and for nearly a year,
24
nothing came to be. Luckily I did not send $.
3. I personally don't think it (IFTA) is a scam, but I do think that IFTA is about the most poorly run business in the lepidoptera arena Some of the garbage, no doubt, is due to the governmental "stuff" that has happened, but still I expect businesses to do business and not to jerk around potential customers and certainly fix issues with people who poured thousands of dollars into a transaction that "went south".
4. Actually, one train of thought on this a while back was that only the biggest of big buyers get any kind of service. This may be true. My group order was in the thousands, but we got screwed big time. Who knows? That is who knows (besides the people at IFTA)?
5. SIMPLY don't order from there!! There are plenty of great dealers with great prices and quality stuff out there.
Respectfully,
Bill Garthe
25
brainy88 Mar-28-04, 09:34 PM (PST)
Member since Mar-28-04
9 posts
"Question about Safe Collecting, and Collectors."
Hello, I wanted to purchase some insects from Insect-Sale.com but are they safe? But my main question is how good is IFTA? Insect FARMING trade AGENCY? are they really good, and on average how much does the shipping and handling cost? If anyone has done any business with them I would love to hear beacause I want to send them money, but I want to be sure that they are honest? ALSO VERNON EVAN?!?!? THIS JUST POPPED INTO MY HEAD. Is he reputable? or honest? Any responses would be great!!!
Happy Collecting!
Bizarro Mar-31-04, 06:08 PM (PST)
Charter Member
81 posts
1. "RE: Question about Safe Collecting, and Collectors."
In response to message #0
Hello!
I've done some business with IFTA a few years ago... everything went well. I still have some Delias and papered Ornithoptera from that package... with CITES certification.
They MUST have to be professionals and honest BECAUSE THEY'RE the oficial Papua New Guinea insect trading agency, the only one allowed to sell native insects abroad. In other words, they just simply can't afford scams... OK!
Actually they make more money from tourist visitation, than from the selling of butterflies because they care that the aborigen papuans that collect/breed the butterflies will keep 75% of the insect selling income (totally oposite to what happens in South America whose insects are bought very cheap and retailed with profits averaging +300%!!).
The major issue with IFTA dealings I think is to get atention from them if you're not a BIG CUSTOMER, or you do irregular, small orders. I think regular curtomers get in front... but this is something I've heard, not actually experienced. I've put a small order and I receive all items after sending the payment first.
Believe me, they are the pioneers of Butterfly Farming Biocomerce in the Tropical World ... just take a look at their website .
Hope it helped
Bizarro
26
Improving the customer service that IFTA provides
It is possible for you to regain the ground that you have lost with dissatisfied customers. This is not only through apologizing and reconciling with the people that have made comments on the insectnet website but through improving your day to day practices.
Remember that the survival of IFTA relies on having satisfied customers. Without them there will be no money coming in and not enough money to keep it a going concern.
Insect and butterfly grading
The correct grading of butterflies and other insects is very important and will make or break your reputation with your customers.
If customers receive an order of A1 specimens from you that they think are actually A2 they will not trust the service which you provide.
You must be very careful when grading insects and if you are not sure if a specimen is A1 then just make it an A2. That way when a customer receives an order from you they will be very happy.
Packing specimens
You may feel that you already know how to pack specimens but it never hurts to revisit your packing techniques to ensure that your customers will always be happy with you.
Packing butterflies
1. Kill live butterflies
2. Cut and fold A4 paper triangle
3. Label specimen (name and location of specimen)
e.g. O. goliath from Gumi, Watut, Morobe Province, Papua New Guinea
(do not use abbreviations for provinces)
4. Put a piece of folded toilet tissue around the specimen and place carefully into the paper triangle
5. Dry the specimen
6. Store the specimen in the appropriate storage cupboard
7. On receipt of an order for the specimen double check to see if the specimen is A1
8. If A1 replace the tissue paper with a fresh sheet
9. If it is a rare specimen double wrap it
Packing beetles
1. Kill the live beetle
2. Make card strips and cut plastic sheet to size
3. Place tissue paper on the card strip
4. Place beetle onto the tissue strip and secure plastic sheet with staples
e.g. eupholus spp. - pack singly (unless otherwise directed by customer)
R. straussi - pack in pairs
E. horridus – pack singly
5. Label the specimen
e.g. E.clarki from Lumi, Torricelli Mountains, Sandaun Province, Papua New Guinea
(do not use abbreviations for provinces)
6. Dry the specimen
7. After 7 days check that the specimen is fully dried
8. After drying store in the appropriate storage cupboard
9. On receipt of an order for the specimen double check to see if the specimen is A1
Packing an order for a customer
1. Check the specimens for their quality and then fumigate them
2. Use a strong box for the packing of an order
3. Place a layer of cotton wool into the box
4. If the order is for butterflies place one single layer of butterflies onto the first layer of cotton wool then add another layer of cotton wool and another single layer of butterflies repeat this until the box is full
5. If the order is for beetles place a single layer of beetles onto the cotton wool face up and then another layer of beetles face down so that the two layers fit snuggly. Then add another layer of cotton wool and another double layer of beetles until the box is full
6. Never over pack the box by pushing specimens down to fit more in as this will only ruin them and upset your customers
7. If you are packing a mixed order of beetles and butterflies into one box place the beetle layers into the box first followed by the butterfly layers
8. Sprinkle a very small quantity of crushed moth balls or naphthalene flakes onto each layer of cotton wool that you place in the box
9. Gently shake the packed box to make sure that none of the specimens are loose
10. Seal the box with sticky tape
11. Place this box into a larger box and surround it on all sides with shredded paper
12. Stick an envelope containing both export and import permits to the outside of the larger box
13. Dispatch the order with EMS
CITES
What is CITES?
CITES (the Convention on International Trade in Endangered Species of Wild Fauna and Flora) is an international agreement between governments. Its aim is to ensure that international trade in specimens of wild animals and plants does not threaten their survival.
Annually, international wildlife trade is estimated to be worth billions of dollars and to include hundreds of millions of plant and animal specimens. The trade is diverse, ranging from live animals and plants to a vast array of wildlife products derived from them, including food products, exotic leather goods, wooden musical instruments, timber, tourist curios and medicines. Levels of exploitation of some animal and plant species are high and the trade in them, together with other factors, such as habitat loss, is capable of heavily depleting their populations and even bringing some species close to extinction. Many wildlife species in trade are not endangered, but the existence of an agreement to ensure the sustainability of the trade is important in order to safeguard these resources for the future.
Because the trade in wild animals and plants crosses borders between countries, the effort to regulate it requires international cooperation to safeguard certain species from over-exploitation. CITES was conceived in the spirit of such cooperation. Today, it accords varying degrees of protection to more than 30,000 species of animals and plants, whether they are traded as live specimens, fur coats or dried herbs.
CITES was drafted as a result of a resolution adopted in 1963 at a meeting of members of IUCN (The World Conservation Union). The text of the Convention was finally agreed at a meeting of representatives of 80 countries in Washington DC., United States of America, on 3 March 1973, and on 1 July 1975 CITES entered in force.
CITES is an international agreement to which States (countries) adhere voluntarily. States that have agreed to be bound by the Convention ('joined' CITES) are known as Parties. Although CITES is legally binding on the Parties – in other words they have to implement the Convention – it does not take the place of national laws. Rather it provides a framework to be respected by each Party, which has to adopt its own domestic legislation to ensure that CITES is implemented at the national level.
For many years CITES has been among the conservation agreements with the largest membership, with now 169 Parties.
How many species are listed in the CITES Appendices?
Around 25,000 plant species and 5,000 animal species are covered by the provisions of the Convention, in the following proportions:
1) Appendix I: about 600 animal species and 300 plant species;
2) Appendix II: about 1,400 animal species and 25,000 plant species; and
3) Appendix III: about 270 animal species and 30 plant species.
How can I know whether I need a permit to import or export wildlife specimens?
Import, export and re-export of any live animal or plant of a species listed in the CITES Appendices (or of any part or derivative of such animal or plant) requires a permit or certificate.
How CITES works in Papua New Guinea
The Department of Environment and Conservation is the CITES management authority for Papua New Guinea. This is the government department that anyone who wants to export CITES listed species from Papua New Guinea has to go to in order to get permits.
CITES traded species exported from Papua New Guinea include crocodile skins, eaglewood, live reef fish and of course birdwing butterflies.
1. IFTA fills out CITES application form and it is sent to DEC.
2. DEC review the application and when approved it is stamped and signed by Wildlife Inspection Officer, the Deputy Secretary for Conservation (Dr. Guy Gowae), the Conservator of Fauna and Flora (Dr. Wari Iamo).
3. Approved CITES permit is sent back to IFTA.
4. Copy of permit made by IFTA and sent to country of import in order to receive import approval from customs.
5. A copy of the import permit sent back to IFTA from importing country.
6. Both the CITES export permit and copy of import permit are put into an envelope on the outside of the order before being sent to the customer via EMS.
QUICK GUIDE: Important Steps and Information
for Dissertation Completion and Graduation
All required forms bolded and italicized below and can be found in the Resource Course and on the Educational Services & Leadership Department Website here: http://www.rowan.edu/colleges/education/programs/edServLead/eduleadership/edd/index.html
1. Entering Dissertation Phase:
* Formation of your Dissertation Committee: Once you have finalized your committee, the following forms should be completed and submitted to the Program Coordinator as required for approval of your Dissertation Committee. Once submitted, notification of approval or any concerns is typically within 2 weeks:
* Formation of Doctoral Committee Form – signatures of chair and committee members required prior to submission
* Committee Rationale Form and Information – completed by student, along with all supporting documents included: rationale for each committee member, copy of CV or resume for each committee member (CVs not needed for any faculty in the Educational Services & Leadership Department), information on your dissertation topic (can put together an introduction or summary including all required information on your dissertation topic or submit a copy of your prospectus)
* MARCH 1: All students must submit committee approval forms and documentation by March 1 following the successful completion of Dissertation Seminar II.
* Registering for Dissertation Credits: Starting the semester after the completion of Dissertation Seminar II, students are eligible to begin registering for dissertation credits. The program requires a minimum of 12 dissertation credits and you must maintain continuous enrollment while in dissertation phase. If you do not complete after taking the minimum 12 credits, you will have to register and pay for additional credits. You will work with your Chair and the Program Coordinator/Advisor to complete your registration for the appropriate number of dissertation credits each term once you have completed all required coursework. You are required to be registered each fall and spring (summer is optional) until you are finished. When you begin work on your dissertation, you will need to email the Advisor/Program Coordinator at least 2 weeks prior to the start of each semester to request registration. Please include the following in dissertation credit registration requests:
* Banner ID
* Number of credits being requested
* Semester requesting credits for (i.e. fall 2016)
* Name of Dissertation Chair (Consulting with and/or notifying your dissertation chair prior to your request is recommended)
Educational Leadership Resources Course | Quick Guide | Updated 9-25-2015 1
STARTING OCTOBER 1, 2016 - REQUIRED REGISTRATION FOR DISSERTATION CREDITS: All active dissertation phase students (those who have successfully completed all 48 coursework credits) will be automatically enrolled in 1 dissertation credit each fall and spring semesters if registration is not requested before then to comply with continuous enrollment requirements. Automatic registration will occur during the second week of the semester if not requested beforehand.
IMPORTANT NOTES:
* Dissertation Committees must be approved prior to any proposal defense. Therefore, it is best practice not to begin work with your committee until all paperwork is submitted and approval is received.
* After a successful defense of your proposal, your committee should sign the Approval of the Dissertation Proposal Form. It is then your responsibility to make sure this signed form is submitted via the Educational Leadership Resource Course along with a copy of your proposal.
* Only AFTER your proposal has been approved can you then begin seeking IRB approval, collecting data, etc.
2. Ready for Symposium: When you and your chair have determined you are ready to schedule your symposium, you need to coordinate with the Department Secretary at minimum 2 weeks prior to the proposed date. A penultimate draft to all committee members must be sent 2 weeks prior to the final symposium as well. All symposiums must be done on campus.
3. After Your Symposium: Your committee will render a decision and provide you with any required changes. Signatures on the Dissertation Approval Form indicating you have passed the symposium presentation and then again once all revisions have been made to the satisfaction of the committee are required. Students must then follow the instructions provided by the Office of Graduate Studies at Rowan Global Learning & Partnerships for final dissertation approval which include submission of the following:
* Completed Dissertation Approval Form (minus Part C signature)
* Electronic Copy of final dissertation in accordance with all format guidelines
* $75 payment for document processing and archiving services (it is recommended you submit this payment in advance in anticipation of the completion of your dissertation to access the tutorial and submission area as required by Rowan Global’s Office of Graduate Studies)
Details on the format and submission of the dissertation to Rowan Global can be found here: http://rowanu.com/academic-resources/thesis-dissertation.
4. Graduation: To officially graduate and receive your conferred degree, a student must complete the Application for Graduation and have completed all required coursework, successfully written and presented the dissertation at symposium, submitted the dissertation and have final approval by the Office of Graduate Studies at Global Learning & Partnerships who then notifies the registrar’s
Educational Leadership Resources Course | Quick Guide | Updated 9-25-2015 2
office. PLEASE NOTE: This can take time as revisions are often required even after a successful symposium and typically 4 weeks is needed for review of the final dissertation document by the Office of Graduate Studies.
* To be eligible to participate in May graduation, all degree requirements and forms must be complete (i.e. the Dissertation Approval Form with parts A, B, and C signed/approved). In accordance with Graduate Policy, there is no walk through option for Ed.D. Students, meaning all degree requirements and approvals must be satisfied in order to participate. Therefore, below are deadlines to follow in order to be eligible to participate in Commencement and/or graduate in May.
* Graduation Timeline:
* Application for Graduation: Submitted through Banner Self-Service by the deadlines posted by the registrar: http://www.rowan.edu/provost/registrar/graduation.html#gradcert (typically by no later than the first 6 weeks of the semester in which you expect to complete/graduate)
* March 1: Final draft to all committee members and schedule your symposium
* April 1: Submission of all final and approved required documents and forms to the Office of the Graduate Studies as noted above
Questions?
Contact the Ed. D. Program Advisor/Coordinator:
Maria Lanza-Gladney
lanza-gladney@rowan.edu
856-256-4500 x3637
Educational Leadership Resources Course | Quick Guide | Updated 9-25-2015 3
ddddddddddddddddddddddddddINTERNATIONAL CONFERENCE ON HARMONISATION OF TECHNICAL REQUIREMENTS FOR REGISTRATION OF PHARMACEUTICALS FOR HUMAN USE
ICH HARMONISED TRIPARTITE GUIDELINE
POST-APPROVAL SAFETY DATA MANAGEMENT:
DEFINITIONS AND STANDARDS FOR EXPEDITED REPORTING
E2D
Current Step 4 version
dated 12 November 2003
This Guideline has been developed by the appropriate ICH Expert Working Group and has been subject to consultation by the regulatory parties, in accordance with the ICH Process. At Step 4 of the Process the final draft is recommended for adoption to the regulatory bodies of the European Union, Japan and USA.
E2D
Document History
First
Codification
History
Date
New
Codification
November
2005
E2D
Approval by the Steering Committee under Step 2 and release for public consultation.
18
July
2003
E2D
Current Step 4 version
E2D
Approval by the Steering Committee under Step 4 and recommendation for adoption to the three ICH regulatory bodies.
12
November
2003
E2D
POST-APPROVAL SAFETY DATA MANAGEMENT:
DEFINITIONS AND STANDARDS FOR EXPEDITED REPORTING
ICH Harmonised Tripartite Guideline
Having reached Step 4 of the ICH Process at the ICH Steering Committee meeting
on 12 November 2003, this guideline is recommended for
adoption to the three regulatory parties to ICH
TABLE OF CONTENTS
1. INTRODUCTION 1
2. DEFINITIONS AND TERMINOLOGY ASSOCIATED
WITH POST-APPROVAL DRUG SAFETY EXPERIENCE 1
2.1 Adverse Event (AE) 1
2.2 Adverse Drug Reaction (ADR) 1
2.3 Serious AE/ADR 1
2.4 Unexpected ADR 2
2.5 Healthcare Professional 2
2.6 Consumer 2
3. SOURCES Of INDIVIDUAL CASE SAFETY REPORTS 3
3.1 Unsolicited Sources 3
3.1.1 Spontaneous Reports 3
3.1.2 Literature 3
3.1.3 Internet 3
3.1.4 Other Sources 4
3.2 Solicited Sources 4
3.3 Contractual Agreements 4
3.4 Regulatory Authority Sources 4
4. STANDARDS FOR EXPEDITED REPORTING 5
4.1 What Should Be Reported? 5
4.1.1 Serious ADRs 5
4.1.2 Other Observations 5
4.1.2.1 Lack of Efficacy 5
i
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
4.1.2.2 Overdose 5
4.2 Minimum Criteria for Reporting 5
4.3 Reporting Time Frames 6
4.4 Non-serious ADRs 6
5. GOOD CASE MANAGEMENT PRACTICES 6
5.1 Assessing Patient and Reporter Identifiability1 6
5.2 The Role of Narratives 7
5.3 Clinical Case Evaluation 7
5.4 Follow-up Information 7
5.4.1 Pregnancy Exposure 8
5.5 How to Report 8
REFERENCES 9
ATTACHMENT 10
ii
POST-APPROVAL SAFETY DATA MANAGEMENT:
DEFINITIONS AND STANDARDS FOR EXPEDITED REPORTING
1. INTRODUCTION
It is important to establish an internationally standardized procedure in order to improve the quality of post-approval safety information and to harmonise the way of gathering and reporting information. The ICH E2A guideline provides guidance on pre-approval safety data management. Although many stakeholders have applied ICH E2A concepts to the post-approval phase, there is a need to provide further guidance on definitions and standards for post-approval expedited reporting, as well as good case management practices. This guideline is based on the content of ICH E2A guideline, with consideration as to how the terms and definitions can be applied in the post-approval phase of the product life cycle.
2. DEFINITIONS AND TERMINOLOGY ASSOCIATED WITH POST-APPROVAL DRUG SAFETY EXPERIENCE
2.1 Adverse Event (AE)
An adverse event is any untoward medical occurrence in a patient administered a medicinal product and which does not necessarily have to have a causal relationship with this treatment. An adverse event can therefore be any unfavorable and unintended sign (for example, an abnormal laboratory finding), symptom, or disease temporally associated with the use of a medicinal product, whether or not considered related to this medicinal product.
2.2 Adverse Drug Reaction (ADR)
Adverse drug reactions, as established by regional regulations, guidance, and practices, concern noxious and unintended responses to a medicinal product.
The phrase “responses to a medicinal product” means that a causal relationship between a medicinal product and an adverse event is at least a reasonable possibility (refer to the ICH E2A guideline).
A reaction, in contrast to an event, is characterized by the fact that a causal relationship between the drug and the occurrence is suspected. For regulatory reporting purposes, if an event is spontaneously reported, even if the relationship is unknown or unstated, it meets the definition of an adverse drug reaction.
2.3 Serious AE/ADR
In accordance with the ICH E2A guideline, a serious adverse event or reaction is any untoward medical occurrence that at any dose:
* results in death,
* is life-threatening
(NOTE: The term “life-threatening” in the definition of “serious” refers to an event/reaction in which the patient was at risk of death at the time of the event/reaction; it does not refer to an event/ reaction which hypothetically might have caused death if it were more severe),
* requires inpatient hospitalisation or results in prolongation of existing hospitalisation,
* results in persistent or significant disability/incapacity,
1
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
* is a congenital anomaly/birth defect,
* is a medically important event or reaction.
Medical and scientific judgment should be exercised in deciding whether other situations should be considered serious such as important medical events that might not be immediately life-threatening or result in death or hospitalisation but might jeopardise the patient or might require intervention to prevent one of the other outcomes listed in the definition above. Examples of such events are intensive treatment in an emergency room or at home for allergic bronchospasm, blood dyscrasias or convulsions that do not result in hospitalization, or development of drug dependency or drug abuse.
2.4 Unexpected ADR
An ADR whose nature, severity, specificity, or outcome is not consistent with the term or description used in the local/regional product labeling (e.g. Package Insert or Summary of Product Characteristics) should be considered unexpected. When a Marketing Authorisation Holder (MAH) is uncertain whether an ADR is expected or unexpected, the ADR should be treated as unexpected.
An expected ADR with a fatal outcome should be considered unexpected unless the local/regional product labeling specifically states that the ADR might be associated with a fatal outcome.
“Class ADRs” should not automatically be considered to be expected for the subject drug. “Class ADRs” should be considered expected only if described as specifically occurring with the product in the local/regional product labeling. This is illustrated in the following examples:
* “As with other drugs of this class, the following undesirable effect occurs with Drug X.”
* “Drugs of this class, including Drug X, can cause...”
If the ADR has not been documented with Drug X, statements such as the following are likely to appear in the local/regional product labeling:
* “Other drugs of this class are reported to cause...”
* “Drugs of this class are reported to cause..., but no reports have been received to date with Drug X.”
In these situations, the ADR should not be considered as expected for Drug X.
NOTE: The term “listedness” is not applicable to expedited reporting but should be used to characterize the ADR according to the Company Core Safety Information (refer to ICH E2C guideline for definitions).
2.5 Healthcare Professional
Healthcare professional is defined as a medically-qualified person such as a physician, dentist, pharmacist, nurse, coroner, or as otherwise specified by local regulations.
2.6 Consumer
Consumer is defined as a person who is not a healthcare professional such as a patient, lawyer, friend, or relative of a patient.
2
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
3. SOURCES Of INDIVIDUAL CASE SAFETY REPORTS
3.1 Unsolicited Sources
3.1.1 Spontaneous Reports
A spontaneous report is an unsolicited communication by a healthcare professional or consumer to a company, regulatory authority or other organization (e.g. WHO, Regional Center, Poison Control Center) that describes one or more adverse drug reactions in a patient who was given one or more medicinal products and that does not derive from a study or any organized data collection scheme.
Stimulated reporting can occur in certain situations, such as notification by a “Dear Healthcare Professional” letter, publication in the press, or questioning of healthcare professionals by company representatives. These reports should be considered spontaneous.
Consumer adverse reaction reports should be handled as spontaneous reports irrespective of any subsequent “medical confirmation”. Regulatory Authorities might require medical confirmation for the purpose of expedited reporting. Emphasis should be placed on the quality of the report and not on its source. Even if reports received from consumers do not qualify for regulatory reporting, the cases should be retained.
3.1.2 Literature
Each MAH is expected to regularly screen the worldwide scientific literature by accessing widely used systematic literature reviews or reference databases. The frequency of the literature searches should be according to local requirements or at least every two weeks. Cases of ADRs from the scientific and medical literature, including relevant published abstracts from meetings and draft manuscripts, might qualify for expedited reporting. A regulatory reporting form with relevant medical information should be provided for each identifiable patient. The publication reference(s) should be given as the report source; additionally a copy of the article might be requested by the local regulatory authority to accompany the report. All company offices are encouraged to be aware of publications in their local journals and to bring them to the attention of the company safety department as appropriate.
The regulatory reporting time clock starts as soon as the MAH has knowledge that the case meets minimum criteria for reportability.
If the product source, brand, or trade name is not specified, the MAH should assume that it was its product, although the report should indicate that the specific brand was not identified.
If multiple products are mentioned in the article, a report should be submitted only by the applicant whose product is suspected. The suspect product is that identified as such by the article's author.
3.1.3 Internet
MAHs should regularly screen websites under their management or responsibility for potential ADR case reports. MAHs are not expected to screen external websites for ADR information. However, if an MAH becomes aware of an adverse reaction on a website that it does not manage, the MAH should review the case and determine whether it should be reported.
3
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
MAHs should consider utilising their websites to facilitate ADR data collection, e.g., by providing ADR forms for reporting or by providing appropriate contact details for direct communication.
Unsolicited cases from the Internet should be handled as spontaneous reports. For the determination of reportability, the same criteria should be applied as for cases provided via other ways.
In relation to such cases from the Internet e.g. e-mail, identifiability of the reporter refers to the existence of a real person, i.e., it is possible to verify that the patient and the reporter exist.
3.1.4 Other Sources
If an MAH becomes aware of a case report from non-medical sources, e.g. the lay press or other media, it should be handled as a spontaneous report. For the determination of reportability, the same criteria should be applied as for other reports.
3.2 Solicited Sources
Solicited reports are those derived from organized data collection systems, which include clinical trials, registries, post-approval named patient use programs, other patient support and disease management programs, surveys of patients or healthcare providers, or information gathering on efficacy or patient compliance. Adverse event reports obtained from any of these should not be considered spontaneous.
For the purposes of safety reporting, solicited reports should be classified as study reports, and therefore should have an appropriate causality assessment by a healthcare professional or an MAH. Further guidance on study-related issues, such as managing blinded therapy cases, can be found in the ICH E2A guideline.
3.3 Contractual Agreements
The marketing of many medicines increasingly takes place through contractual agreements between two or more companies, which may market same product in the same or different countries/region. Arrangements vary considerably with respect to inter-company communication and regulatory responsibilities. Overall, this can be a complex issue.
In such relationships, it is very important that explicit licensing/contractual agreements specify the processes for exchange of safety information, including timelines and regulatory reporting responsibilities. Safety personnel should be involved in development of any agreements from the beginning. Processes should be in place to avoid duplicate reporting to the regulatory authority, e.g. assigning responsibility to one company for literature screening.
Whatever the nature of the arrangement, the MAH is ultimately responsible for regulatory reporting. Therefore, every reasonable effort should be made between the contracting partners to minimize the data exchange period so as to promote compliance with MAH responsibilities.
3.4 Regulatory Authority Sources
Individual serious unexpected adverse drug reaction reports originating from foreign regulatory authorities are subject to expedited reporting to other authorities by each MAH. Re-submission of serious ADR cases without new information to the
4
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
originating regulatory authority is not usually necessary, unless otherwise specified by local regulation.
4. STANDARDS FOR EXPEDITED REPORTING
4.1 What Should Be Reported?
4.1.1 Serious ADRs
Cases of adverse drug reactions that are both serious and unexpected are subject to expedited reporting. The reporting of serious expected reactions in an expedited manner varies among countries. Non-serious adverse reactions, whether expected or not, would normally not be subject to expedited reporting.
For reports from studies and other solicited sources, all cases judged by either the reporting healthcare professional or the MAH as having a possible causal relationship to the medicinal product would qualify as ADRs. For purposes of reporting, spontaneous reports associated with approved drugs imply a suspected causal relationship.
4.1.2 Other Observations
In addition to single case reports, any safety information from other observations that could change the risk-benefit evaluation for the product should be communicated as soon as possible to the regulatory authorities in accordance with local regulation. Examples include any significant unanticipated safety findings from an in vitro, animal, epidemiological, or clinical study that suggest a significant human risk, such as evidence of mutagenicity, teratogenicity, carcinogenicity, or lack of efficacy with a drug used in treating a life-threatening or serious disease.
4.1.2.1 Lack of Efficacy
Evidence of lack of efficacy should not normally be expedited, but should be discussed in the relevant periodic safety update report. However, in certain circumstances and in some regions, individual reports of lack of efficacy are considered subject to expedited reporting. Medicinal products used for the treatment of life-threatening or serious diseases, vaccines, and contraceptives are examples of classes of medicinal products where lack of efficacy should be considered for expedited reporting. Clinical judgment should be used in reporting, with consideration of the local product labeling and disease being treated.
4.1.2.2 Overdose
Reports of overdose with no associated adverse outcome should not be reported as adverse reactions. Cases associated with serious adverse reactions are considered subject to expedited reporting, unless otherwise specified by local regulation. They should be routinely followed up to ensure that the information is as complete as possible with regard to symptoms, treatment, and outcome. The MAH should collect any available information on overdose related to its products.
4.2 Minimum Criteria for Reporting
It is recommended that as much information as possible be collected at the time of the initial report. However, for the purpose of regulatory reporting, the minimum data elements for an ADR case are: an identifiable reporter, an identifiable patient, an adverse reaction, and a suspect product. Lack of any of these four elements means
5
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
that the case is considered incomplete; however, MAHs are expected to exercise due diligence to collect the missing data elements.
4.3 Reporting Time Frames
In general, expedited reporting of serious and unexpected ADRs is required as soon as possible, but in no case later than 15 calendar days of initial receipt of the information by the MAH. Time frames for other types of serious reports vary among countries, depending on source, expectedness and outcome.
The regulatory reporting time clock is considered to start on the date when any personnel of the MAH first receive a case report that fulfills minimum criteria as well as the criteria for expedited reporting. In general, this date should be considered day 0.
When additional medically relevant information is received for a previously reported case, the reporting time clock is considered to begin again for submission of the follow-up report. In addition, a case initially classified as a non-expedited report, would qualify for expedited reporting upon receipt of follow-up information that indicates the case should be re-classified (e.g., from non serious to serious).
4.4 Non-serious ADRs
Cases of non-serious ADRs, whether expected or not, would not normally be considered reportable on an expedited basis. Non-serious ADRs should be included in the periodic safety update report according to the ICH E2C guideline.
5. GOOD CASE MANAGEMENT PRACTICES
Accurate, complete, and bona fide information is very important for MAHs and regulatory agencies identifying and assessing ADR reports. Both are faced with the task of acquiring sufficient information to help ensure that the reports are authentic, accurate, as complete as possible, and non-duplicative.
5.1 Assessing Patient and Reporter Identifiability1
Patient and reporter identifiability is important to avoid case duplication, detect fraud, and facilitate follow-up of appropriate cases. The term identifiable in this context refers to the verification of the existence of a patient and a reporter.
Local data privacy laws regarding patient and reporter identifiability might apply.
One or more of the following should automatically qualify a patient as identifiable: age (or age category, e.g., adolescent, adult, elderly), gender, initials, date of birth, name, or patient identification number. In addition, in the event of second-hand reports, every reasonable effort should be made to verify the existence of an identifiable patient and reporter.
All parties supplying case information or approached for case information should be identifiable: not only the initial reporter (the initial contact for the case), but also others supplying information.
In the absence of qualifying descriptors, a report referring to a definite number of patients should not be regarded as a case until the minimum four criteria for case reporting are met. For example, “Two patients experienced...” or “ a few patients experienced” should be followed up for patient-identifiable information before regulatory reporting.
6
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
5.2 The Role of Narratives
The objective of the narrative is to summarize all relevant clinical and related information, including patient characteristics, therapy details, medical history, clinical course of the event(s), diagnosis, and ADR(s) including the outcome, laboratory evidence (including normal ranges), and any other information that supports or refutes an ADR. The narrative should serve as a comprehensive, standalone “medical story”. The information should be presented in a logical time sequence; ideally this should be presented in the chronology of the patient’s experience, rather than in the chronology in which the information was received. In follow-up reports, new information should be clearly identified.
Abbreviations and acronyms should be avoided, with the possible exception of laboratory parameters and units. Key information from supplementary records should be included in the report, and their availability should be mentioned in the narrative and supplied on request. Any relevant autopsy or post-mortem findings should also be summarized in the narrative and related documents should be provided according to local regulation and if allowed by local data privacy laws.
Terms (e.g., AEs/ADRs, indication, and medical conditions) in the narrative should be accurately reflected in appropriate data fields.
5.3 Clinical Case Evaluation
The purpose of careful medical review is to ensure correct interpretation of medical information. Preferably, information about the case should be collected from the healthcare professionals who are directly involved in the patient’s care. Regardless of the source of an ADR report, the recipient should carefully review the report for the quality and completeness of the medical information. The review should include, but is not limited to, the following considerations:
* Is a diagnosis possible?
* Have the relevant diagnostic procedures been performed?
* Were alternative causes of the reaction(s) considered?
* What additional information is needed?
ADR terms should be used consistently and in accordance with recommended standards for diagnosis, if possible. The report should include the verbatim term as used by the reporter, or an accurate translation of it. Any company personnel receiving reports should provide an unbiased and unfiltered report of the information from the reporter. While the report recipient is encouraged to actively query the reporter to elicit the most complete account possible, inferences and imputations should be avoided in report submission. However, clearly identified evaluations by the MAH are considered appropriate and are required by some regulatory authorities.
When a case is reported by a consumer, his/her description of the event should be retained, although confirmatory or additional information from any relevant healthcare professionals should also be sought and included.
5.4 Follow-up Information
The information from ADR cases when first received is generally incomplete. Ideally, comprehensive information would be available on all cases, but in practice efforts
7
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
should be made to seek additional information on selected reports, including second – hand reports (see Attachment, Recommended Key Data Elements, of this guideline).
In any scheme to optimize the value of follow-up, the first consideration should be prioritization of case reports by importance. The priority for follow-up should be as follows: cases which are 1) serious and unexpected, 2) serious and expected, and 3) non-serious and unexpected. In addition to seriousness and expectedness as criteria, cases “of special interest” also deserve extra attention as a high priority (e.g., ADRs under active surveillance at the request of the regulators), as well as any cases that might lead to a labeling change decision.
Follow-up information should be obtained, via a telephone call and/or site visit and/or a written request. The company should provide specific questions it would like to have answered. Follow-up methods should be tailored towards optimizing the collection of missing information. Written confirmation of details given verbally should be obtained whenever possible. In exceptional circumstances, if requests for information have been refused by the reporter, a regulatory authority might be able to assist an MAH in obtaining follow-up data.
To facilitate the capture of clinically relevant and complete information, use of a targeted questionnaire/specific form is encouraged, preferably at the time of the initial report. Ideally, healthcare professionals with thorough pharmacovigilance training and therapeutic expertise should be involved in the collection and the direct follow-up of reported cases (particularly those of medical significance). For serious ADRs, it is important to continue follow-up and report new information until the outcome has been established or the condition is stabilized. How long to follow up such cases is a matter of judgment.
It is important that at the time of the original report, sufficient details about the patient and reporter be collected and retained to enable future investigations, within the constraints imposed by local data privacy laws.
5.4.1 Pregnancy Exposure
MAHs are expected to follow up all pregnancy reports from healthcare professionals or consumers where the embryo/foetus could have been exposed to one of its medicinal products. When an active substance, or one of its metabolites, has a long half-life, this should be taken into account when considering whether a foetus could have been exposed (e.g., if medicinal products taken before the gestational period should be considered).
5.5 How to Report
The CIOMS I form has been a widely accepted standard for expedited adverse event reporting. However, no matter what the form or format used, it is important that certain basic information/data elements, when available, be included with any expedited report, whether in a tabular or narrative presentation. It is recommended that the Medical Dictionary for Regulatory Activities (MedDRA) be used for coding medical information. The standards for electronic submission of Individual Case Safety Reports (ICSRs), according to the ICH E2B/M2 guidelines, should be implemented.
The listing in the Attachment of this guideline addresses those data elements regarded as desirable; if all relevant elements are not available at the time of expedited reporting, efforts should be made to obtain them.
8
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
REFERENCES
1. Current Challenges in Pharmacovigilance: Pragmatic Approaches (Report of CIOMS Working V), Geneva 2001
2. Rules Governing Medicinal Products in the European Union, Volume 9, PHARMACOVIGILANCE: Medicinal Products for Human Use, 2001 http://pharmacos.eudra.org/F2/eudralex/vol-9/home.htm
3. Guidance for Industry: Postmarketing Safety Reporting for Human Drug and Biological Products Including Vaccines, Food and Drug Administration, March 2001 (draft) http://www.fda.gov/cder/guidance/4153dft.pdf
4. Safety Reporting Requirements for Human Drug and Biological Products, Proposed Rule, Food and Drug Administration, March 2003
5. Notification No 421 on Enforcement of the Law Revising Partially the Pharmaceutical Affairs Law, the Director General, Pharmaceutical Affairs Bureau, Ministry of Health and Welfare, March 1997
9
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
ATTACHMENT
RECOMMENDED KEY DATA ELEMENTS FOR INCLUSION
IN EXPEDITED REPORTS
OF SERIOUS ADVERSE DRUG REACTIONS
Some data elements might not be relevant, depending on the circumstances. Attempts should be made to obtain follow-up information on as many other listed items as are pertinent to the case. Refer to the ICH E2B/M2 guidelines for detailed data elements for electronic transmission of ICSRs.
1. Patient Details
* Initials
* Other relevant identifier (patient number, for example)
* Gender
* Age, age category (e.g., adolescent, adult, elderly), or date of birth
* Concomitant conditions
* Medical history
* Relevant family history
2. Suspected Medicinal Product(s)
* Brand name as reported
* International Non-Proprietary Name (INN)
* Batch/lot number
* Indication(s) for which suspect medicinal product was prescribed or tested
* Dosage form and strength
* Daily dose (specify units - e.g., mg, ml, mg/kg) and regimen
* Route of administration
* Starting date and time
* Stopping date and time, or duration of treatment
3. Other Treatment(s)
The same information as in item 2 should be provided for the following:
* Concomitant medicinal products
(including non-prescription, over-the-counter medicinal products, herbal remedies, dietary supplements, complementary and alternative therapies, etc.)
* Relevant medical devices
Post-Approval Safety Data Management: Definitions and Standards for Expedited Reporting
4. Details (all available) of Adverse Drug Reaction(s)
* Full description of reaction(s), including body site and severity
* The criterion (or criteria) for regarding the report as serious
* Description of the reported signs and symptoms
* Specific diagnosis for the reaction
* Onset date (and time) of reaction
* Stop date (and time) or duration of reaction
* Dechallenge and rechallenge information
* Relevant diagnostic test results and laboratory data
* Setting (e.g., hospital, out-patient clinic, home, nursing home)
* Outcome (recovery and any sequelae)
* For a fatal outcome, stated cause of death
* Relevant autopsy or post-mortem findings
* Relatedness of product to reaction(s)/event(s)
5. Details on Reporter of an ADR
* Name
* Mailing address
* Electronic mail address
* Telephone and/or facsimile number
* Reporter type (consumer, healthcare professional, etc.)
* Profession (specialty)
6. Administrative and MAH Details
* Source of report (spontaneous, epidemiological study, patient survey, literature, etc.)
* Date the event report was first received by manufacturer/company
* Country in which the event occurred
* Type (initial or follow-up) and sequence (first, second, etc.) of case information reported to authorities
* Name and address of MAH
* Name, address, electronic mail address, telephone number, and facsimile number of contact person of MAH
* Identifying regulatory code or number for marketing authorisation dossier
* Company/manufacturer's identification number for the case (the same number should be used for the initial and follow-up reports on the same case).
11
Marketing
Component Guidelines
2015-2016
Version 3.2 | Published October 15, 2015
Eligible Activities
1. Funding under the Marketing Component can be used to support a range of activities in the marketing and promotion of a Qualifying Album. FACTOR-funded full-length sound recordings are by definition Qualifying Albums.
2. To be considered a Qualifying Album, the sound recording must:
* Be performed by a Canadian Artist and meet FACTOR’s MAPL criteria for Canadian content (see Business Policies: MAPL Certification);
* Be a full-length album of at least six tracks, or include at least 20 minutes of recorded material;
* Be comprised of all new, previously unreleased sound recording masters;
* Have no more than 50% French lyrics. The lyrics may otherwise be in English or any language other than French. French-language sound recordings may be supported by Musicaction.
3. Typical marketing activities funded under this Component include (but are not limited to):
* Advertising and publicity campaigns in print, TV, radio and online media;
* Digital and social media marketing initiatives;
* Subscriptions to SoundScan;
* Promotional appearances by the artist;
* Production of promotional videos and EPKs;
* Production expenses related to an album release party (excluding hospitality costs).
4. Non-promotional live appearances and the production of music videos must be funded under the Tour Support, Showcase Support, and Video Components.
5. If you are claiming commercial radio advertising and promotion costs, you may also be eligible for additional funding under the Radio Marketing Fund. Please see the guidelines for details.
Marketing Component Guidelines | Version 3.2 | Published October 15, 2015 1
Funding Limits
6. The funding limit for the Marketing Component varies by each Program. Please see the Program Guidelines for the Program you are applying under.
Eligible Costs - Marketing
7. The Eligible Costs in this section are specific to the Marketing Component. Please also review Eligible Costs – General Terms as well as the Eligible Costs – This Program in the Program Guidelines for the Program under which you are applying. Applicants are strongly advised to read and understand these sections, and to contact FACTOR prior to submitting an Application to ensure eligibility of the proposed costs.
8. FACTOR funding under the Marketing Component will cover either 50% or 75% of the Total Eligible Budget, to the Program maximum. Please see the Program Guidelines for the Program under which you are applying.
9. FACTOR will only reimburse the costs which it deems Eligible Costs and will remove ineligible costs from the budget. The FACTOR contribution may be reduced accordingly. If you have any questions about the eligibility of a particular cost, please contact FACTOR in advance.
10. Marketing funding can be used to pay for the cost of marketing initiatives that are specific to the Artist and approved Qualifying Album. Eligible costs can include, for example, publicity, radio tracking, print, radio and TV advertising, and digital and social marketing initiatives. In addition:
a) International costs are allowable to a maximum of 50% of the total Eligible Costs.
b) In-House and Related Party costs are allowable to a maximum of 25% of the total Eligible Costs. Please see Business Policies: Eligible Costs – General Terms for additional details regarding In-House and Related Party costs.
c) Printed materials including posters, handbills, and banners are eligible without limit as to number of units or cost.
d) The cost of manufacturing or purchasing Promotional CDs, vinyl LPs, dropcards, USB keys or any other sound recording media format will be allowed at a maximum aggregate number of 500 units. Manufacturing receipts or account statements from the label or distributor showing the cost must be provided at Completion to claim promotional items.
e) Other promotional merchandise, such as T-shirts, buttons and other soft goods, will be allowed at a maximum aggregate charge of $500. Manufacturing or purchase receipts showing the cost must be provided at Completion to claim all promotional items.
f) Development of significant new Artist-specific web assets (e.g. new website development, addition of e-commerce portal)
g) Ongoing web maintenance costs (such as page re-design, updating content, programming, domain and hosting fees) that are Artist-specific will be recognized at a maximum of $1,000 per month.
h) A SoundScan subscription is an Eligible Cost as long as it is Artist-specific (such as title reports, venue settlement).
i) Lyric Videos are eligible for funding only under the Marketing Component. There is no limit to the number of Lyric Videos that can be funded; however, the budget per Lyric Video will be capped at $500.
j) Promotional videos (such as documentaries, EPK material) are eligible but the budget will be capped to a maximum FACTOR contribution of $1000.
k) Costs incurred by a third-party distributor and charged back to the Recipient’s account may be eligible, provided that the Recipient submits (a) an itemized list of expenses incurred, along with a detailed accounting of the charge-backs (supplier name, invoice date, invoice number, dollar amount, purpose of item) and (b) if requested by FACTOR, all of the required delivery materials and supporting documentation that would be demanded of the Recipient as if it had incurred those costs out-of-pocket, such as invoices, receipts, and proofs of payment.
Marketing Component Guidelines | Version 3.2 | Published October 15, 2015 2
11. Ineligible costs include:
* Costs which are not Artist-specific; such as general company wages, services, supplies, and any other company overhead core cost or expense.
* Any costs for initiatives previously claimed within another FACTOR supported project.
* Contest prizes, unless they meet the definition of “promotional merchandise” set out above.
* Costs associated with applications for prize or award nominations.
* Donated Services.
12. Administration Fees for the Marketing Component are calculated automatically by the online system, as follows:
* Comprehensive Artist Program: 15% of total Eligible Costs to a maximum of $5,000
* Comprehensive Music Company Program: 15% of total Eligible Costs to a maximum of $5,000
* Juried Sound Recording Program: 15% of total Eligible Costs to a maximum of $3,000
* Marketing and Promotion for FACTOR-Funded Sound Recordings Program:
o 15% of total Eligible Costs to a maximum of $5,000 for sound recordings funded under a Program where FACTOR’s contribution is up to 50% of Total Eligible Budget
o 15% of total Eligible Costs to a maximum of $3,000 for sound recordings funded under a Program where FACTOR’s contribution is up to 75% of Total Eligible Budget
* Marketing and Promotion for Non-FACTOR-Funded Sound Recordings Program: 15% of total Eligible Costs to a maximum of $5,000
Component Completion
13. In addition to the Program completion requirements, to complete the Marketing Component the Applicant must:
* Provide FACTOR with verification of a confirmed Commercial Release Date;
* Complete the online Completion Budget; and
* Submit Letters of Direction to Pay (maximum three per Project). These Letters are voluntary on the part of the Applicant, who may prefer to have FACTOR pay certain suppliers directly.
Marketing Component Guidelines | Version 3.2 | Published October 15, 2015 3
How to Make a Preparation Outline
Public Speaking Center * www.uwlax.edu/CommStudies/PSC * 251 Murphy Library
MATLAB
BEGINNER’S
GUIDE
0
About MATLAB
MATLAB is an interactive software which has been used recently in various areas of engineering and scientific applications. It is not a computer language in the normal sense but it does most of the work of a computer language. Writing a computer code is not a straightforward job, typically boring and time consuming for beginners. One attractive aspect of MATLAB is that it is relatively easy to learn. It is written on an intuitive basis and it does not require in-depth knowledge of operational principles of computer programming like compiling and linking in most other programming languages. This could be regarded as a disadvantage since it prevents users from understanding the basic principles in computer programming. The interactive mode of MATLAB may reduce computational speed in some applications.
The power of MATLAB is represented by the length and simplicity of the code. For example, one page of MATLAB code may be equivalent to many pages of other computer language source codes. Numerical calculation in MATLAB uses collections of well-written scientific/mathematical subroutines such as LINPACK and EISPACK. MATLAB provides Graphical User Interface (GUI) as well as three-dimensional graphical animation.
In general, MATLAB is a useful tool for vector and matrix manipulations. Since the majority of the engineering systems are represented by matrix and vector equations, we can relieve our workload to a significant extent by using MATLAB. The finite element method is a well-defined candidate for which MATLAB can be very useful as a solution tool. Matrix and vector manipulations are essential parts in the method. MATLAB provides a help menu so that we can type the help command when we need help to figure out a command. The help utility is quite convenient for both beginners and experts.
1
Vector and Matrix Manipulations
Once we get into MATLAB, we meet a prompt » called the MATLAB prompt. This prompt receives a user command and processes it providing the output on the next line. Let us try the following command to define a matrix.
>> A= [1 ,3,6;2,7,8;0,3,9]
Then the output appears in the next line as shown below.
A =
1
3
6
2
7
8
0
3
9
Thus, a matrix is entered row by row, and each row is separated by the semicolon(;). Within each row, elements are separated by a space or a comma(,). Commands and variables used in MATLAB are case-sensitive. That is, lower case letters are distinguished from upper case letters. The size of the matrix is checked with
>> size(A)
ans = 3 3
Transpose of a matrix In order to find the transpose of matrix A, we type
>> A'
The result is
1 2 0
ans = 3 7 3
6 8 9
Column or row components MATLAB provides columnwise or rowwise operation of a matrix. The following expression
>> A(:,3)
yields
ans =
2
6
8
9
which is the third column of matrix A. In addition,
>>A(:,1)
represents the first row of A as
ans = 1 3 6
We can also try
>> A(:,1)+A(:,3)
as addition of the first and third rows of A with the result
ans= 1 6 15
Now let us introduce another matrix B as
>> B = [3,4,5; 6,7,2:8,1,0];
Then there seems to be no output on the screen. MATLAB does not prompt output
on the screen when an operation ends with the semicolon (;).
If we want to check the B matrix again, we simply type
>>B The screen output will be
345
B = 6 7 2
8 1 0
Matrix addition Adding two matrices is straightforward like
>> C = A + B
4
7
11
C = 8
14
10
8
4
9
Matrix subtraction In order to subtract matrix B from matrix A, we type
3
>> C = A-B
-2 -1 1
C = -4 0 6
- 8 2 9
Note that C is now a new matrix, not the summation of A and B anymore.
Matrix multiplication Similarly, matrix multiplication can be done as
>> C = A*B
69 31 11 C= 112 65 24 90 30 6 Matrix Functions
Manipulation of matrices is a key feature of the MATLAB functions. MATLAB is a useful tool for matrix and vector manipulations. Collections of representative MATLAB matrix functions are listed in Table 1. Examples and detailed explanations are provided for each function below.
Table 1. Basic Matrix Functions
Symbol Explanations
inv inverse of a matrix
det determinant of a matrix
rank rank of a matrix
cond condition number of a matrix
eye(n) the n by n identity matrix
zeros(n,m) the n by m matrix consisting of all zeros
4
Matrix inverse The inverse of a matrix is as simple as
>> inv(A)
ans=
1.8571 -0.4286 -0.8571
-0.8571 0.4286 0.1905
0.2857 -0.1429 0.0476
In order to verify the answer, we can try
>> A*inv(A);
which should be a 3 by 3 identity matrix.
Determinant of a matrix
>> d =det(A)
produces the determinant of the matrix A. That is, d = 21
Rank of a matrix The rank of a matrix A, which is the number of independent rows or columns, is obtained from
>> rank(A);
Identity matrix
>> eye(3)
yields
1
0
0
ans = 0
1
0
0
0
1
eye(n) produces an identity matrix of size n by n. This command is useful when we initialize a matrix.
Matrix of random numbers A matrix consisting of random numbers can be generated using the following MATLAB function.
>> rand(3,3)
0.2190 0.6793 0.5194
5
ans = 0.0470 0.9347 0.8310
0.6789 0.3835 0.0346
That is, rand(3,3) produces a 3 by 3 matrix whose elements consist of random numbers. The general usage is rand(n, m).
trace Summation of diagonal elements of a matrix can be obtained using the trace operator.
For example,
>> C= [1 3 9; 672; 8 -1 -2];
Then, trace(C) produces 6, which is the sum of diagonal elements of C.
zero matrix
>> zeros( 5 ,4)
produces a 5 by 4 matrix consisting of all zero elements. In general, zeros(n,m) is used for an n by m zero matrix.
condition number The command cond(A) is used to calculate the condition number of a matrix A. The condition number represents the degree of singularity of a matrix. An identity matrix has a condition number of unity, and the condition number of a singular matrix is infinity.
>>cond(eye(6))
ans = 1
An example matrix which is near singular is
A=[1 1;1 1+1e-6]
The condition number is
>>cond(A)
ans =
4.0000e+006
Further matrix functions are presented in Table 2. They do not include all matrix functions of the MATLAB, but represent only a part of the whole MATLAB functions. Readers can use the MATLAB Reference Guide or help command to check when they need more MATLAB functions.
6
Table 2. Basic Matrix Functions (Continued)
Symbol Explanations
expm exponential of a matrix
eig eigenvalues/eigenvectors of a matrix
lu LU decomposition of a matrix
svd singular value decomposition of a matrix
qr QR decomposition of a matrix
\ used to solve a setof linear algebraic equations
Matrix exponential The expm(A) produces the exponential of a matrix A. In other words,
>> A =rand(3, 3)
0.2190 0.6793 0.5194
A = 0.0470 0.9347 0.8310
0.6789 0.3835 0.0346
>> expm(A)
1.2448 0.0305 0.6196
ans= 1.0376 1.5116 1.3389
1.0157 0.1184 2.0652
Eigenvalues The eigenvalue problem of a matrix is defined as
A??????
where A is the eigenvalue of matrix A, and (j> is the associated eigenvector.
>> e =eig(A)
gives the eigenvalues of A, and
>> [V,D]=eig(A)
produces the V matrix, whose columns are eigenvectors, and the diagonal matrix D
whose values are eigenvalues of the matrix A. For example,
7
>> A = [5 3 2; 146; 972];
>> [ V,D]=eig(A)
0.4127 0.5992 0.0459
V ? 0.5557 -0.7773 -0.6388
0.7217 0.1918 0.7680
12.5361 0 0
D ? 0 1.7486 0
0 0 -3.2847
LU Decomposition The LU decomposition command is used to decompose a matrix into a combination of upper and lower triangular matrices.
>> A = [1 3 5; 248; 47 3];
>> [L, U] =l
0.2500 1.0000 0
L = 0.5000 0.4000 1.0000
1.0000 0 0
4.0000 7.0000 3.0000
U= 0 1.2500 4.2500
0 0 4.8000
In order to check the result, we try
>> L*U
1
3
5
ans ??2
4
8
4
7
3
The lower triangular matrix L is not perfectly triangular. There is another command
8
available
>> [L,U,P]=lu(A)
1
0
0
L ? 0.25
1
0
0.5
0.4
1
4.0000 7.0000 3.0000
U= 0 1.2500 4.2500
0 0 4.8000
0
0
1
P= 1
0
0
0
1
0
Here, the matrix P is the permutation matrix such that P * A = L * U.
Singular value decomposition The svd command is used for singular value
decomposition of a matrix. For a given matrix,
A?U?V?
where ? is a diagonal matrix consisting of non-negative values. For example, we define a matrix D like
>> D = [1 3 7; 295; 285]
The singular value decomposition of the matrix is
>> [U, Sigma, V]=svd(D)
which results in
0.4295
0.8998
- 0.0775
U ? 0.6629
- 0.3723
- 0.6495
0.6133
- 0.2276
0.7564
15.649
0
0
Sigma ? 20
4.133
0
0
0
0.139
0.1905
- 0.0726
0.9790
V ? 0.7771
- 0.5982
- 0.1956
0.5999
0.7980
- 0.0576
QR decomposition A matrix can be also decomposed into a combination of an orthonormal matrix and an upper triangular matrix. In other words,
A = QR
where Q is the matrix with orthonormal columns, and R is the upper triangular matrix. The QR algorithm has wide applications in the analysis of matrices and associated linear systems. For example,
0.2190 0.6793 0.5194
A?0.0470 - 0.9347 0.8310
0.6789 0.3835 0.0346
Application of the qr operator follows as
>> [Q,R]=qr(A)
and yields
- 0.3063 - 0.4667 - 0.8297
Q??- 0.0658 - 0.8591 0.5076
- 0.9497 0.2101 0.2324
0.7149 - 0.6338 - 0.2466
R ? 0 -1.0395 -0.9490
0 0 0.0011
Solution of linear equations The solution of a linear system of equations is frequently needed in the finite element method. The typical form of a linear system of algebraic equations is written as
Ax = y
and the solution is obtained by
>>x =inv(A) * y
or we can use \ sign as
>>x = A\y
For example
10
>> A=[l 3 4; 5 7 8; 2 3 5];
and
>>y = [10; 9; 8];
Let us compare two different approaches.
>> [inv(^) * y A\y]
? 4.250 ? 4.250
ans ? 1.7500 1.7500
2.2500 2.2500
Loop and Logical Statements
There are some logical statements available in MATLAB which help us in writing combinations of MATLAB commands. Furthermore, loop commands can be used as in other programming languages. In fact, we can duplicate the majority of existing programs using MATLAB commands, which significantly reduces the size of the source codes. A collection of loop and logical statements in MATLAB is presented in Table 3
for loop The for is a loop command which ends with end command.
>>for i = l: 100
a(i, i) = 2 * i;
end
In the above example, i is a loop index which starts from 1 and ends at 100. There may be also multiple loops.
>> for i=1: 100
for j = 1 : 50
for k = 1 : 50
a(i,j) = b(i,k) *c(k,j) +a(i,j);
end
end
end
Table 3 Loop and Logical Statements
Symbol Explanations
for loop command similar to other languages used
while for a loop combined with conditional statement
if produces a conditional statement
elseif, else used in conjunction with if command
break breaks a loop when a condition is satisfied
while The while command is useful for an infinite loop in conjunction with a conditional statement. The general synopsis for the while command is as follows:
while condition
statements
end
For example,
i= 1
while (i < 100)
i = i + l; end
Another example of the while command is
n = 1000; var = [];
while (n > 0) n = n/2-1; var = [var,n];
end
The result is
var =
Columns 1 through 7
499.0000 248.5000 123.2500 60.6250 29.3125 13.6563 5.8281
Columns 8 through 9
1.9141 -0.0430
12
where we used [ ] in order to declare an empty matrix.
if, elseif, else The if, elseif, and else commands are conditional statements which are used in combination.
if condition #1 statement #1
elseif condition #2 statement
#2
else
statement #3
end
For example,
n = 100;
if (rem(n,3) ==0)
x = 0;
elseif (rem(n, 3) == 1)
x=1;
else
x=2;
end
,where rem(x,y) is used to calculate the remainder of x divided by y.
Table 4 Loop and Logical Statements
Symbol Explanations
?? two conditions are equal
~= two conditions are not equal
<=(>=) one is less (greater) than or equal to the other
<(>) one is less (greater) than the other
& and operator - two conditions are met
~ not operator
| or operator - either condition is met
13
break The break command is used to exit from a loop such as if and while. For example,
for i = l : 100 f = i+l; if(i == 10) break; end
end
Logical and relational operators The logical and relational operators of MATLAB are as listed in Table 4
The above command sets are used in combination.
Writing Function Subroutines
MATLAB provides a convenient tool, by which we can write a program using collections of MATLAB commands. This approach is similar to other common programming languages. It is quite useful especially when we write a series of MATLAB commands in a text file. This text file is edited and saved for later use. .
The text file should have filename.m format normally called m-file. That is, all MATLAB subroutines should end with .m extension, so that MATLAB recognizes them as MATLAB compatible files. The general procedure is to make a text file using any text editor. If we generate a file called func1.m, then the file func1.m should start with the following file header
function[ ov1,ov2, • • •] = func1(iv1,iv2,...)
where iv1,iv2,... are input variables while ov1 ,ov2,... are output variables. The input variables are specific variables and the output variables are dummy variables, for which we can use any variables.
For example, let us solve a second order algebraic equation ax2 + bx + c = 0. The solution is given in analytical form as
??? 2 ?
b b 4ac
x
2a
We want to write an m-file with the name secroot.m, which produces the analytical
14
solution.
function [r1,r2]=sec root (a,b,c); %
% Find Determinant-- Any command in MATLAB which starts with
% % sign is a comment statement
Det= 6^2 - 4 * a * c;
if (Det < 0),
r1=- (-b + j* sqrt(-Det))/2/a;
r2 = (-b -j* sqrt(-Det))/ 2/ a;
disp('The two roots are complex conjugates');
elseif(Det == 0),
r1 = -b/2/a;
r2 = -b/2/a;
disp('There are two repeated roots');
else(Det > 0)
r1 = (-6 + sqrt(Det))/2/a;
r2 = (-b - sqrt(Det))/2/a;
disp('The two roots are real');
end
Some commands appearing in the above example will be discussed later. Once the secroot.m is created, we call that function as
>> [rl,r2]=secroot(3,4 ,5)
or
>> [pl,p2] =secroot(3,4,5)
One thing important about the function command is to set up the m-file pathname. The m-file should be in the directory which is set up by the MATLAB configuration set up stage. In the recent version of MATLAB, the set up procedure is relatively easy by simply adding a directory which we want to access in a MATLAB configuration file.
15
Another function subroutine fct.m is provided below.
function [f] =fct(x)
f= (l-x)^2;
The above function represents f(x)= (l-x)^2. In the MATLAB command prompt, we call the function as
>>y =fct(9);
The function subroutine utility of MATLAB allows users to write their own
subroutines. It provides flexibility of developing programs using MATLAB.
File Manipulation
Manipulating files is another attractive feature of MATLAB. We can save MATLAB workspace, that is, all variables used, in a binary file format and/or a text file format. The saved file can also be reloaded in case we need it later on. The list of file manipulation commands is presented in Table 5.
Table 5 File Manipulation Commands
Symbol Explanations
save save current variables in a file
load load a saved file into Matlab environment
diary save screen display output in text format
save The save command is used to save variables when we are working in MATLAB. The synopsis is as follows
save filename var1 var2 ...
where filename is the filename and we want to save the variables, var1 var2 .... The filename generated by save command has extension of .mat, called a mat-file. If we do not include the variables name, then all current variables are saved automatically. In case we want to save the variables in a standard text format, we use
save filename var1 var2 .../ascii/double
load The load command is the counterpart of save. In other words, it reloads the variables in the file which was generated by save command. The synopsis is as follows
16
load filename var1 var2 ...
where filename is a mat-file saved by save command. Without the variables name specified, all variables are loaded.
For example,
>> a=[l 3 4];
>>6 = 3;
>> save test
>> clear all % clear all variables
>>who % display current variables being used
>> load test
>>who
diary Using diary command, we can capture all MATLAB texts including
command and answer lines which are displayed on the screen. The texts will be
saved in a file, so that we can edit the file later. For example,
>> diary on
>>a = 1; b = 4; c = 5;
>> [a b c]
>>d = a*b
>> e = g * h
>> diary off
Now we can use any text editor to modify the diary file. The diary command is
useful displaying the past work procedures. Also, it can be used to save data in a
text format.
Basic Input-Output Functions
Input/output functions in MATLAB provide users with a friendly programming environment. Some input/output functions are listed in Table 6.
Table 6 Input-Output Functions
Symbol Explanations
input save current variables in a file
disp load a saved file into MATLAB
format check the file status in the directory
17
input The input command is used to receive a user input from the keyboard. Both numerical and string inputs are available. For example,
>> age= input('How old are you?')
>> name—input('What is your name','s')
The 's' sign denotes the input type is string.
disp The disp command displays a string of text or numerical values on the screen. It is useful when we write a function subroutine in a user-friendly manner. For example,
>> disp('This is a MATLAB tutoriall')
>> c=3*4;
>> disp('T/ie computed value of c turns out to be')
>> c
format The format command is used to display numbers in different formats.
MATLAB calculates floating numbers in the double precision mode. We do not want
to, in some situations, display the numbers in the double precision format on the
screen. For a display purpose, MATLAB provides the following different formats
>> x = 1/9
x = 0.1111
>> format short e
z = l.lllle-001
>> format long
1 = 0.11111111111111
> >format long e
x- l.llllllllllllllle-001
>> format hex
x = 3/6c71c71c71c71c
Plotting Tools
MATLAB supports some plotting tools, by which we can display the data in a desired format. The plotting in MATLAB is relatively easy with various options available. The collection of plotting commands is listed in Table 7.
A sample plotting command is shown below.
18
>> t = 0:0.1: 10;
>> y = sin(t);
>> plot(y)
>> title('plot(y)')
The resultant plot is presented at the top of Fig. 1.
>> t = 0 :0.1 : 10;
>> y = sin(t);
>> plot(t,y)
title('plot(t,y)')
Table 7 Plotting Commands
Symbol Explanations
plot basic plot command
xlabel(ylabel) attach label to x(y) axis
axis manually scale x and y axes
text place a text on the specific position of graphic
title place a graphic title on top of the graphic
ginput produce a coordinate of a point on the graphic
gtext receives a text from mouse input
grid add a grid mark to the graphic window
pause hold graphic screen until keyboard is hit
subplot breaks a graphic window into multiple windows
The resultant plot is presented at the bottom of Fig. 1. In the above example, t = 0 : 0.1 : 10 represents a vector t which starts from 0 and ends at 10 with an interval of 0.1. We can use just y or both y and t together. In the first case, the horizontal axis represents number of data, from 0 to 101. In the second case, the horizontal axis is the actual time scale t hi the plot(t,y) command.
19
Figure 1 A Sample Plot
Plotting multiple data We plot multiple data sets as shown below.
>> t = 0 : 1 : 100;
>> y1 = sin(t).*t;
>>y2 = cos(t).*t;
>>plot(t,y1,'-',t,y2,'-')
where '-' and '-' represent line styles. The line styles, line marks, and colors are
listed in Table 8.
For example, if we want to plot data in a dashed blue line, the command becomes >>plot(y,'-b');
20
xlabel, ylabel The xlabel('iext') and ylabel('text') are used to label the x and y axes.
axis The axis command sets up the limits of axes. The synopsis is axis[xmin, xmax, ymin, ymax]
text The text command is used to write a text on the graphic window at a designated point. The synopsis is
text(x,y,'text contents')
where x and y locate the (x,y) position of the 'text contents'. A text can be also
added in 3-D coordinates as shown below:
text(x, y,z,'text contents')
Table 8 Line, Mark, and Color Styles
Style Line marks Color
Solid
‘-‘
Point
.
red
r
dashed
' '
star
*
green
g
dotted
':'
circle
o
blue
b
dashdot
'-.'
plus
+
white
w
x-mark
x
invisible
i
ginput This command allows us to pick up any point on a graphic window. The synopsis is
[x,y] = ginput
We can pick as many points as we want on the graphic screen. The vector [x,y] then contains
all the points.
gtext The gtext command is used to place text on the graphic window using the mouse input. The synopsis is
gtext('text')
Once the above command is entered or read in a function subroutine, the cursor on the graphic window is activated waiting for the mouse input, so that the 'text' is located at the point selected by the mouse.
21
grid The grid command adds grids to the graphic window. It is useful when we want to clarify axis scales.
An example plot constructed using some of the commands described above is presented in Fig. 2. The following commands are used for the plot output.
>> t=0:0.1:20;
>> plot(t,sin(t))
>> xlabel('Time(sec)'))
>> ylabel('ydata')
>> title('This is a plot example')
>> grid
>> gtext('sm(t)')
>> axis([0 20 - 1.5 1.5])
pause This command is useful when we display multiple graphic windows sequentially. It
allows us to display one at a time with the keyboard interrupt.
subplot The subplot is used to put multiple plots on the same MATLAB figure window. The command is
>> subplot(pqr)
Figure 2 A Plot Example With Some Commands
22
Figure 3 A Subplot Example
The plot size is adjusted by a p by q matrix on the whole size of the graphic window. Then the third index r picks one frame out of the p by q plot frames. An example subplot is presented in Fig. 3 with the following commands entered.
>> x = 0 : 0.1 : 3 * pi',y = sin(x); z = cos(x);
>> subplot(222)
>> plot(x,y)
>> title('x and y')
>> subplot(223)
>> title('x and z')
>> subplot(224)
>> plot(x,y,'-',x,z,' -- ')
>> title('x and [y z]')
where pi is an internally defined variable equivalent to ?.
23
STUDENT RESEARCH COMPETITION
Information, Guidelines, and Grant Proposal Components
(Revised Summer 2019)
INTRODUCTION
Ball State University offers a variety of internal grant programs that support student projects. The University Research and Creative Arts Committees would like to encourage students to consider applying for a Student ASPiRE Grant. Funds can be provided to assist in project costs such as travel or supplies.
ELIGIBILITY
Applicants must be enrolled for the entirety of the funding period at Ball State University, Burris Laboratory School or the Indiana Academy as one of the following:
* Graduate Student
* Undergraduate Student
* Burris High School Student - grades 11 or 12 (follow instructions/deadlines for Undergraduate Students)
* Indiana Academy High School Student - grades 11 or 12 (follow instructions/deadlines for Undergraduate Students)
o While the instructions/deadlines are the same for Undergraduates and Burris/Academy
students, it should be noted that they are reviewed separately and are not in competition with each other.
* Students may not hold this award for the same project as, or within the same fiscal year as an ASPiRE Student Travel award, a Tucker Autism Research or Travel Grant, or a Hollis Award.
RESEARCH PROGRAM DESCRIPTION
In general, projects submitted to the research competition involve a process of study or discovery that will produce new insights, theories, or will apply such knowledge to problem-solving within the
discipline or in society. Results of research should be worthy of submission for publication in refereed journals and could serve as the foundation for an external funding application.
TYPES OF FUNDING AVAILABLE FOR THE RESEARCH COMPETITION
There are two competition pools for this program, the Graduate and the Undergraduate.
Proposals are capped at $1,000 for Graduate students and $300 for Undergraduate, Burris, and Indiana Academy Students. Funds can be used for Supplies, Equipment, Expenses and Travel (S.E.E.T.) costs associated with research. Costs need to be itemized and justified and must follow accepted university accounting policies and procedures. Funding is not for the purpose of attending conferences or solely
ASPiRE Student Research Guide
for dissemination. It is anticipated that 40 total Graduate awards will be made between the creative arts and research competitions.
Graduate awardees may not hold this award concurrently with an ASPiRE Travel award, a Tucker Autism Research or Travel Grant, or a Hollis Award.
ROLE OF THE MENTOR
All ASPiRE student projects require the support of a faculty mentor or advisor.
* Ball State students must ask a BSU faculty member to serve as a mentor and to provide assistance in preparing the project application.
* Burris and Academy students must ask a school faculty member to serve as a project advisor. A BSU faculty member may also serve in conjunction with the school faculty member.
The mentor is someone who will be involved in the project as a teacher and guide. He or she can provide assistance with defining the scope of the project, determining the best research or creative methods to achieve the desired goals of the project, and suggest or help obtain resources to complete the project. In fact, many faculty mentors welcome students into their research laboratories and programs, and are willing to provide personal day-to-day guidance in the progress of research or creative activity.
TO THE STUDENT
* Discuss the project with your mentor.
o If your research is an extension of your mentor’s, be sure to make a clear distinction on
the unique contribution that your project proposes.
* Ask your mentor to review and proofread your proposal before you submit it.
* Ask your mentor what institutional resources are available to help complete your project.
* Request your required letter of support from your mentor during the initial project meetings with your mentor.
TO THE MENTOR
* Be an active participant in the student’s project.
* Review the guidelines for the required letter of support for the student (see Proposal Attachments).
* Request that you be allowed to review and proofread the proposal prior to submission.
* Complete final report on behalf of the student should student be unavailable.
GENERAL REQUIREMENTS FOR PROPOSALS
Proposals are due no later than 11:59 pm on the submission deadline date listed on the website. Proposals must be submitted via Submittable. PLEASE NOTE: Applications missing information or not adhering to the guidelines will be returned without review, no exceptions.
* PROJECT DESIGN
Proposal Narrative: Limited to two Pages; double?space the text using no smaller than a size 11 font, New Times Roman or Arial, suggested type font, with at least 1 inch margins in all directions. Label the required sections as stated below and address each topic completely but concisely. Because review committees are comprised of members from a variety of disciplines, all proposals should be written in clear, non-technical language readily understood by an educated layperson. Material should be
2
ASPiRE Student Research Guide
organized in such a way that a clear outcome of the project is readily discernable.
This section is intended to introduce the background and methodology of the project. Keeping within the two page limit, please address the following topics:
1. Executive summary - An overview of the proposed project and its significance.
2. Background information - The history of the idea, the current status of the research in this area, and a definition of terms needed to facilitate a lay reader’s understanding of the project. Note: if your project is an extension of your mentor’s research, you must make a clear distinction between your project and the unique and innovative contribution your research proposes.
3. Goals, objectives, and significance - A discussion of the scope of the project, focusing on the overall goals and specific objectives of the research. Material should be organized in such a way that a clear outcome of the project is readily discernable. Project significance should clearly delineate the anticipated impact in the field or on one’s research agenda. Plans for dissemination, publication, or presentation must be described in this section.
4. Research Methods & Timeline - An identification of research questions and a detailed description of the proposed project. Describe in detail the methods, procedures, steps or activities to be undertaken, and a timetable for completing the work. Overall, the goal of the research methods section is to outline what will happen during the course of the research and how it will lead to an outcome. The research methodology description should provide a reader with a clear mental picture of the proposed work.
Regardless of whether the research involves quantitative or qualitative methodology, a good description contains the following information:
* What research question(s) will the data propose to answer, or how will the defined inquiry be described?
* How will the data be collected and analyzed?
* If human subjects are to be involved, how will they be selected and why?
* What is the expected outcome or what further research questions may be prompted by the results?
* BUDGET & BUDGET NARRATIVE
The goal of the budget section of the proposal is to identify the type and amount of funds required to carry out the project. There should be a clear connection between the budget items and their contribution to the proposed project. Funding is not for the purpose of dissemination only.
Budget Form is a separate section of the Submittable application.
3
ASPiRE Student Research Guide
Budget Form Category Definitions
Supplies, Materials, Minor Equipment--Any supplies needed to complete the project. Supplies include:
Office supplies, postage, software, laboratory supplies, books, etc.
Travel--Any expenses incurred during or directly related to travel: Airfare, lodging, mileage, per diem, etc.
Other:
Participant Costs--Stipend to research subjects or project participants
NOTE: Participant Incentives must be in line with the Controller's Office policy on awards, prizes,
and stipends. For details please see the latest policy. Per this policy, Ball State faculty and staff
are not eligible to serve as paid participants on Aspire-funded projects.
Contractual—Non BSU agreements (including software licenses) and non BSU project consultants.
Budget Narrative is limited to 1 page, double spaced; serves as a text description of items denoted in the Budget Form. Do not include a table within the budget narrative. Use the budget narrative section to:
* Give a complete explanation of the amounts listed on the Budget Form.
* Explain rationale for figures in budget itemization.
* Name the source of the contribution for the "BSU-Other" column (e.g. Department, College, External Grant, personal funds, etc.)
Please keep in mind the following when preparing the Budget and Budget Narrative.
* Student Awards may NOT pay salaries or wages.
* Provide sufficient detail regarding how various budget items were calculated and proposed budget amounts are justified.
* All equipment, tangible materials, and books purchased on a grant are property of the University.
* ASPiRE funds not encumbered or spent by the end of the project period will revert back to the ASPiRE Program.
o Mileage is based on current University rates: BSU Travel Regulations and Procedures Manual
* PROPOSAL ATTACHMENTS
Please include the following required and optional materials as attachments to your proposal.
* Required:
o Literature references (1 page maximum; use the format that is standard for publishing in your field.)
o Letter of support from your faculty mentor. The letter of support should address the following issues:
* The viability of the project
* Innovative contributions of the project
* How the project fits with the student’s academic progress
* Student’s ability to complete the project
* What role, if any, the student’s project plays in your own research
o 1 page Curriculum Vitae for student applicant and co-applicants
4
ASPiRE Student Research Guide
* Optional:
o If available at the time of application, include approved IRB or IACUC letter.
o Appendix of related materials (e.g., survey questions)
o Additional letter(s) of support from collaborators or other agencies. (It is recommended that a letter of support be obtained if an outside agency is involved in any capacity.)
o Glossary of Terms (1 page maximum). It is recommended that even though your proposal should be written for the educated lay person, it may be beneficial to the reviewer to refer to a glossary of terms used in the narrative or methodology.
* ADDITIONAL SUPPORT MATERIALS
Occasionally, a proposal can be strengthened by the inclusion of Additional Support Materials. These
include, but are not limited to, the following types of information.
* Examples of artwork, photographs, etc.
* Examples of previously published material.
* Anything else the review committee could use to evaluate the proposal.
Submission
A complete proposal submission includes:
* Proposal narrative, budget narrative, required appendices (curriculum vitae and letter of support, and optional appendices. These items should be clearly labeled (i.e., Application Cover Sheet, Budget Narrative, etc.).
* Optional Additional Support Materials.
Review Criteria for Competitions
All proposals are reviewed by the appropriate research subcommittee. Reviewers assign points and an overall rating (fund, possibly fund, do not fund) to the proposals prior to the review meeting. At the review meeting, those proposals that fall into the “possibly fund” category are discussed. Those that are clearly fund or do not fund are not discussed unless there is a specific issue raised by a member of the (sub)committee. Proposals are scored out of a possible 30 points ? see below for exact criteria.
Proposal Background (maximum of 5 points)
Origin of idea or project
Literature review
Significance of Project (maximum of 10 points)
Goal of proposal is clearly stated
Benefits to discipline and practice
Impact of expected outcome of the project
Research Methods (maximum of 10 points)
Project plan (e.g., development of creative design, or data collection and analytical procedure)
Feasibility of project within time frame
Use of resources and budget
5
ASPiRE Student Research Guide
Dissemination (maximum of 5 points)
Clear and specific plans for dissemination, publication, commercialization, or presentation
Appropriate to the field, applicant and/or project
REVIEW PROCESS
The University Research Committee is comprised of faculty members and students nominated by the University Senate. The Research Committee reviews the proposals submitted for the research program. The University Research Committee is divided into two subcommittees according to discipline:
Subcommittee #1: Humanities, Arts, Music, and Architecture
Subcommittee #2: Education, Business, and Social Sciences
Subcommittee #3: Mathematical, Biological, Physical, Health & Applied Sciences
All proposals submitted will undergo the same review process.
* Administrative review by SPA, for:
o Previous internal award documentation
o Adherence to proposal submission guidelines
* Subcommittee Review against Review Criteria (see below)
* Funding Recommendation – Subcommittees make suggestions for funding to the Director of SPA, who then makes the final funding decision based on availability of funds.
AWARD DECISIONS
After a funding decision is made, the applicant will receive an email indicating that a decision has been made. If the project is not recommended for funding, the decision letter may suggest ways to strengthen or improve the proposal, or suggest other funding sources. If the committee feels the proposal could be strengthened with minor changes, the applicant will be asked to revise and resubmit to the next round of the competition.
PROJECT PERIOD
The project period for graduate and undergraduate competitions are as follows:
Graduate Students: Fall Submission: January 1 – December 31
Spring Submission: May 1 – April 30
Undergraduate Students: Fall Submission: January 1 – December 31
Spring Submission: May 1 – April 30
6
ASPiRE Student Research Guide
FUNDING AWARD STIPULATIONS
* All awards must comply with the ASPiRE Post Award Guidelines.
* Ball State students are limited to one Research or Creative Arts award per student classification level (Burris/Indiana Academy, undergraduate, graduate).
* Awarded funds cannot be used outside of the Project Period. Any remaining funds will be returned to the program.
* Students may not hold this award for the same project as, or within the same fiscal year as an ASPiRE Student Travel award, a Tucker Autism Research or Travel Grant, or a Hollis Award.
* For projects that involve co?directors, the award is per project, not per student.
* Funding requests may be reduced based on funding limitations; however students are still only eligible for one Research or Creative Arts award per degree, no matter the amount awarded.
* Projects involving the use of human subjects, animals, radioactive or other biohazardous material must receive approval from the appropriate research compliance review committee to ensure compliance with federal regulations and established university guidelines. See the Office of Research Integrity (ORI) for forms and submission requirements. Note: Applications that require review from ORI need not be submitted prior to submitting an ASPiRE application. However, approval must be obtained before the project can begin or grant funds can be released.
* Students are highly encouraged to participate in the annual SPA Student Symposium, occurring each Spring.
* All equipment and materials purchased on a grant will remain the property of Ball State University.
FINAL REPORT
The final report deadline is April 30 following the project period for fall competitions and December 31
following the project period for spring competitions.
The Final Report Form can be submitted via this link or through the link on the Aspire website. Grant recipients and /or their mentors failing to submit acceptable final reports will be declared ineligible for further support under programs supervised by the ASPiRE Program. If reports cannot be submitted by the date specified in the program guidelines, the ASPiRE Program Manager will consider written requests for an extension by email to aspire@bsu.edu.
7
The University of Texas at El Paso
Graduate School
Thesis &
Dissertation
Formatting Guidelines
http:graduate.utep.edu
Current Students
Forms
Doctoral Completion Packet
or
Master’s Completion Packet
Olympia Caudillo
Assistant Director of Graduation Evaluation
Graduate School
747-7902
ocaudillo2@utep.edu
University of Texas
at El Paso
Graduate School
Formatting
Guidelines
House-rules
specific to the
UTEP Graduate
School
UTEP Graduate
School house-rules
ensure the
consistency and
uniformity of UTEP
theses and
dissertations.
Formatting
guidelines refer to
the
page layout
not the writing or
citation style
Consult your
committee for
discipline specific
writing or citation
style
The Graduate School
does not require the
submission of
Non-Theses
Consult with your
program for
instructions on format
and final presentation
of the non-thesis
document.
Submit thesis or
dissertation for
format check
at least one day
PRIOR
to your scheduled
defense date
Email the file or submit it
on a flash drive
ocaudillo2@utep.edu
Paper copy not necessary!
Thesis &
Dissertation
Word Template
Template is made available for
the convenience and use of
UTEP’s graduate students.
Students are not obligated to
use the template.
Use only the pages/sections
that are applicable or required
by your committee.
Order of Thesis or
Dissertation
4 Signature Page
Copyright Notice
Dedication and/or Epigraph
4 Title Page
Acknowledgments and/or Preface
4 Abstract
4 Table of Contents
List of Tables
List of Figures
List of Illustrations / Maps /Slides
4 Text
4 References
Glossary
Appendix
4 Curriculum Vita
Required Pages
1. Signature Page
2.Abstract (Dissertation Only)
3. Title Page
4. Table of Contents
5. List of Figures*
6. List of Tables*
7. Text
8. Bibliography
9.Curriculum Vita
*A separate List of Figures & Tables required if there are 3 or more figures
or tables.
All other Pages are Optional
unless your committee requires the
submission of optional pages.
SAMPLE OF
REQUIRED PAGES
Signature Page
TITLE OF THESIS or DISSERTATION
STUDENT’S FULL OFFICIAL NAME
Title of the academic department or program
APPROVED :
Art Johnson, Ph.D., Chair
Maryann Smith, Ph.D.
George Lopez, Ph.D.
Title Page
TITLE OF THESIS or DISSERTATION
By
STUDENT’S FULL OFFICIAL NAME, DEGREES
THESIS or DISSERTATION
Presented to the Faculty of the Graduate School of
The University of Texas at El Paso
in Partial Fulfillment
of the Requirements
for the Degree of
Title of the academic department or program
Table of Contents
Sample pages using
> Chapters
> Sections
The format of all Major Headings
must remain consistent so if all
Major Headings are written using
Only Upper Case Letters as
shown in the sample on the
previous page, then Major
Headings as listed in the Table of
Contents should also be written
using only Upper Case Letters as
shown in the sample Table of
Contents on the next page.
This only applies to major headings –
not subchapter headings.
Page
ACKNOWLEDGEMENTS iii
ABSTRACT v
TABLE OF CONTENTS vi
LIST OF TABLES vii
LIST OF FIGURES ix
Chapter
1. INTRODUCTION 1
2. PROBLEM 4
2.1 Mechanics 7
2.2 Thermodynamics 8
3. SOLUTION ALTERNATIVES 12
3.1 What Is a Solution? 13
3.2 How Do You Apply a Solution? 15
3.3 Why Use Our Solution? 17
4. SUMMARY AND CONCLUSIONS 20
4.1 Summary 20
4.2 Conclusions 22
LIST OF REFERENCES 24
TITLE OF APPENDIX 27
CURRICULUM VITA 28
vi
Major headings using initial capital letters only.
The format of all Major Headings
must remain consistent so if
Major Headings are written
Capitalizing the First Letter Only
as shown in the sample on the
previous page, then Major
Headings as listed in the Table of
Contents should also be written
Capitalizing the First Letter Only
as shown in the sample Table of
Contents on the next page.
This only applies to major headings –
not subchapter headings.
Table of Contents
Page
Acknowledgements iii
Abstract v
Table of Contents vi
List of Tables viii
Introduction 1
Problem 6
Solution Alternatives 11
Conclusions 18
Works Cited 22
Curriculum Vita 25
iv
Sample of Table of Contents using
Sections instead of Chapters
Maria Ornelas was born in Tucson, Arizona. The second daughter of Edmundo Ornelas and Alicia Avila Ornelas, she graduated from Burges High School, El Paso, Texas, in the spring of 1980 and entered The University of Texas at El Paso in the fall with the Stevens Scholarship. While pursuing a bachelor’s degree in computer science, she worked with Spartan Software, a software development company in Austin, Texas, during the summers of 1983 and 1984, and later worked with the company full-time after receiving her bachelor’s of science degree from The University of Texas at El Paso in 1985. She was a guest speaker during the Conference in Natural Language Processing in 1985 in San Jose, California and published Techniques in Natural Language Parsing in 1988 in the ACM Journal of Artificial Intelligence. In the fall of 1989, she entered the Graduate School at The University of Texas at El Paso.
Permanent Address: 1967 Alan Turin
El Paso, Texas 79924 Or
ocaudillo2@utep.edu
In keeping with UTEP Graduate School Thesis & Dissertation Formatting
Guidelines, the vita is a brief biographical sketch of the writer which provides
information for future readers and researchers of unpublished works. Ideally
the vita should be one page and is not a resume or advertisement for
employment.
28
FORMATTING
SUMMARY
Margins
1. One Inch Margins – 4 sides
2. Margins apply to the entire
document
One Inch
One Inch
Important Notes:
1. First page of every new section
or chapter begins on a new
page but numbering is
continuous.
2. Use a Standard Font, Style and
Size – 12.
3. Largest Font size allowed – 14.
4. Smaller font may be used for
table and/or figure/graphic
captions.
5. Use the same font style
throughout the document.
6. Use the same standard font
style and size for page
numbers.
7. Double space the entire
document.
8. Keep the format of all major
headings consistent.
Major headings are the titles of
your chapters or sections.
9. Captions for Tables are placed
above the Table.
Table 1
10. Captions for figures/graphics are
placed below the figure/graphic.
Figure 1
Separate List of Figures, Tables, graphics, etc. needed if there are more than three
tables, figures, etc.
Page Numbers
1. Page Numbers are centered,
one inch from bottom of
page.
*Margins on the Template are set at .7
Please do not adjust the margins on the
template.
iii
2. The page AFTER the title page is
the first page to appear with a
number.
3. All pages are included in the
Total Count even though they are
not numbered.
4. Preliminary pages are numbered
using lowercase roman numerals.
5. The first page of the actual body
of the text is always Arabic
Numeral “1”.
6. Subsequent pages are numbered
consecutively through the last
page, the CV.
FINAL SUBMISSION
Final Submission
Only after you obtain Graduate School
Approval!!
1. Please perform a SPELL
CHECK before final
submission!!!
2.Submit the thesis or dissertation
in PDF format on a CD in a case –
do not include any WORD files
(preferably a rewritable CD).
3.Keep the title of the PDF thesis or
dissertation file simple. Remember it
will be available to the public.
Mbarraza Thesis
4. Submit a hard copy of the
signature page and the defense
form along with the CD.
Signature Page in Thesis
or Dissertation PDF file is
Blank. Do not scan the signature
page and insert it in the final file.
5.Label the CD with your name,
semester and year. Indicate if it is a
thesis or dissertation.
M. Flores
Thesis
Spring 2013
6. The CD copy submitted to the
Graduate School is catalogued at the
UTEP Library.
7. Submit the thesis or dissertation PDF
file at the UMI website for on-line
publication –
http://dissertations.umi.com/utep
8. The UMI/ProQuest website will ask
you for a user ID and password that
you create.
9. The Thesis/Dissertation On-Line
Publication fee paid at UTEP is for
Traditional Publication. You are not
obligated to purchase additional
services from UMI/ProQuest but if you
choose to make a purchase, you are
responsible for any additional charges
incurred.
10. Theses and Dissertations are
regarded as publications once they
are approved and submitted to the
Graduate School and will be available
to the public.
11. The author of the thesis or
dissertation holds copyright
privileges.
12. If portions of the thesis or
dissertation have been published, you
must include written permission from
the publisher to include those
sections in your thesis or dissertation.
13. Please notify the Graduate School
if your thesis or dissertation research
has patent or discovery paperwork
pending.
Copyright
In accordance with the University of Texas System Board of Regents' Rules, the Board (University) does not exert ownership in scholarly works UNLESS it is a scholarly work (i) created by someone who was specifically hired or required to create it or (ii) commissioned by the System or a component institution of System, in either of which cases, Board, not the creator, will own the intellectual property. Scholarly works include: educational materials, artworks, musical compositions, and dramatic and nondramatic literary works related to the author's academic or professional field, regardless of the medium of expression.
This applies to works authored by students, professionals, faculty, and non-faculty researchers. The Board retains certain rights in these works as set forth in the Policy and Guidelines for Management and Marketing of Copyrighted Works.
http://research.utep.edu/Default.aspx?tabid=72170
The University encourages authors to manage their copyrights carefully.
For more information about copyrightable materials and use, go to the Copyright Crash Course at:
http://www.utsystem.edu/ogc/IntellectualProperty/
Inclusion of Previous Publications
The verbatim inclusion of previously published papers or articles by the student in a thesis or dissertation is the prerogative of the major department, but copyright laws must be followed. Even in fields where collaborative research has become the norm, the student must be the sole author of the manuscript being submitted to meet degree requirements and therefore must specify his or her contribution to the overall research program and delineate contributions of colleagues. Where published papers, books, or ready-for-publication papers are included, the following guidelines should be followed:
1. Multiple Authorship
Multiple authorship of a published paper should be addressed by clearly designating in an introduction the role that the thesis or dissertation author had in the research and the production of the published paper. The student must have made a major contribution to the research and writing of papers included in the thesis or dissertation. This applies only to the verbatim inclusion of a previously published paper. The inclusion of short passages from a paper previously published by the student is cited in the same manner as a work by any other author.
2. Referencing
There must be a full citation of where individual papers have been published.
3. Permission for Copyrighted Materials
Written permission must be obtained for all copyrighted materials used in the thesis or dissertation, including manuscripts submitted for publication but not yet published, and these permissions must be submitted with the final manuscript.
Submit thesis or
dissertation for
format check
at least one day
PRIOR
to your scheduled
defense date
Email the file or submit it
on a flash drive to
ocaudillo2@utep.edu
Paper copy not necessary!
Remember that accuracy and
consistency are the
all?important matters. These
qualities make the thesis or
dissertation a usable research tool
for other readers.
Perform a spell check before
final submission to the
Graduate School. Your document
will reside in the UTEP Library for
all time and will be accessible
electronically to millions of
researchers in and out of your
field.
Please give your thesis or
dissertation the care
and attention that it deserves.
Questions, comments or concerns:
Olympia Caudillo
Assistant Director of
Graduation Evaluation
The Graduate School
500 W. University, ASB 223
El Paso, Texas 79968-0587
(915) 747-7902
Fax (915) 747-5788
ocaudillo2@utep.edu
http://graduate.utep.edu
User guide for Modularized Surrogate Model Toolbox
Juliane M¨uller
Tampere University of Technology
Department of Mathematics
email: juliane.mueller2901@gmail.com
October 2, 2012
1 Introduction
This user guide accompanies the surrogate model toolbox for global optimization problems. The toolbox is made for computationally expensive black-box global optimization problems with continuous, integer, or mixed-integer variables. Problems where several or all variables have to be integers may also have black-box constraints, whereas purely continuous problems may only have box constraints.
For problems with computationally cheap function evaluations the toolbox may not be very efficient. Surrogate models are intended to be used when function evaluations take from several minutes to several hours or more. When reading this manual it is recommended to simultaneously take a look at the code.
The code is set up such that the user only has to define his/her optimization problem in a Matlab file (see Section 6.1). Additional input such as the surrogate model to be used, the sampling strategy, or starting points are optional (see Section 6).
This document is structured as follows. In Section 2 the general structure of a surrogate model algorithm is summarized. The installation is described in Section 3. The dependencies of the single functions in the code are shown in Section 4. Section 5 briefly summarizes how the surrogate model algorithm works in general. Section 6 describes the options for the input of the main function. In Section 7 the input and output of the single subfunctions of the algorithm are described. Examples for using the surrogate model algorithm are given in Section 8. In Section 9 it is explained how the user can define an own (mixture) surrogate model and an example is given. The elements of the saved results are described in Section 10.
Finally, if you have any questions, or you encounter any bugs, please feel free to contact me at the email address juliane.mueller2901@gmail.com.
2 Surrogate Model Algorithms
Surrogate models (also known as response surfaces, metamodels) are used during the optimization phase to approximate expensive simulation models [1]. During the optimization phase information from the surrogate model is used in order to guide the search for improved solutions. Using the surrogate model instead of the true simulation model reduces the computation time considerably. Most surrogate model algorithms consist of the same steps as shown in the algorithm below.
Algorithm General Surrogate model Algorithm
1. Generate an initial experimental design.
1
2. Do the costly function evaluations at the points generated in Step 1.
3. Fit a response surface to the data generated in Steps 1 and 2.
4. Use the response surface to predict the objective function values at unsampled points in the variable domain to decide at which point to do the next expensive function evaluation.
5. Do the expensive function evaluation at the point selected in Step 4.
6. Use the new data point to update the surrogate model.
7. Iterate through Steps 4 to 6 until the stopping criterion has been met. Surrogate model algorithms in the literature differ mainly with respect to
* the generation of the initial experimental design;
* the chosen surrogate model;
* the strategy for selecting the sample point(s) in each iteration.
Typically used stopping criteria are a maximum number of allowed function evaluations, a maximum allowed CPU time, or a maximum number of failed iterative improvement trials.
3 Installation
Download the file SurrogateOptimizationModule.zip and unzip it in a location known to the Matlab search path. Alternatively, you can add a new folder to the Matlab search path by clicking in the Matlab window on
File -+ Set Path... -+ Add with Subfolders -+ Save
You can try if the algorithm works by typing
>> SurrogateModelModule_v1(’datainput_hartman6’,300,’MIX_RcKg’, ’CAND’, ’SLHD’, 15);
into the command prompt.
4 Code Structure
The structure of the code is outlined here. Depending on if there are integrality constraints in the problem either option (a), (b), or (c) is used. If the problem has only continuous variables, the user can choose different sampling strategies (options (i)-(v)). For problems with integer constraints only the candidate point sampling strategy can be used at this moment.
SurrogateModelModule v1.m
StartingDesign.m
SLHD.m/lhsdesign.m/bestlh.m/cornerpoints.m
(a) OptimizationPhase continuous.m
FitSurrogateModel.m
RBF.m/POLY.m/dacefit.m/aresbuild.m/DempsterFor2models.m/DempsterFor3models.m
(i) CandidatePointSampling.m
Perturbation.m
PredictFunctionValues.m
RBF eval.m/POLY eval.m/predictor.m/arespredict.m
DistanceCriterion.m
PredictedValueCrit.m
(ii) SurfaceMinSampling.m
ODDS.m
FitSurrogateModel.m
RBF.m/POLY.m/dacefit.m/aresbuild.m/DempsterFor2models.m/DempsterFor3models.m
2
(iii) ExpImprovementSampling.m
distanceupdate.m
likelihood new.m
ExpectedImprovement.m
(iv) BumpinessMinSampling.m
ODDS.m
RBF.m
bumpiness measure.m
(v) ScoreMinSampling.m
Multi DDS.m
FitSurrogateModel.m
RBF.m/POLY.m/dacefit.m/aresbuild.m/DempsterFor2models.m/DempsterFor3models.m
(b) OptimizationPhase integer.m
SOI OP1.m
FitSurrogateModel.m
RBF.m/POLY.m/dacefit.m/aresbuild.m/DempsterFor2models.m/DempsterFor3models.m
Perturbtation SOI.m
PredictFunctionValues.m
RBF eval.m/POLY eval.m/predictor.m/arespredict.m
DistanceCriterion.m
SOI OP2.m
FitSurrogateModel.m
RBF.m/POLY.m/dacefit.m/aresbuild.m/DempsterFor2models.m/DempsterFor3models.m
Perturbtation SOI.m
PredictFunctionValues.m
RBF eval.m/POLY eval.m/predictor.m/arespredict.m
DistanceCriterion.m
PredictedValueCrit.m
(c) OptimizationPhase mixedinteger.m
FitSurrogateModel.m
RBF.m/POLY.m/dacefit.m/aresbuild.m/DempsterFor2models.m/DempsterFor3models.m
Perturbation SOMI.m
PredictFunctionValues.m
RBF eval.m/POLY eval.m/predictor.m/arespredict.m
DistanceCriterion.m
PredictedValueCrit.m
5 The Main File SurrogateModelModule v1.m
The file from which to run the surrogate model toolbox is SurrogateModelModule v1.m. The file expects several inputs (see Section 6), of which only the first is mandatory. The algorithm starts by checking the input and assigns default values to variables that have not been set by the user. An initial experimental design is created (StartingDesign.m). After evaluating the expensive function evaluations at the selected points in the initial experimental design one of the three (continuous, integer, mixed-integer) optimization routines is called.
6 Input
The main file SurrogateModelModule v1.m requires several input arguments (see Table 1), out of which only the first is mandatory to run the algorithm. If no input is given for the remaining arguments, default values are used.
3
Table 1: Input parameters
Input Description
data file string with name of data file containing optimization problem data (mandatory!)
maxeval integer defining maximum number of allowed function evaluations (default 400)
surrogate model string defining which surrogate model to use (default ’RBFcub’)
sampling technique string defining the technique for picking the next sample site (default ’CAND’)
initial design string defining which initial experimental design should be used (default ’SLHD’)
number startpoints integer defining the number of initial starting points (default 2(d + 1), d=dimension)
starting point matrix with points added to initial experimental design
6.1 Input data file
The data file contains all the necessary problem information. See for example the file datainput hartman6.m. The data file has no input argument, and one output argument (the structure Data). The data structure has to contain the information shown on Table 2.
Table 2: Contents data file
Variable Description
Data.xlow variable lower bounds, row vector with d (=dimension) entries
Data.xup variable upper bounds, row vector with d (=dimension) entries
Data.dim problem dimension, integer
Data.integer row vector containing indices of variables with integer constraints (Data.integer=[] if no integrality constraints)
Data.continuous row vector containing indices of continuous variables (Data.continuous=[] if no continuous variables)
Data.objfunction handle to objective function, must return a scalar value
Data.constraint cell array with handles to constraint functions; only for integer and mixed-integer problems; optional
6.2 Input surrogate model
There are several options for choosing the surrogate model. Surrogate models can be interpolating (for example radial basis functions (RBF), kriging) or non-interpolating (polynomial regression models, multivariate adaptive regression splines (MARS)). The toolbox allows the user to choose between different (mixture) models (see Tables 3, 4, 5, 6). The surrogate model to be used can be defined by the user as string as function input surrogate model. Numerical experiments showed that if nothing about the behavior of the black-box objective function is known, a mixture surrogate model containing a radial basis function should be used. Mixture models prevent accidentally selecting the worst surrogate model.
Table 3: Radial basis function surrogate models
surrogate model Description
’RBFcub’ cubic radial basis function interpolant
’RBFtps’ thin plate spline radial basis function interpolant
’RBFlin’ linear radial basis function interpolant
4
Table 4: Kriging surrogate models
surrogate model Description
’KRIGexp0’ Kriging model with exponential correlation function and 0-order regression polynomial
’KRIGexp1’ Kriging model with exponential correlation function and first order regression polynomial
’KRIGexp2’ Kriging model with exponential correlation function and second order regression polynomial
’KRIGgexp0’ Kriging model with generalized exponential correlation function and 0-order regression polynomial
’KRIGgexp1’ Kriging model with generalized exponential correlation function and first order regression polynomial
’KRIGgexp2’ Kriging model with generalized exponential correlation function and second order regression polynomial
’KRIGgauss0’ Kriging model with Gaussian correlation function and 0-order regression polynomial
’KRIGgauss1’ Kriging model with Gaussian correlation function and first order regression polynomial
’KRIGgauss2’ Kriging model with Gaussian correlation function and second order regression polynomial
’KRIGlin0’ Kriging model with linear correlation function and 0-order regression polynomial
’KRIGlin1’ Kriging model with linear correlation function and first order regression polynomial
’KRIGlin2’ Kriging model with linear correlation function and second order regression polynomial
’KRIGspline0’ Kriging model with spline correlation function and 0-order regression polynomial
’KRIGspline1’ Kriging model with spline correlation function and first order regression polynomial
’KRIGspline2’ Kriging model with spline correlation function and second order regression polynomial
’KRIGsphere0’ Kriging model with spherical correlation function and 0-order regression polynomial
’KRIGsphere1’ Kriging model with spherical correlation function and first order regression polynomial
’KRIGsphere2’ Kriging model with spherical correlation function and second order regression polynomial
’KRIGcub0’ Kriging model with cubic correlation function and 0-order regression polynomial
’KRIGcub1’ Kriging model with cubic correlation function and first order regression polynomial
’KRIGcub2’ Kriging model with cubic correlation function and second order regression polynomial
5
Table 5: Non-interpolating surrogate models
surrogate model Description
’POLYlin’ linear regression polynomial
’POLYquad’ quadratic regression polynomial
’POLYquadr’ reduced quadratic regression polynomial
’POLYcub’ cubic regression polynomial
’POLYcubr’ reduced cubic regression polynomial
’MARS’ multivariate adaptive regression spline
Table 6: Mixture surrogate models
surrogate model Description
’MIX RcKg’ mixture of cubic radial basis function interpolant and kriging model with Gaussian correlation function (first order regression polynomial)
’MIX RcM’ mixture of cubic radial basis function interpolant and multivariate adaptive regression spline
’MIX RcPc’ mixture of cubic radial basis function interpolant and full cubic regression polynomial
’MIX KgM’ mixture of kriging model with Gaussian correlation function (first order regression polynomial) and multivariate adaptive regression spline
’MIX KgPc’ mixture of kriging model with Gaussian correlation function (first order regression polynomial) and cubic regression polynomial
’MIX KgPqr’ mixture of kriging model with Gaussian correlation function (first order regression polynomial) and reduced quadratic regression polynomial
’MIX KgPcr’ mixture of kriging model with Gaussian correlation function (first order regression polynomial) and reduced cubic regression polynomial
’MIX RcKgM’ mixture of cubic radial basis function interpolant, kriging model with Gaussian correlation function (first order regression polynomial) and multivariate adaptive regression spline
6.3 Input SampleStrategy
For problems with only continuous variables the user can determine the strategy for selecting the sample point in each iteration (see Table 7, input sampling technique).
In the candidate point approach [10] (CAND) two groups of candidates for the next sample point are created. The points in the first group are generated by perturbing the best point found so far. The points in the second group are generated by uniformly selecting points from the whole box-constrained variable domain. The response surface is used to predict the objective function values at the candidate points (response surface criterion). The distance of each candidate point to the set of already sampled points is computed (distance criterion), and a weighted sum of both criteria is used to determine the best candidate point.
When using the minimum of the response surface (’SURFmin’) as sampling strategy, the dynamically dimensioned search (DDS) algorithm [12] is used to find the surface minimum. Other algorithms (such as Matlab’s fmincon) could be used as well. Note that it is not necessary to find the global minimum of the surface. Rather, response surfaces that are able to model multimodalities are preferable because sample points are likely to be derived from local optima in different regions of the variable domain, which in turn makes the search more global (see also [11]).
Using the point that maximizes the expected improvement (’EImax’) has been introduced by Jones et al. [6] (EGO algorithm). When using this sampling strategy, the kriging model must be used
6
as response surface because it returns an error estimate together with the objective function value prediction. Thus, if the maximum expected improvement is the sampling strategy of choice, the surrogate model cannot be set by the user.
The strategy of using the point that minimizes the scoring criterion (’SCOREmin’) corresponds to finding the point that minimizes the weighted score defined by the response surface and the distance criterion. In contrast to the candidate point approach (where the best candidate from a limited number of points is selected) the actual minimum is tried to be found. When minimizing the scoring function the DDS algorithm is used.
Minimizing the bumpiness measure [3] (’BUMPmin’) aims at finding the point in the variable domain where it seems most ”reasonable” that the objective function takes on a given target value, i.e. it is more reasonable that the objective function takes on a very low value in regions of the variable domain where already low function values have been encountered, rather than in regions where large objective function values have been observed and very steep valleys would result. The sampling strategy can only be used in connection with a radial basis function interpolant. During the optimization phase the algorithm cycles through several different target values as described by Holmstr¨om [4].
Computational experiments showed that the candidate point approach (’CAND’), maximizing the expected improvement (’EImax’), or using the minimum of the response surface (’SURFmin’) lead in general to the best results.
Table 7: Options for sampling techniques
sampling technique Description
’CAND’ candidate point approach (default)
’SURFmin’ uses the minimum point of response surface
’EImax’ uses the point that maximizes the expected improvement (currently working only for kriging with generalized exponential correlation function)
’SCOREmin’ uses the scoring criteria from the candidate point approach, but tries to find the best point in the whole domain with respect to these criteria instead of choosing the best point among a limited number of points as done in ’CAND’)
’BUMPmin’ minimizes bumpiness measure (see [3]), only for cubic, linear and thin-plate spline radial basis function interpolant
6.4 Input initial design
The user can choose between the initial experimental design strategies shown in Table 8.
Table 8: Initial experimental design strategies
sampling technique Description
’SLHD’ symmetric Latin hypercube design
’LHS’ Matlab’s built-in Latin hypercube design (default)
’CORNER’ uses (subset) of corner points of variable domain
’SPACEFIL’ space filling design (as in EGO, see [2])
When choosing the SLHD option, it is advised to use at least 2d (d=problem dimension) starting points, due to the possibility of linear dependencies in the design matrix. When many points are required for the initial experimental design, using the ’SPACEFIL’ strategy becomes computationally
7
more expensive because of the optimization routine when creating the design. Also note that if more points are required for the initial experimental than there are corner points, the corner point sampling strategy as initial experimental design fails.
6.5 Inputs number startpoints and starting point
The user can define the number of points s/he wishes to have in the initial experimental design. The minimum number of required points depends on the surrogate model and initial design strategy.
* When using the symmetric Latin hypercube sampling (SLHD), at least 2d (d=dimension) points should be used due to the possibility of linear dependencies in the design matrix.
* When using the kriging model together with the second order regression polynomial (KRIGexp2, KRIGgexp2, KRIGgauss2, KRIGlin2, KRIGspline2, KRIGsphere2, KRIGcub2), at least 1 + 2d + (d )
2
points are needed, otherwise the DACE toolbox fails.
* When using the full quadratic polynomial regression model (POLYquad) at least 1 + 2d + (d )
2
points are needed. Otherwise the least squares problem is underdetermined.
* When using the reduced quadratic polynomial regression model (POLYquadr), at least 2d + 1 points are needed. Otherwise the least squares problem is underdetermined.
* When using the full cubic polynomial regression model (POLYcub), at least 1 + 3d + (d ) + (d )
2 3
points are needed. Otherwise the least squares problem is underdetermined.
* When using the reduced cubic polynomial regression model (POLYcubr), at least 3d + 1 points are needed. Otherwise the least squares problem is underdetermined.
* When using mixture models in general at least the minimum number of required points plus one additional point are needed because of the leave-one-out cross-validation. For example, if using MIX RcKg for a d-dimensional problem, at least d + 2 points are required (first order regression polynomial needs at least d + 1 points, and one additional point is needed because of the leave-one-out cross-validation).
* When using the mixture ’MIX RcPc’ or ’MIX KgPc’, at least 2+3d+(d )+(d ) points are needed.
2 3
Otherwise the least squares problem is underdetermined.
* When using the mixture ’MIX KgPqr’ at least 2 + 2d points are needed. Otherwise the least squares problem is underdetermined.
* When using the mixture ’MIX KgPcr’ or ’MIX KgPc’, at least 2+3d points are needed. Otherwise the least squares problem is underdetermined.
In numerical experiments it was found that 2(d + 1) points (which is twice the minimum number of points needed to fit an RBF model of dimension d) work in general well. If the total number of allowed function evaluations is very low, it might however be better to use fewer points in the starting design, and spend more evaluations in the iterative improvement. If number startpoints is not given, the default value 2(d + 1) is used.
Additionally, the user can define m point(s) s/he wants to add to the initial experimental design. The points must be given in matrix form (m × d), and the points are added at the beginning of the starting design.
6.6 Input Example
The following example calls the surrogate model toolbox for finding the minimum of the six-dimensional Hartmann function defined in the file datainput hartman6.m. The maximum number of function evaluations is set to 300, the surrogate model to be used is a mixture of the cubic radial basis function interpolant and the kriging model with Gaussian correlation function (’MIX RcKg’). The candidate point sampling strategy is used (’CAND’), and the initial experimental design is built with a symmetric Latin hypercube design (’SLHD’) with 15 points. Starting points are not given.
8
The user is encouraged to try out the example by typing into the Matlab command window (make sure the location of the files is known to Matlab’s search path):
>> SurrogateModelModule_v1(’datainput_hartman6’,300,’MIX_RcKg’, ’CAND’, ’SLHD’, 15);
7 File Description
This section contains the input and output variables of the single functions.
7.1 SurrogateModelModule v1.m
Input: See Table 1.
Output: None. The algorithm saves the results in the structure Data in the file Results.mat.
This is the main file from which the algorithm is started. At first the user input is checked for correctness. If all input is correct, the initial experimental design is generated (file StartingDesign.m). If there are integrality constraints for the variables, the corresponding values of the variables in the initial design are rounded to the closest integers. The computationally expensive function evaluations are done at the points in the initial experimental design1. Depending on whether there are only continuous, only integer, or mixed-integer variables, the corresponding optimization function is called (OptimizationPhase continuous.m, OptimizationPhase integer.m, OptimizationPhase mixedinteger.m).
7.2 StartingDesign.m
Input
Variable Description
initial design string, name of design strategy (see Table 8)
number startpoints integer, number of points desired for the initial experimental design
Data structure, contains all problem information
Output: Matrix with points in initial experimental design:
where xij is the jth component of the ith sample point. StartingDesign.m calls either one of the functions
Function Description
lhsdesign.m Matlab’s built-in latin hypercube design
SLHD.m symmetric Latin hypercube design [13]
bestlh.m space-filling design, implementation by [2] cornerpoints.m corner point strategy
1Note that this may take a while depending on the time required for one function evaluation. Alternatively, the evaluations can be done in parallel. Uncomment the parfor loop in that case as indicated in the code.
9
7.3 OptimizationPhase continuous.m
This function is executed when the optimization problem has only continuous variables.
Input:
Variable Description
Data structure array, contains the optimization problem information
Surrogate string, determines which surrogate model to be used SampleStrategy string, determines strategy for selecting the next sample site
maxeval integer, maximum number of allowed function evaluations
Output: updated structure array Data containing all information about the problem and solution
The function starts by assigning the parameter tolerance, which is used to decide when the distance of two points is so low that they are considered equal. The value can be changed by the user as desired. Note however that ill-conditioning of the design matrix may become a problem if the value is reduced. The best point and the corresponding function value in the initial experimental design are determined and stored in the variables Data.xbest and Data.fbest, respectively. These values are being updated throughout the optimization. Large function values are replaced by the median of all function values obtained so far in order to prevent the response surface from oscillating wildly.
The desired surrogate model is fit to the data (Data.S and Data.Ymed). For fitting the kriging models the Matlab toolbox DACE [7] has been used, and the user is referred to the DACE manual for instructions of use and parameter settings. Note that due to ill-conditioning of the sample site matrix the DACE toolbox may fail, in which case the algorithm fails. The DACE toolbox issues an error and the user may want to use a different surrogate model. For fitting the multivariate adaptive regression spline (MARS), the Matlab toolbox ARESLab [5] is used. The user is referred to the ARESlab manual for further details. If mixture models are to be used, several surrogate models have to be fit to the data. Dempster-Shafer theory is used to determine the influence each model should have in the mixture [8]. The user may define other (mixture) surrogate models where indicated in the code.
The surrogate models for which implementations exist are summarized in Tables 3, 4, 5, and 6, respectively.
7.3.1 FitSurrogateModel.m
This function fits the response surface(s) to the given data.
Input:
Variable Description
Data structure, contains all information about the problem Surrogate string, containing surrogate model to be used
10
Output:
Variable Description
lambda, gamma vectors, parameters of RBF model; lambda=gamma=[] if RBF model not used
dmodel structure, parameters for kriging model; dmodel=[] if kriging model not used
mmodel structure, parameters for MARS model; mmodel=[] if MARS model not used
beta vector, parameters of polynomial regression model; beta=[] if polynomial model not used
w m vector, contains the weights for the models in mixtures; w m=[] if no mixture model used
RBF.m and RBF eval.m
The parameters of the radial basis function (RBF) interpolant are computed by the function RBF.m,
and RBF eval.m predicts the objective function values using the RBF surrogate model.
Input RBF.m:
Variable Description
S matrix, contains the sample points
Y column vector, contains function values corresponding to points in S
flag string, determines what type of RBF interpolant is used (see Table 3)
Output RBF.m:
Variable Description
lambda column vector, contains multipliers of radial basis functions
gamma column vector, contains parameters for polynomial tail
11
Input RBF eval.m:
Variable Description
X matrix, contains points where objective function value will be predicted
S matrix, contains the sample points already evaluated
lambda column vector, contains multipliers of radial basis functions
gamma column vector, contains parameters for polynomial tail
flag string, determines what type of RBF interpolant is used (see Table 3)
Output RBF eval.m: vector Yest containing predicted objective function values
POLY.m and POLY eval.m
The parameters of the polynomial regression model are computed by the function POLY.m, and
POLY eval.m predicts the objective function values.
Input POLY.m:
Variable Description
S matrix, contains the sample points
Y column vector, contains function values corresponding to points in S
flag string, determines the order of the polynomial (lin, quad, quadr, cub, cubr)
Output POLY.m: vector b containing the parameters of the polynomial regression model Input POLY eval.m:
Variable Description
X matrix, contains points where objective function value will be predicted
b column vector, contains parameters
flag string, determines the order of the polynomial (lin, quad, quadr, cub, cubr)
Output POLY eval.m: vector Yest containing predicted objective function values
dacefit.m and predictor.m
See the tutorial for the Matlab toolbox DACE [7] for input and output arguments, and parameter
settings. The output is stored in the structure variable dmodel.
aresbuild.m and arespredict.m
See the tutorial for the Matlab toolbox ARESLab [5] for input and output arguments, and parameter
settings. The output is stored in the structure variable mmodel.
12
7.3.2 CandidatePointSampling.m
The function is called when the candidate point sampling strategy (setting ’CAND’) is used. The function creates candidates for the next sample site by perturbing the best point found so far, and by uniformly selecting points from the whole variable domain (function Perturbation.m). The objective function values of the candidates are predicted using the response surface, and the distances of the candidate points to the set of already sampled points are computed. Based on these two criteria the best candidate point is selected for doing the next objective function evaluation.
Input:
Variable Description
Data structure, contains all information about the optimization problem
maxeval integer, maximum number of allowed function evaluations
Surrogate string, surrogate model type to be used
lambda,gamma vectors, parameters of RBF model; lambda=gamma=[] if RBF model not used
dmodel structure, parameters for kriging model; dmodel=[] if kriging model not used
mmodel structure, parameters for MARS model; mmodel=[] if MARS model not used
beta vector, parameters of polynomial regression model; beta=[] if polynomial model not used
w m vector, contains the weights for the models in mixtures; w m=[] if no mixture model used
tolerance scalar, distance when two points are considered equal
Output: updated structure Data with all information about the problem.
Perturbation.m
The function generates candidate points for the next sample site by randomly adding or subtracting perturbations to randomly chosen variables of the best point found so far. Furthermore, candidates that are uniformly selected from the whole variable domain are generated.
Input:
Variable Description
Data structure, contains all information about the problem NCandPoint integer, number of points for each candidate point group sigma stdev vector, contains perturbation ranges (small, medium, large)
P scalar, perturbation probability for each variable
Output: matrix CandPoint of size (2·NCandPoint×d), where d is the problem dimension.
13
PredictFunctionValues.m
The function calls RBF eval., POLY eval., predictor.m, and/or arespredict.m, respectively for doing
objective function value predictions at the candidate points.
Input:
Variable Description
Data structure, contains all information about the optimization problem
Surrogate string, surrogate model type to be used
CandPoint matrix, contains points for doing the objective function value predictions lambda,gamma vectors, parameters of RBF model; lambda=gamma=[] if RBF model not used
dmodel structure, parameters for kriging model; dmodel=[] if kriging model not used
mmodel structure, parameters for MARS model; mmodel=[] if MARS model not used
beta vector, parameters of polynomial regression model; beta=[] if polynomial model not used
w m vector, contains the weights for the models in mixtures; w m=[] if no mixture model used
Output: vector CandValue with predicted objective function value for each point in CandPoint. Distancecriterion.m
The function computes the distance of each candidate point to the set of already sampled points, and scales the values to the interval [0,1], where the largest distance is scaled to 0 and the smallest distance is scaled to 1.
Input:
Variable Description
S matrix, contains all already sampled points
CandPoint matrix, contains all candidate points for next function evaluation
Output:
Variable Description
ScaledDist vector, contains distances of candidate points to the set of already sampled points, scaled to [0,1]
Dist vector, contains distances of candidate points to the set of already sampled points
PredictedValueCrit.m
Scales the predicted objective function values at the candidate points to the interval [0,1], where the lowest predicted objective function value is scaled to 0, and the largest predicted value is scaled to 1.
Input: Vector CandValue containing the predicted objective function values of all candidate points.
Output: Vector ScaledCandValue containing the predicted objective function values scaled to the interval [0,1].
14
7.3.3 SurfaceMinSampling.m
The function is called when the minimum site of the response surface is used as sampling strategy (setting ’SURFmin’). The dynamically dimensioned search algorithm [12] is used for finding the minimum of the surface.
Input:
Variable Description
Data structure, contains all information about the optimization problem
maxeval integer, maximum number of allowed function evaluations
Surrogate string, surrogate model type to be used
lambda,gamma vectors, parameters of RBF model; lambda=gamma=[] if RBF model not used
dmodel structure, parameters for kriging model; dmodel=[] if kriging model not used
mmodel structure, parameters for MARS model; mmodel=[] if MARS model not used
beta vector, parameters of polynomial regression model; beta=[] if polynomial model not used
w m vector, contains the weights for the models in mixtures; w m=[] if no mixture model used
tolerance scalar, distance when two points are considered equal
Output: updated structure Data with all information about the problem.
ODDS.m
Ordinary dynamically dimensioned search algorithm [12] for finding the minimum of the response surface.
Input:
Variable Description
Data structure, contains all information about the optimization problem
x0 vector, starting guess for search
objective string, surrogate model type to be used
lambda,gamma vectors, parameters of RBF model; lambda=gamma=[] if RBF model not used
dmodel structure, parameters for kriging model; dmodel=[] if kriging model not used
mmodel structure, parameters for MARS model; mmodel=[] if MARS model not used
beta vector, parameters of polynomial regression model; beta=[] if polynomial model not used
w m vector, contains the weights for the models in mixtures; w m=[] if no mixture model used
tolerance scalar, distance when two points are considered equal
Output: vector xbest, best point found during the optimization routine.
7.3.4 ExpImprovementSampling.m
The function is called when the point that maximizes the expected improvement is used as sampling strategy (setting ’EImax’). Matlab’s genetic algorithm is used to maximize the expected improvement. Parts of the implementation are from [2].
Input:
Variable Description
Data structure, contains all information about the optimization problem maxeval integer, maximum number of allowed function evaluations
15
Output: updated structure Data with all information about the problem.
distanceupdate.m
Computes the pairwise distances of all sample points.
Input: Structure Data containing all information about the problem
Output:
Variable Description
D matrix, contains pairwise distances between variable values in each dimension
ij matrix, contains indices of points for which pairwise distances computed
likelihood new.m
Implementation by A.I.J. Forrester [2]. See the book Engineering Design via Surrogate Modelling -
A Practical Guide by [2] for usage and file description.
ExpectedImprovement.m
Computes the expected improvement at a given point n the variable domain.
Input:
Variable Description
Data structure, contains information about the problem
x vector, the point in the variable domain at which the expected improvement is computed
Output: scalar ExpImp, the expected improvement at the point x. See the book Engineering Design via Surrogate Modelling - A Practical Guide by [2] for usage and file description.
7.3.5 BumpinessMinSampling.m
The function is called when the minimum site of the bumpiness measure is used as sampling strategy (setting ’BUMPmin’).
Input:
Variable Description
Data structure, contains all information about the optimization problem
maxeval integer, maximum number of allowed function evaluations
Surrogate string, surrogate model type to be used
lambda,gamma vectors, parameters of RBF model; lambda=gamma=[] if RBF model not used
tolerance scalar, distance when two points are considered equal
Output: updated structures array with all information about the problem.
16
bumpiness measure.m
The function is minimized in order to find the point where it is most reasonable that the objective function takes on the given target value.
Input:
Variable Description
x vector at which the bumpiness measure is being computed
Data structure, contains all information about the optimization problem
flag string, containing information of the type of RBF model to be used
target scalar, target value for objective function
tolerance scalar, distance when two points are considered equal
lambda,gamma vectors, parameters of RBF model
Output: bumpiness measure value hn.
7.3.6 ScoreMinSampling.m
The function is called when the minimum site of the score function is used as sampling strategy (setting ’SCOREmin’).
Input:
Variable Description
Data structure, contains all information about the optimization problem
maxeval integer, maximum number of allowed function evaluations
Surrogate string, surrogate model type to be used
lambda,gamma vectors, parameters of RBF model; lambda=gamma=[] if RBF model not used
dmodel structure, parameters for kriging model; dmodel=[] if kriging model not used
mmodel structure, parameters for MARS model; mmodel=[] if MARS model not used
beta vector, parameters of polynomial regression model; beta=[] if polynomial model not used
w m vector, contains the weights for the models in mixtures; w m=[] if no mixture model used
tolerance scalar, distance when two points are considered equal
Output: updated structure Data with all information about the problem.
Multi DDS.m
Implementation of the DDS algorithm [12] for finding the minimum of the scoring function. Input:
Variable Description
Data structure, contains all information about the optimization problem
valueweight scalar, weight for the response surface criterion
mindistweight scalar, weight for the distance criterion
tolerance scalar, distance when two points are considered equal
myfun function handle to the (mixture) surrogate model
Output: best point found during optimization xbest
17
7.4 OptimizationPhase integer.m
This function is executed when the optimization problem has only integer variables. The algorithm is able to solve problems that have in addition to the integer and box constraints also black-box constraints. The constraints must be defined in form of function handles in the Data file as variables Data.constraint{1}, Data.constraint{2}, etc. See the file datainput G4 I.m for an example. If there are black-box constraints, and if there is no feasible point contained in the starting design, then the function SOI OP1.m is executed. After a first feasible point has been found (or the problem does not have any black-box constraints), SOI OP2.m is executed. If the problem has integer constraints for all variables, only the candidate point sampling strategy can be used.
7.4.1 SOI OP1.m
This function tries to find a first feasible point of an optimization problem with additional black-box constraints by minimizing a constraint violation function. The candidate point approach is used during the optimization. SOI OP1.m stops after the first feasible point has been found, or the maximum number of allowed function evaluations has been reached.
Input:
Variable Description
Data structure, contains all information about the optimization problem
maxeval integer, maximum number of allowed function evaluations
P scalar, perturbation probability for creating candidates
stdev int vector of integers, perturbation ranges
Surrogate string, contains the name of the (mixture) surrogate model to be used for predictions
Output: Updated structure Data with all problem information.
Perturbation SOI.m
This function generates the candidate points for the next expensive function evaluation. The best point found so far is perturbed for generating the candidates in the first group. For candidates in the second group are uniformly selected integer points form the whole variable domain.
Input:
Variable Description
Data structure, contains all information about the optimization problem NCandidates integer, number of candidates
stdev int vector of integers, perturbation ranges
P scalar, perturbation probability for creating candidates
Output: Matrix CandPoint with candidate points.
18
7.4.2 SOI OP2.m
This function is used after a first feasible point of an optimization problem with additional black-box constraints has been found (or if the problem does not have any additional constraints). The candidate point approach is used during the optimization.
Input:
Variable Description
Data structure, contains all information about the optimization problem
maxeval integer, maximum number of allowed function evaluations
P scalar, perturbation probability for creating candidates
stdev int vector of integers, perturbation ranges
Surrogate string, contains the name of the (mixture) surrogate model to be used for predictions
Output: Updated structure Data with all problem information.
7.5 OptimizationPhase mixedinteger.m
This function is called when mixed-integer black-box problems are considered. If the mixed-integer problem has additional black-box constraints, the user must supply at least one feasible point in the variable starting point in the input of SurrogateModelModule v1.m. For mixed-integer problems only the candidate point sampling approach can be used.
Input:
Variable Description
Data structure, contains all information about the optimization problem
Surrogate string, contains the name of the (mixture) surrogate model to be used for predictions
maxeval integer, maximum number of allowed function evaluations
Output: Updated structure Data with all problem information.
7.5.1 Perturbation SOMI.m
This function is used to create candidate points by uniformly selecting points from the whole variable domain, and by perturbing the best feasible point found so far as follows
* randomly add or subtract small, medium, or large perturbations only to continuous variables
* randomly add or subtract small, medium, or large perturbations only to integer variables
* randomly add or subtract small, medium, or large perturbations to continuous and integer variables
19
Input:
Variable Description
Data structure, contains all information about the optimization problem NCandidates integer, number of candidates
stdev int vector of integers, perturbation ranges for integer variables
stdev cont vector, perturbation ranges for continuous variables
P scalar, perturbation probability for creating candidates
Output: Matrix CandPoint with candidate points.
8 Examples
This section contains several examples for continuos, integer, and mixed-integer black-box optimization problems. The input data files are supplied, and the user is encouraged to take a look at these files when defining data files for own problems. The examples have computationally cheap objective functions in order to reduce the computation time when experimenting with the code.
8.1 Examples for Continuous Problems
The algorithm is applicable to box-constrained continuous problems. If the problem has additional constraints, they should be incorporated with a penalty term in the objective function. The input
data files below are all for continuous problems.
* datainput Ackley15.m
* datainput Ackley30.m
* datainput Branin.m
* datainput DixonPrice15.m
* datainput hartman3.m
* datainput hartman6.m
* datainput Levy20.m
* datainput Michalewicz25.m
* datainput Powell24.m
* datainput Rastrigin12.m
* datainput Rastrigin30.m
* datainput Rosenbrock10.m
* datainput Rosenbrock20.m
* datainput Schoen 10 4 3.m
* datainput Schoen 17 2 5.m
* datainput Shekel5.m
* datainput Shekel7.m
* datainput Shekel10.m
* datainput Shubert.m
20
* datainput Sphere27.m
* datainput Zakharov.m
For the six-dimensional Hartmann function, type
>> SurrogateModelModule_v1(’datainput_hartman6’,200,’RBFtps’,’BUMPmin’,’LHS’,7);
With this input the following settings are used:
* 200 function evaluations are allowed
* the thin-plat spline radial basis function model is used as response surface (RBFtps)
* the bumpiness measure (BUMPmin) is used as sampling strategy
* the Latin hypercube sampling strategy (Matlab’s built-in Latin hypercube sampling) is used (LHS) as experimental design strategy
* 7 points are used in the initial starting design
For the 4-dimensional Schoen function with 7 local minima, type
>> SurrogateModelModule_v1(’datainput_Shekel7’,400,’MIX_RcPc’,’CAND’,’SLHD’,24);
With this input the following settings are used:
* 400 function evaluations are allowed
* the mixture model of cubic RBF and full cubic polynomial model is used as response surface (MIX RcPc)
* the candidate point sampling strategy (CAND) is used as sampling strategy
* the symmetric Latin hypercube sampling strategy (SLHD) is used as experimental design strategy
* 24 points are used in the initial starting design
8.2 Examples for Integer Problems
The algorithm is applicable to optimization problems where all variables have integer constraints, and may have additional black-box constraints. The black-box constraints are treated with a penalty approach. Currently only the candidate point sampling strategy can be used with purely integer problems. The example data input files for integer problems all end in I.m:
* datainput convex I.m
* datainput ex I.m
* datainput ex1221 I.m
* datainput G1 I.m
* datainput G2 I.m
* datainput G4 I.m
* datainput G6 I.m
* datainput G9 I.m
* datainput GWSS20 I.m
* datainput hmittelmann I.m
* datainput hydropower 1plant1 I.m
21
* datainput hydropower 1plant2 I.m
* datainput hydropower 1plant3 I.m
* datainput hydropower 2plant1 I.m
* datainput hydropower 2plant2 I.m
* datainput hydropower 2plant3 I.m
* datainput linearproblem I.m
* datainput nvs09 I.m
* datainput nvs09alt I.m
* datainput Rastrigin12 I.m
* datainput Rastrigin30 I.m
* datainput throughput I.m
* datainput throughput small I.m
For the convex problem, type for example
>> SurrogateModelModule_v1(’datainput_convex_I’,300,’MARS’,’CAND’,’LHS’,10);
With this input the following settings are used:
* 300 function evaluations are allowed
* the MARS model is used as response surface (MARS)
* the candidate point sampling strategy (CAND) is used as sampling strategy (this is the only sampling strategy that can be used with integer problems)
* Matlab’s built-in Latin hypercube sampling strategy (LHS) is used as experimental design strategy
* 10 points are used in the initial starting design
For the small throughput problem, type for example
>> SurrogateModelModule_v1(’datainput_throughput_small_I’,500,’KRIGgauss2’,’CAND’,’LHS’,21);
With this input the following settings are used:
* 500 function evaluations are allowed
* the kriging model with second order regression polynomial and Gaussian correlation is used as response surface (KRIGgauss2)
* the candidate point sampling strategy (CAND) is used as sampling strategy (this is the only sampling strategy that can be used with integer problems)
* Matlab’s built-in Latin hypercube sampling strategy (LHS) is used as experimental design strategy
* 21 points are used in the initial starting design
22
8.3 Examples for Mixed-Integer Problems
The algorithm can be applied to mixed-integer problems that may have black-box constraints in addition to the variable upper and lower bounds. For constrained problems the user must supply at least one feasible starting point. During the computational experiments described in [9] the starting points given in the files StartPoints 1.mat, StartPoints 2.mat, etc. in Table 9 have been used. Each file contains 30 feasible points, and for each algorithm trial one of these points has been used. The data input files for mixed integer problems all end in MI.m. For mixed-integer problem currently only the candidate point sampling strategy works. For testing the surrogate model algorithm for test problem datainput G4 MI.m, for example, type
>> load(’StartPoints_7.mat’);
>> x0=StartPoints(1,:);
>> SurrogateModelModule_v1(’datainput_G4_MI’, [],[],[],[],[],x0)
In this example only the problem defining data file is given and the starting point. For all other input parameters the default values are used.
Table 9: Problem files and feasible starting points for mixed-integer problems
Test Problem File Feasible Starting Points
datainput BermanAshrafi MI.m StartPoints 1.mat
datainput convex MI.m StartPoints 2.mat
datainput ex MI.m StartPoints 13.mat
datainput ex1221 MI.m StartPoints 3.mat
datainput Floudas MI.m StartPoints 4.mat
datainput G2 MI.m StartPoints 6.mat
datainput G4 MI.m StartPoints 7.mat
datainput G6 MI.m StartPoints 8.mat
datainput G9 MI.m StartPoints 9.mat
datainput linearproblem MI.m StartPoints 11.mat
datainput nvs09 MI.m StartPoints 14.mat
datainput nvs09alt MI.m StartPoints 15.mat
datainput Rastrigin12 MI.m StartPoints 16.mat
datainput Rastrigin12alt MI.m StartPoints 12.mat
datainput Rastrigin30 MI.m StartPoints 10.mat
datainput Yuan MI.m StartPoints 5.mat
For the problem G2, type for example
>> load(’StartPoints_6.mat’);
>> x0=StartPoints(11,:);
>> SurrogateModelModule_v1(’datainput_G2_MI’,300,’RBFtps’,’CAND’,’SLHD’,50,x0);
With this input the following settings are used:
* 300 function evaluations are allowed
* the radial basis function with thin-plate spline is used as response surface (RBFtps)
* the candidate point sampling strategy (CAND) is used as sampling strategy (this is the only sampling strategy that can be used with mixed-integer problems)
* the symmetric Latin hypercube sampling strategy (SLHD) is used as experimental design strategy
* 50 points are used in the initial starting design
23
For the Yuan problem, type for example
>> load(’StartPoints_5.mat’);
>> x0=StartPoints(19,:);
>> SurrogateModelModule_v1(’datainput_Yuan_MI’,300,’MIX_RcKgM’,’CAND’,’SLHD’,22,x0);
With this input the following settings are used:
* 300 function evaluations are allowed
* the mixture of cubic radial basis function, kriging with gaussian correlation function, and MARS is used as response surface (MIX RcKgM)
* the candidate point sampling strategy (CAND) is used as sampling strategy (this is the only sampling strategy that can be used with mixed-integer problems)
* the symmetric Latin hypercube sampling strategy (SLHD) is used as experimental design strategy
* 22 points are used in the initial starting design
9 So you want to define your own surrogate model?
The files that need to be altered when defining an own (mixture) surrogate model are
* SurrogateModelModule v1.m
* FitSurrogateModel.m
* DempsterFor2models.m or DempsterFor3models.m
* PredictFunctionValues.m
* SurfaceMinSampling.m (if necessary)
* ScoreMinSampling.m (if necessary)
9.1 Example
Suppose you want to add a mixture model consisting of the thin-plate spline RBF model and a reduced quadratic polynomial model. Let’s abbreviate the model by MIX RBFtpsPqr. We have to alter the file SurrogateModelModule v1.m at the indicated position (see the code) as follows:
if isempty(surrogate model)
display(’Using default surrogate model.’)
surrogate model=’RBFcub’; %if no surrogate model defined, use cubic RBF
else %add own surrogate model to the list
if ?any(strcmp(surrogate model,{’RBFcub’, ’RBFtps’, ’RBFlin’,’KRIGexp0’,...
’KRIGexp1’,’KRIGexp2’,’KRIGgexp0’, ’KRIGgexp1’, ’KRIGgexp2’,...
’KRIGgauss0’, ’KRIGgauss1’,’KRIGgauss2’,’KRIGlin0’,’KRIGlin1’,...
’KRIGlin2’,’KRIGspline0’,’KRIGspline1’,’KRIGspline2’,’KRIGsphere0’,...
’KRIGsphere1’,’KRIGsphere2’,’KRIGcub0’,’KRIGcub1’,’KRIGcub2’,...
’POLYlin’, ’POLYquad’, ’POLYquadr’, ’POLYcub’, ’POLYcubr’,...
’MARS’, ’MIX RcKg’,’MIX RcM’,’MIX RcPc’, ’MIX KgM’,...
’MIX KgPc’, ’MIX KgPqr’, ’MIX KgPcr’,’MIX RcKgM’,’MIX RBFtpsPqr’}))
error(’The surrogate model you want to use is not contained in the toolbox. Check spelling.’)
end
In the file FitSurrogateModel.m the following branch in the if-elseif tree must be added under the comment
% add more 2-model combinations here if necessary. Update function ”DempsterFor2models” accordingly:
elseif strcmp(Surrogate,’MIX RBFtpsPqr’)
[lambda, gamma]=RBF(Data.S,Data.Ymed,’TPS’); % RBF-tps
beta=POLY(Data.S,Data.Ymed,’quadr’); %reduced quadratic polynomial
w m=DempsterFor2models(Data, Surrogate); %uses dempster shafer theory to adjust model weights
In the file DempsterFor2models.m the following branch in the if-elseif tree must be added under the comment
%other 2-model mixtures here:
%Mixture model: thin-plate spline RBF & reduced quadratic polynomial
elseif strcmp(Surrogate,’MIX RBFtpsPqr’)
if ?forcekfold %leave-one-out cross-validation
for jj=1:m
[lambda, gamma]=RBF([Data.S(1:jj-1,:);Data.S(jj+1:end,:)],...
[Data.Ymed(1:jj-1,:);Data.Ymed(jj+1:end,:)],’TPS’); % RBF
yPred all(jj,1)=RBF eval(Data.S(jj,:),[Data.S(1:jj-1,:);...
Data.S(jj+1:end,:)],lambda,gamma,’TPS’); %repredict jj-th objective function value
reduced quadratic polynomial model prediction
beta=POLY([Data.S(1:jj-1,:);Data.S(jj+1:end,:)],...
[Data.Ymed(1:jj-1,:);Data.Ymed(jj+1:end,:)],’quadr’);
yPred all(jj,2)=POLY eval(Data.S(jj,:),beta,’quadr’);
end
else %k-fold cross-validation
Ss=Data.S(mix,:); %resorted sample sites
Ys=Data.Ymed(mix);
for jj=1:ceil(m/kfold)
%validation set
if jj*kfold <=m
validation S=Ss((jj-1)*kfold+1:jj*kfold,:);
else %left-overs
validation S=Ss((jj-1)*kfold+1:end,:);
end
%validation set
if jj ==1 %first group
trainset S=Ss(jj*kfold+1:end,:);
trainset Y=Ys(jj*kfold+1:end,:);
elseif 2<=jj && jj<= floor(m/kfold) && mod(m,kfold) >0 ||...
2<=jj && mod(m,kfold)==0 && jj*kfold<m
%group 2 till last full group
trainset S=[Ss(1:(jj-1)*kfold,:);Ss(jj*kfold+1:end,:)];
trainset Y=[Ys(1:(jj-1)*kfold,:);Ys(jj*kfold+1:end,:)];
else %left-overs
trainset S=Ss(1:(jj-1)*kfold,:);
trainset Y=Ys(1:(jj-1)*kfold,:);
end
if 1<=jj && jj <= floor(m/kfold) $& mod(m,kfold) >0 ||...
2<=jj && mod(m,kfold)==0 && jj*kfold<m
% RBF model for prediction
[lambda, gamma]=RBF(trainset S,trainset Y,’TPS’);
yPred all((jj-1)*kfold+1:jj*kfold,1)=RBF eval(validation S,...
trainset S,lambda,gamma,’TPS’);
25
% rduced quadratic polynomial model prediction
beta=POLY(trainset S,trainset Y,’quadr’); %Polynomial
yPred all((jj-1)*kfold+1:jj*kfold,2)=POLY eval(validation S,beta,’quadr’);
else %left-overs
id a=(jj-1)*kfold+1;
if mod(m,kfold)>0
id e=(jj-1)*kfold+mod(m,kfold);
else
id e=jj*kfold;
end
% RBF model for prediction
[lambda, gamma]=RBF(trainset S,trainset Y,’TPS’);
yPred all(id a:id e,1)=RBF eval(validation S,trainset S,...
lambda,gamma,’TPS’);
% reduced quadratic polynomial model prediction
beta=POLY(trainset S,trainset Y,’quadr’);
yPred all(id a:id e,2)=POLY eval(validation S,beta,’quadr’);
yPred all=sortrows([yPred all,mix(:)],Data.numberOfModels+1);
yPred all=yPred all(:,1:Data.numberOfModels);
end
end
end
If a new three-model mixture is introduced, the file DempsterFor3models.m must accordingly be altered.
The file PredictFunctionValues.m has to be extended by adding a branch to the if-elseif tree under the comment
%add more surrogate models here if desired:
elseif strcmp(Surrogate,’MIX RBFtpsPqr’) %Mixture model: RBF with tps & reduced quadratic polynomial
CandValue RBF = RBF eval(CandPoint,Data.S,lambda,gamma,’TPS’); % objective function value prediction with cubic RBF model for all candidate points
CandValue Pqr=POLY eval(CandPoint,beta,’quadr’); %objective function value prediction with reduced quadratic polynomial for all candidate points
CandValue= w m(1)*CandValue RBF+w m(2)*CandValue Pqr; %weighted objective function value prediction
The file SurfaceMinSampling.m need to be altered only if the minimum of the response surface is chosen as sampling strategy (setting SURFmin). The if-elseif tree has to be extended under the comment
%add other 2-model mixtures here:
elseif strcmp(Surrogate,’MIX RBFtpsPqr’) %mixture model of Kriging with gaussian correlation and 1st
order regression polynomial, and reduced quadratic polynomial
myfun1=@(x)RBF eval(x,Data.S,lambda,gamma,’TPS’); % cubic RBF model
myfun2=@(x)POLY eval(x,beta,’quadr’); %reduced quadratic model
myfun=@(x)w m(1)*myfun1(x)+w m(2)*myfun2(x); %weighted mixture
The file ScoreMinSampling.m needs to be altered only if the minimum of the scoring function is used as sampling criterion. In that case the if-elseif tree needs to be extended under the comment
%add more 2-model mixtures here as follows:
26
elseif strcmp(Surrogate,’MIX RBFtpsPqr’) %mixture model of Kriging with gaussian correlation and 1st
order regression polynomial, and reduced cubic polynomial
myfun1=@(x)RBF eval(x,Data.S,lambda,gamma,’TPS’); % cubic RBF model
myfun2=@(x)POLY eval(x,beta,’quadr’); %reduced quadratic model
myfun=@(x)w m(1)*myfun1(x)+w m(2)*myfun2(x); %weighted mixture
If a completely new surrogate model is to be defined, two functions are needed, namely a first function that computes the model parameters (the output of FitSurrogateModel.m must be adjusted accordingly), and a second function that predicts the objective function values, where the predictions should be scalars.
10 Results
The algorithm saves the results of the optimization in every iteration to the file Results.mat. If the algorithm is not interrupted (e.g. by pressing CTRL+C), the following elements are contained in the saved Data structure (see Table 10).
Table 10: Saved data structure elements
Elements Description
Data.xlow vector with lower variable bound
Data.xup vector with upper variable bounds
Data.dim integer, problem dimension
Data.integer vector with indices of integer variables
Data.continuous vector with indices of continuous variables
Data.objfunction function handle to objective function
Data.constraint cell array with function handles to constraints; only for constrained problems
Data.S matrix with sample sites
Data.Y vector with objective function values corresponding to Data.S
Data.fevaltime vector with time for each function evaluation
Data.gevaltime vector with time for constraint evaluations; only for constrained problems
Data.pen vector with squared constraint violations; only for constrained problems
Data.Feasible binary, 1 if feasible solution found, 0 otherwise; only for constrained problems
Data.Fbest inf scalar, best infeasible function value found; only for constrained problems
Data.xbest inf vector, best infeasible point found; only for constrained problems
Data.xbest vector, best (feasible) point found
Data.Ymed vector with function values where large values have been replaced by median
Data.fbest scalar, best feasible objective function value found
Data.Ypenalized vector with penalty-augmented objective function values; only for constrained problems
Data.Problem string, contains name of data file that specifies the optimization problem
Data.SurrogateModel string, contains name of (mixture) surrogate model used in optimization
Data.SamplingTechnique string, contains name of iterative sampling technique used in optimization
Data.InitialDesign string, contains name of experimental design strategy
Data.NumberStartPoints integer, number or points in initial experimental design
Data.StartingPoint matrix, contains user-defined points that are added to the initial experimental design; only if defined, and for mixed-integer problems
Data.TotalTime scalar, total computation time needed by the algorithm
References
[1] A.J. Booker, J.E. Dennis Jr, P.D. Frank, D.B. Serafini, V. Torczon, and M.W. Trosset. A rigorous framework for optimization of expensive functions by surrogates. Structural Multidisciplinary
27
Optimization, 17:1–13, 1999.
[2] A. Forrester, A. S´obester, and A. Keane. Engineering Design via Surrogate Modelling - A Practical Guide. Wiley, 2008.
[3] H.-M. Gutmann. A radial basis function method for global optimization. Journal of Global Optimization, 19:201–227, 2001.
[4] K. Holmstr¨om, N.-H. Quttineh, and M.M. Edvall. An adaptive radial basis algorithm (ARBF) for expensive black-box mixed-integer constrained global optimization. Journal of Global Optimization, 41:447–464, 2008.
[5] G. Jekabsons. ARESLab: Adaptive Regression Splines toolbox for Matlab. available at http://www.cs.rtu.lv/jekabsons/, 2010.
[6] D.R. Jones, M. Schonlau, and W.J. Welch. Efficient global optimization of expensive black-box functions. Journal of Global Optimization, 13:455–492, 1998.
[7] S.N. Lophaven, H.B. Nielsen, and J. Søndergaard. DACE a Matlab kriging toolbox. Technical report, Technical Report IMM-TR-2002-12, 2002.
[8] J. M¨uller and R. Pich´e. Mixture surrogate models based on Dempster-Shafer theory for global optimization problems. Journal of Global Optimization, 51:79–104, 2011.
[9] J. M¨uller, C.A. Shoemaker, and R. Pich´e. SO-MI: A surrogate model algorithm for computationally expensive nonlinear mixed-integer black-box global optimization problems. Computers and Operations Research, http://dx.doi.org/10.1016/j.cor.2012.08.022, 2012.
[10] R.G. Regis and C.A. Shoemaker. A stochastic radial basis function method for the global optimization of expensive functions. INFORMS Journal on Computing, 19:497–509, 2007.
[11] R.G. Regis and C.A. Shoemaker. A quasi-multistart framework for global optimization of expensive functions using response surface models. Journal of Global Optimization, DOI 10.1007/s10898-012-9940-1, 2012.
[12] B.A. Tolson and C.A. Shoemaker. Dynamically dimensioned search algorithm for computationally efficient watershed model calibration. Water Resources Research, 43:W01413, 16 pages, 2007.
[13] K.Q. Ye, W. Li, and A. Sudjianto. Algorithmic construction of optimal symmetric Latin hyper-cube designs. Journal of Statistical Planning and Inference, 90:145–159, 2000.
28
1
Texas A&M University - Commerce
Department of Biological and Environmental Sciences
Masters Thesis Guidelines
General
The following guidelines are to be followed by all Masters students in the Department of Biological and Environmental Sciences (BESC). While other guidelines may be used, adherence to these guidelines will ensure that the thesis conforms to standards appropriate to the biological and environmental sciences and that are acceptable to the Graduate School and Thesis and Dissertaion Services (TDS). Students should contact their research advisors if they have any questions.
Proposal
All students writing a thesis must submit a thesis proposal to the Graduate School at least one semester prior to graduation. This proposal, to be completed in consultation with the major advisor, must be approved by all members of the thesis committee, the Department Head of BESC, the Dean of Science and Engineering, and the Dean of the Graduate School. The thesis proposal form is available from the Graduate School.
Human Subjects Protection
In preparing your thesis proposal you should be aware that any research that involves human subjects must be in compliance with University Rule 15.99.01.R1 on Human Subjects Protection. A copy of that policy is available at http://www.tamuc.edu/aboutUs/policiesProceduresStandardsStatements/rulesProcedures/15Rese archPrograms/15.99.01.R1HumanSubjectsProtection.pdf
If you have questions concerning compliance, contact your major advisor or Chair of the Institutional Review Board (IRB).
Care and Treatment of Animals used in Research
All students using live animals in their research must have an approved protocol on file with the Institutional Animal Care and Use Committee (IACUC) prior to starting their research. Students must be aware of and agree to comply with all local, state, and federal regulations. Students must be aware that the Department of BESC adheres to the guidelines set forth in the National Institutes of Health’s publication number 92-23, “Guide for the Care and Use of Laboratory Animals” and follows A&M-Commerce Rule15.99.07.R1, Use of Vertebrate Animals A copy of the rule is available at http://www.tamuc.edu/aboutUs/policiesProceduresStandardsStatements/rulesProcedures/15Rese archPrograms/15.99.07.R1.pdf.
A copy of the Guide for the Care and Use of Laboratory Animals can be found here: https://grants.nih.gov/grants/olaw/guide-for-the-care-and-use-of-laboratory-animals.pdf
Failure to ensure adequate care and minimization of suffering to research animals will result in immediate termination of that research. Questions about compliance should be addressed to the major advisor.
2
Use of Biohazardous Agents
All students using biohazardous agents in their research must have an approved protocol on file with the Institutional Biosafety Committee (IBC) prior to starting their research. Students must be aware of and agree to comply with all local, state, and federal regulations. Students must be aware that the Department of BESC adheres to theA&M-Commerce Rule 15.99.06.R1, “Use of Biohazards in Research Teaching and Testing”. A copy of the rule is available at http://www.tamuc.edu/aboutUs/policiesProceduresStandardsStatements/rulesProcedures/15Rese archPrograms/15.99.06.R1.pdf.
Questions about compliance should be addressed to the major advisor.
Preparation of Thesis
Text must be 12-point, Times New Roman, and double-spaced throughout, including figures and tables. All headings should be in bold font. Major headings should be centered and all capitalized. Minor headings are left justified and only the first letter is capitalized. An extra space is left between major headings and the first paragraph following the major heading. No extra spaces should occur between paragraphs or after minor headings. Margins should be 1” on top, bottom, left and right sides. The only exception is the Abstract where the top margin must be 2.5”. All pages must be numbered at the top right side. The Title Page and the Signature Page have no page numbers. However, the pages following the Signature Page will begin with the number “iii” and use the same Roman numeral format ending with the List of Figures (lower case Roman numerals iii, iv, v, vi, etc.). Pages in the main body of the thesis (Introduction, Methods, Results, Discussion, Literature Cited, Appendices, Vita) should be numbered using Arabic numerals (1, 2, etc.).
Common names for organisms may be used, but the Latin name (genus and species) must be specified at their first mention. Latin names must be in italics. Use standard scientific abbreviations for units of measure. Do not use footnotes in the body of the thesis. Do not use direct quotes.
Organization of Sections
As per the graduate school guideleins use the following arrangement of materials.
1) Title Page
2) Signature Page
3) Copyright Page (if copyrighting)
4) Abstract
5) Acknowledgments (if applicable)
6) Table of Contents
7) List of Tables (if applicable)
8) List of Figures (if applicable)
9) Body
10) Literature Cited
11) Appendix (if applicable)
12) Vita
Note that sections 1 through 8 use lower case Roman numeral page numbers. All other sections use regular page numbering.
3
Major Headings:
TITLE of the thesis on the Title Page
TITLE of the thesis on the Signature Page
Headings ABSTRACT, ACKNOWLEDGMENTS, TABLE OF CONTENTS, LIST OF
FIGURES, LIST OF TABLES on the preliminary pages
INTRODUCTION
METHODS
RESULTS
DISCUSSION
LITERATURE CITED
APPENDIX designations and titles
VITA on the Vita page. (See the Office of Thesis and Dissertation Services for formatting)
Note that INTRODUCTION, METHODS, RESULTS, DISCUSSION, all form the body of the thesis (section 9 in the preceding list under Organization of Sections).
The rules for major headings are:
- All must be centered at the top of a new page and in ALL CAPITAL letters.
- The headings must be placed one inch from the top of the page.
- Each line of the title must be shorter than the one above it.
- The font must be the same size as the body text.
- Major headings may not be in italic type.
- No punctuation after a major heading.
- All major headings must be handled consistently.
- Major headings over one line are double-spaced between the lines
Subheadings:
Subheadings may be used throughout the thesis when appropriate.
- Use the same font size for all subheadings.
- Use consistent capitalization for each level of subheading.
- The style used for subheadings must clearly show their various levels.
- Maintain consistency in vertical spacing around major headings and subheadings.
- Subheadings and subsections do not need to begin on a new page.
- Subheadings near the bottom of a page must have at least two lines of text under it.
- Subheadings cannot be centered and have all capital letters.
Figures and Tables
Color is acceptable in figures, but black and white is preferred for line and bar graphs. The minimum size for letters and numbers is 7 point type, but fonts equal to or larger than 11 point are preferred with the exception of footnotes. All lettering must be of publishable quality; this requirement includes scanned images. Images must be clear and without blurred or dark areas.
All Figures and Tables in result section are placed on a separate page from the text and accompanied by an appropriate figure caption or table header. Figure captions always go below
4
the figure and do not have to be double spaced. Figures and tables are placed in the thesis on the page immediately following their first mention in the body of the thesis.
Note that all drawings, figures, graphs, maps, etc. are referred to as Figure 1, Figure 2, etc. in the text and in the figure caption. DO NOT use any other terminology (ie. Graph 1, Map 1, Chart 1, Picture 1, Drawing 1, etc.). Figures should be quality graphic images. Graphs must be computer generated and of at least 300 dpi density. Makes sure the fonts on the figure are in Times New Roman.
Tables must be put on the page immediately following their first mention in the text (same rule applies to figures). The text in the table header and column headings do not have to be double spaced. Similarly, the text in any given cell of the table does not have to be double spaced if the text must be wrapped within that cell. However, the rows themselves must by double spaced. An example is provided below (see Table 1).
Note that all tables have a table header. The table header should clearly explain the purpose of the table. Horizontal lines should separate the table header and major column headings from the body of the table. Do not use vertical lines in tables. Do not put a box around the table. Space the columns evenly. Left column must be left justified if it is a text column, otherwise all columns must be centered below the column heading. Align characters and decimal places. Use footnotes on tables where appropriate. Make sure the font in the table is Times New Roman, 12 point, and double-spaced.
Tables or figures longer than one page have the complete header and the number on the first page. Subsequent pages have the table or figure number and the word “Continued” plus the complete header. For example, the second page of a long table should have the following format:
Table 2 Continued. The fledging success of swallows nesting in culverts and bridges in northeast Texas between 1998 and 2002.
Bridges
(mean + std. dev)
Culverts
(mean + std. dev)
Kruskal-Wallis Test X2 (1 df)
p-value
2001 first broods
4.25 + 0.23
5.23 + 1.25
6.55
0.010
2001 second broods
3.61 + 0.33
4.77 + 0.56
1.12
0.320
2002 first broods
5.01 + 0.25
5.11 + 0.66
0.69
0.864
2002 second broods
2.69 + 0.52
3.50 + 0.97
1.35
0.264
5
If the table or figure is placed lengthwise (landscape position), the top of the table or figure must be at the left-hand, “binding” side of the page and facing in the same direction as the figure or table. The page number stays in regular (portrait) position. Use a text box and rotate it 90 degrees.
Below are examples of an appropriately formatted figure and table. Note that each is on a separate page. Note also that the table contains an appropriately formatted footnotes. Tables are the only place in the thesis where footnotes are allowed. These footnotes can be either numerals or letters.
Notice also that both the table and the figure are placed at the top of the page, so that in each case they maintain a 1” top margin.
Examples of title page, signature page, copyright page, abstract, list of tables, list of figures and vita are found in the TDS website and should be followed precisely. The table of contents should differ because the biology department does not follow the same format with regard to sections as the graduate school. An example of a the format for a table of contents is on page 8 of this document.
APPENDICES
Use appendices only if necessary to include important data not included in the body of the thesis. In Biology, appendices are treated in the same way as tables and come directly after the Literature Cited section. Appendices should be numbered separately from other tables, but the general format of appendices is the same.
There should be a title page with the word APPENDICES centered on the page and directly following the LITERATURE CITED section.
Appendices then follow after that title page and are formatted like normal tables. DO NOT CAPITALIZE the table headings. For multiple page appendices the table heading on subsequent pages should follow the following format:
Appendix 1 Continued. Cell lines used in this study along with their sources.
Cell Line Source
BCE C/D-1b (CRL-2048TM) ATCC
ZF4 (CRL-2050TM) ATCC
jb6 rt101 Thermo Fisher Scientific
Page numbering in appendices continues from the literature cited section.
6
Table 1. Summary of the best fit logistic regression model based on AICc. This procedure was used to predict the survival of monarch eggs or larvae based on the abundance of arthropod groups found on control host plants where the proportion of predators to non-predators was low. Concordance of this model was 31.6%.
Parameter
DF
Estimate
Standard
Error
Wald
Chi-Square
Pr > ChiS
q
Intercept
1
-1.2480
0.2236
31.1489
<0.0001
RIFAa
1
-3.6684
2.4628
2.2188
0.1363
aParameter confidence interval approaches infinity and is overestimated.
Figure 1. Control host plants upon which monarch larvae survived to the third instar had fewer predators relative to the total number of arthropods on the plant than did host plants upon which monarch larvae failed to reach the third instar. Kruskal-Wallis ANOVA, Chi-square Approximation, X2 = 6.7482, df = 1, p = 0.0094. Numbers over bars are sample sizes.
TABLE OF CONTENTS
LIST OF TABLES #
LIST OF FIGURES #
INTRODUCTION #
Subsection (if applicable) #
Subsection (if applicable) #
METHODS #
Subsection (if applicable) #
Subsection (if applicable) #
RESULTS #
Subsection (if applicable) #
Subsection (if applicable) #
DISCUSSION #
Subsection (if applicable) #
Subsection (if applicable) #
LITERATURE CITED #
APPENDICES #
Appendix # #
Appendix # #
VITA #
9
REFERENCES CITED IN THE BODY OF YOUR PAPER
Cite all references in text by author and year. If more than one citation then they should be in chronological (not alphabetical) order:
Jones (1971)
Jones (1971, 1975)
(Jones 1971)
(Jones 1971, 1975)
(Jones 1971, Smith 1973, Davis 1975)
If two authors for the paper:
Smith and Davis (1985) or (Smith and Davis 1985)
If more than two authors, use:
Jones et al. (1976) or (Jones et al. 1976).
Citations of publications by the same author(s) in the same year should be designated alphabetically:
(Smith 1979a, 1979b).
LITERATURE CITED SECTION:
Journal abbreviations in the Literature Cited section should follow those listed in Biological Journals and Abbreviations which can be found here:
https://home.ncifcrf.gov/research/bja/
Be certain that all citations in the text are included in the Literature Cited section and vice versa. Consecutively-paged journal volumes and other serials should be cited by volume, number and pagination. Serials with more than one number and that are not consecutively paged should be cited by number as well (Smithson. Misc. Coll., 37(3):1-30). Citations use hanging paragraph format as shown below.
In your Literature Cited section, alphabetize entries according to the last name of the first
author of each work cited. If you have multiple sources from the same author(s), arrange by year (earlier first); if you have multiple sources from the same author(s), in the same year, add a lower case letter to identify each when citing in the text and when making the bibliographical entry (Jones 1983a; Jones l983b). If you have multiple papers using the same first author but different second authors, then they a listed after any single author papers by that author and in alphabetical order of the second authors and date (and so on if there are many multi-author papers by the same first author).
10
The following are examples of a variety of citations:
JOURNALS ARTICLES:
Jones, T. L. 1971. Vegetational patterns in the Guadalupe Mountains, Texas. Am. J. Bot., 76(3):266-278.
Smith, J. D. 1973. Geographic variation in the Seminole bat, Lasiurus seminolus. J. Mammal., 54(1):25-38.
Smith, J. D. and G. L. Davis. 1985. Bats of the Yucatan Peninsula. Occas. Pap. Mus., Texas Tech Univ., 97:1-36.
Tames, H. W., J. D. Smith, G. L. Davis, and R. W. Hoffstamer. Directional shifts in the magnetic compass of migratory insects. J. Entomol., 85: 489-499.
BOOKS:
Jones, T. L. 1975. An introduction to the study of plants. John Wiley & Sons, New York, xx+386 pp.
CHAPTER IN A BOOK:
Jones, T. L., A. L. Bain and E. C. Burns. 1976. Grasses of Texas. Pp. 205-265, in Native grasses of North America (R. R. Dunn, ed.), Univ. Texas Studies, 205:xx+1-630.
11
ABSTRACT
The abstract should start on the page following the signature sheet. The abstract should
follow the graduate school guidelines.
ACKNOWLEDGMENTS
Acknowledgments should follow the graduate school guidelines.
TABLE OF CONTENTS
The Table of should follow the graduate school guidelines.
LIST OF TABLES
The list of tables should follow the graduate school guidelines.
LIST OF FIGURES
The list of figures should follow the graduate school guidelines.
INTRODUCTION
The introduction starts on a new page after the last page of the list of figures. The heading, INTRODUCTION, is centered a the top of the page. Arabic numbers (e.g. 1, 2, 3, etc.) are now used for page numbers, starting with 1. The page numbers continue and are numbered consecutively through all subsequent sections to the end of the Vita. The page number for the first page of each major section, all other pages are numbered in the upper right cornter of the page (except figure pages which are unnumbered).
The introduction should normally start with a literature review). This should lead into a description of the purpose or objectives for the current study, significance, and the hypotheses to be tested.
MATERIALS AND METHODS
The methods section starts on a new page after the last page of the introduction. The heading, MATERIALS AND METHODS, is centered a the top of the page. This section should describe all of the methods used to collect and analyze the data. Enough detail must be provided that anyone reading the methods should be able to replicate the study. Describe any apparatus that was used and provide a figure if appropriate. Describe what was measured and how it was measured. For field studies there should be a description of the study areas used. For lab studies, the source of materials must be identified. Be sure to identify the make and model number of important pieces of equipment and the suppliers of critical reagents.
RESULTS
The results section starts on a new page after the last page of the methods section. The heading, RESULTS, is centered a the top of the page. This section should describe the results of the study. For most projects the data are presented as figures and tables. The text of the results section should describe the important trends and/or features shown by the figures and tables. Do not simply reiterate all the details of the table or figure. Provide statistical analyses whenever appropriate.
12
DISCUSSION
The discussion starts on a new page after the last page of the results section. This section describes the implications of the results relative to the existing literature. Provide interpretation of the results and support these statements with corroborative evidence from current literature. The final paragraphs should provide a brief summary and possible future directions for research.
LITERATURE CITED
The literature cited starts on a new page after the last page of the discussion. The
heading, LITERATURE CITED, should be centered and in upper case. The literature cited must
be in alphabetical order and modeled after that used by a prominent biological journal using
author surname alphabetized citation lists (e.g. Ecology). See the examples below:
Diamond, D. D. and F. E. Smeins. 1988. Gradient analysis of remnant true and upper coastal prairie grasslands of North America. Can. J. Bot., 66:2152-2161.
Gould F. W. 1975. The Grasses of Texas. Texas A&M University Press, Texas, vii + 653 pp. Eidson, J. A. and F. E. Smeins. 1999. Texas Blackland Prairies. Pp. 231-237 in Terrestrial
Ecoregions of North America: A Conservation Assessment (T. Ricketts, E. Dinerstein & D.
Olson, eds.), Island Press, Washington, D.C., 508 pp.
STATISTICA. 2003. Version 6.1. StatSoft Inc. Tulsa, Oklahoma.
Weaver, J. E. and T. J. Fitzpatrick. 1934. The prairie. Ecol. Monog., 4:109-225.
AVOID citing web pages. However if a journal article is downloaded from a web source, then the journal URL should follow the citataion:
Wassenaar, L. I., and Hobson, K. A. 1998. Natal origins of migratory monarch butterflies at wintering colonies in Mexico: new isotopic evidence. Proceedings of the National Academy of Sciences, 95(26), 15436-15439. Retrieved on July 5, 2007 from: http://www.pnas.org/content/95/26/15436.long
Submission of Thesis
Revise and proofread your thesis before turning it in to your thesis committee members. All theses must be submitted to your committee members for review at least 1 month before the Graduate School Submission Deadline as published on the graduate school website. Students will not be allowed to schedule an oral defense until AFTER their committee has viewed their thesis.
The Graduate School will only accept theses that have been approved by the student’s committee and are in final form. One copy of the thesis must be submitted to the Graduate School by the deadline. The Graduates School has specific guidelines for thesis submission and these guidelines should be followed exactly as stated on the Graduate School website.
Selecting (an) Investment Bank(s) for the Initial Public Offering of a Venture Capital-backed Company
An EVCA Venture Capital Committee Paper
October 2006
Written by François Lainée and Eric Plantier
About EVCA
The European Private Equity and Venture Capital Association (EVCA) was established in 1983 and is based in Brussels. EVCA represents the European private equity sector and promotes the asset class both within Europe and throughout the world.
With over 925 members in Europe, EVCA’s role includes representing the interests of the industry to regulators and standard setters; developing professional standards; providing industry research; professional development and forums, facilitating interaction between its members and key industry participants including institutional investors, entrepreneurs, policymakers and academics.
EVCA’s activities cover the whole range of private equity: venture capital (from seed and start-up to development capital), buyouts and buyins.
Table of Contents 3
Executive Summary 5
Introduction 7
Section 1. Background on Preparing an IPO 8
1.1. The IPO process: a long journey that does not stop the first trading day 8
1.2. Segmenting the venture capital-backed company’s IPO landscape 9
1.2.1. Company financial profile at IPO drives the scope of achievable objectives 9
1.2.2. “IPO size also matters” 10
1.2.3. Impact of IPO segmentation on the choice of banking partner(s) 10
1.3. The Investment Bank: a central and multiple-headed actor in the IPO journey 12
Section 2. Sanity Checks before Launching an IPO Process 13
2.1. Aligning the company and its stakeholders among themselves 13
2.2. Understanding stakeholders’ new roles and ensuring that management is fit
for his/her new environment 13
Section 3. Going Beyond the Surface in Selecting (a) Banking Partner(s) for the IPO 15
3.1. Building a bank syndicate: a valuable objective paired with
economics-driven constraints 15
3.2. Running an efficient beauty contest 16
3.3. Assessing investment banks through a structured list of questions 16
3.3.1. Overall fit of the bank with the company’s case 16
3.3.2. Probing credibility on driving a solid process in syndicate 17
3.3.3. Assessing the investment bank’s team: measuring individuals’
and team skills, their ability to understand and communicate
the company’s story, the dedication of, and the fit with management 17
3.3.4. Research strength: a requirement increasing with IPO’s size 18
3.3.5. Assessing the ability to size, value and price the offering 18
3.3.6. Assessing the ability to market and distribute efficiently 19
3.3.7. Assessing the ability to manage lock-up and provide after market support 19 3.3.8. Fees structure: probing for optimisation and staying focused on a good deal 20 3.4. Synthesis: a proposed RFP for beauty contest organisation and investment
bank(s) selection 20
Acknowledgement 21
Annexes 22
Annex 1. Main contributors 22
Annex 2. Suggested RFP for investment bank(s) selection for an IPO 23
Annex 3. Suggested IPO track record map 26
4
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Executive Summary 5
Considering the importance of Initial Public Offerings (IPOs) as an exit route for portfolio companies, the favourable development trend of the European IPO market, and the relative youth/immaturity of European venture capitalists (VCs) as regards to IPOs, the European Venture Capital Association (EVCA) has commissioned this short study to provide the European venture capital community with a practical approach to the complex subject of selecting investment banks for their portfolio companies’ IPOs.
Conducted during the summer of 2006 and based on interviews, this study has gathered the advice and know-how available both within the EVCA Venture Capital Committee and within the base of experienced European investment bankers, venture capitalists and IPO-savvy lawyers.
Before considering the selection criteria, a few key points regarding the nature of an IPO and its processes should be outlined:
First, a technology company IPO is mostly a financing event and, in some cases, a partial liquidity event with full liquidity usually reached over a long period for incumbent venture capitalists. As the process to IPO usually spans six months on regulated markets (and three months on non-regulated markets), full liquidity is generally achieved within two years. The IPO process for an AIM listing normally takes three months from when the strategic decision is taken to proceed (including advisor selection).
Second, although all IPO processes are broadly similar, two key dimensions, namely the credibility of the company future growth story and the size of the offering drive the level of risk attached to the IPO, the liquidity horizon for venture capital and the preferred type of banking partners (e.g. large international banks versus sector and geography-specialised boutiques).
Finally, investment banks are not homogenous in their service offering and positioning and all have several functions/ departments involved during the IPO process. This diversity coupled with multi-headed internal structures contribute to the relative complexity of assessing potential partners in the selection process.
In addition, as companies and their stakeholders engage on the IPO route, they need to follow two golden rules in order to increase the chances of success:
* the first rule is that all stakeholders and the company need to have aligned objectives and expectations of the IPO, translating into a shared-fundraising story and vision on liquidity;
* the second rule is that the stakeholders should be aware of public market constraints and their new roles in the public company. The CEO needs to drive the IPO process and be fit for running the company when it becomes public. Venture capitalists need to embrace public market investors’ attitudes.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
6 Executive Summary
When all these risks are under control, selecting the investment banker(s) is the central first step of the IPO process. Overall, the level of comfort and trust of the stakeholders, and especially the CEO, with the team proposed by the bank is a key selection criterion. However, even if preferences emerge, it is best practice to conduct an in-depth structured assessment of a few banks (typically three to four) in order to choose the lead partner in full awareness of its strengths and weaknesses, and eventually build the best syndication. Syndicating banks to support the IPO appears to be a preferable approach (giving more access to investors and more research coverage), provided economic constraints (fee sharing) can be managed.
Given the complexity of the IPO process, the assessment of potential investment banking partners should cover a number of dimensions, namely the credibility of the bank/team to drive a robust and timely process, the team’s and individuals’ skills including the team’s understanding of the company’s story and ability to communicate it, the dedication of the team and its fit with management, the quality and long term reliability of the research arm, the ability to size, value and price the offering, the ability to market and distribute efficiently and, very importantly given the duration to access full liquidity, the ability to manage lock-up and provide after-market support. Finally, fees levels and structure matter, keeping in mind that the lowest price might not achieve the optimum outcome.
The lead bank and syndicate (if possible) being chosen, the company is now ready to embark on the IPO process which, if successful, will bring the company into a new stage of its development: becoming a public company.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Introduction 7
An Initial Public Offering (IPO) in which the public market pays a premium over strategic and financial acquirers is considered the preferred exit strategy for venture capitalists (VCs). However, IPOs are far from being the majority of venture capital exits – during the boom of 1998-2000, IPOs were as high as 40% of all venture capital exits and during the dry spell of 2001-2003 they were as low as 7% (averaging out between 25-30% over long term).
Since late 2004, IPO activity has increased, especially in Europe which accounted for more than 50% of all worldwide venture capital-backed IPOs in 2005. In the United States, the introduction of the Sarbanes-Oxley law has negatively impacted the technology IPO market. Due to the restrictions of this law, it is now more challenging to introduce a small capitalisation company into the US market.
However, even if Europe takes the greatest share of IPOs with fast developing markets such as AIM (500 IPOs between September 2005 and 2006) and Alternext (50 IPOs in its first year of existence) for small capitalisation companies, it is far from being ideal with problems such as high market fragmentation, lack of public investors, lack of liquidity, lack of research, too few successful secondary offerings and a less experienced technology financial community (as compared to the US market). In particular, venture capital is a relatively new market in Europe and many venture capital professionals and venture-backed company managements have little or no IPO experience.
For these reasons, the European Private Equity and Venture Capital Association (EVCA) has commissioned a study with the aim of helping this community choose their banking partner(s) for an IPO. This study was conducted over the summer 2006 with contributions by the EVCA Venture Capital Committee, European technology investment bankers (within large and medium-sized investment banks), leading European venture capital professionals and successful entrepreneurs.
The synthesis of those contributions is summarised in this report which has been divided into three sections:
* background on preparing an IPO;
* sanity checks before launching an IPO process;
* selecting (a) banking partner(s) for the IPO.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
8 Section 1. Background on Preparing an IPO
1.1. The IPO process: a long journey that does not stop on the first trading day
The IPO process from the day when the company board of directors makes the formal decision to go public to the first day of trading generally lasts about six months for venture capital-backed companies on regulated markets and as little as three months on non-regulated markets. It is composed of the following key steps:
* selection of the bank: the focus of this study;
* preparation of the prospectus and the investor presentation, classification and completion of financial and legal documents (typically a two to three month process);
* investor road shows: for typical IPOs, the company management presents in general to 20-40 institutional investors (a number correlated to the amount of capital raised). The length of this stage is around one month depending on the number and location of the investors approached;
* approval of the market authority (in the case of regulated markets): this consists of a first meeting with the market authority followed by an exchange of questions and answers. Depending on the size of the offering and financial markets, this step lasts between one and two months. In case of IPOs on non-regulated stock markets, the approval process is delegated by the stock market to the investment bank. Please note that this regulatory authority approval process is not applicable to most IPOs on AIM where the investment bank itself (nominated advisor or nomad) is responsible for vetting the documentation under what is effectively delegated authority;
* public and analyst presentation: a stressful day for the management who presents the company for the first time to this new audience and whose stage performance can create enthusiasm or scepticism a few weeks prior to market introduction;
* first trading day.
If the IPO process including going to the full liquidity of venture capitalists is now considered, then it is usually a nine to 30-month process including six months to two years post trading day to establish full liquidity (the longest timings being generally associated with “small” IPOs, as measured by amount of capital raised). This post trading day phase covers the following specific action items:
* investor relations and communications on company financials and important news;
* investor road shows associated with either important business news or preparation of secondary offerings (sales of historical shareholders’ shares, and/or new capital increases);
* market making and support: the intervention of market operators to facilitate liquidity of the shares;
* lock-up and orderly selling management: the objective is to enable the sales of incumbent investor shares without disturbing the share price evolution. This phase is critical and has to be handled with care. It has to be brought to the reader’s attention that some markets (by regulation), market authorities or lead investment banks require a lock-up period for incumbent investors (generally 12 months).
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
1.2. Segmenting the venture capital-backed company’s IPO landscape
Although all IPOs follow a similar process, they can be different due to two main factors which strongly impact the
success of the offering and the type of banking partner required: the financial profile and the size of the offering.
1.2.1. Company financial profile at IPO drives the scope of achievable objectives
The financial profile of the company determines which investors will be attracted to invest in the company and what their expectations will be of the company and its existing shareholders attitudes. It will also determine which type of banks will pitch for the IPO.
The most important driver of the investor appetite in the IPO is the credibility of the future growth story (in revenue and results) of the company. The size of the company (in terms of revenue) and past growth track record are important factors in reassuring investors in the growth story. Companies with highly credible future growth plans are the “core profile” for an IPO.
However, some recent IPOs have also been completed with companies deviating significantly from such core profiles – such as several AIM floated companies that may have potential growth stories, but have hardly any growth track record, and still have a long path to profitability.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
10 Section 1. Background on Preparing an IPO
In fact, these IPOs are of a different nature to the core IPOs described above; they are refinancing events to support the next step of a company’s development, but are not likely to offer short term liquidity to the historical shareholders. The reason for this is that new investors, perceiving a high level of risk (due to the unproven growth track record), want existing shareholders to show their commitment to the development plan by staying on board, and even sometimes by contributing to the new financing event. As regards to liquidity, these “refinancing IPOs” stocks generally show low level of transaction volumes until these companies prove strong performances.
In contrast, in core IPOs, institutional investors might not care, or even might encourage rapid venture capitalists’ exits because they perceive venture capitalists as being “strange” managers of public company stocks, with sometimes irrational behaviour (in selling their stocks).
Core IPOs are thus considered less risky (from an investor close up) than the second category described above. Therefore in difficult market conditions, core IPO companies might be successful in their IPO, but with a reduced price whereas the latter category companies might have to postpone their offering.
This segmentation also influences which banking partner to choose, because the institutional investors tend to be different for the two groups of companies mentioned above.
1.2.2. “IPO size also matters”
Another very important segmentation of IPOs, and very relevant for our study, is the size of the offering and the associated valuation. For European technology companies, we estimate that the division between small and large IPOs is $80 million initial offering or $500 million post-money valuation. The companies belonging to the smaller cap IPO group form the large majority (in 2005, they represented 85-90% in number but only approximately 35% in amount raised).
Small IPO companies are quoted on small to medium-sized public markets (AIM, Euronext C, Alternext), and are generally advised by small to mid-sized banks with a strong, local foothold. By contrast, companies belonging to the large IPO category quoted on main markets (LSE, Euronext B), are generally advised by large banks with strong brands and an international foothold.
Size is also an important driver of the IPO risk. Indeed, the likelihood of success is strongly correlated with the size of the offering – large offerings having a higher chance of succeeding than small ones.
1.2.3. Impact of IPO segmentation on the choice of banking partner(s)
Although IPO segmentation impacts the chances of IPO success and the expectations for shareholders, the IPO process, broadly speaking, is similar for the different types of IPOs mentioned above, and the main criteria in choosing the banking partner are also similar.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
11
However, there are some general rules which apply both to large international and small medium local banks:
* some markets authorities (i.e. in cases of non-regulated markets) select a list of investment banks or advisers (they are called nominated advisors for the AIM market). It is thus mandatory for the company to choose from such a list, if relevant;
* the brand name of large international banks can be a strong added value to attract large institutional investors, which may be required in large IPOs. Such brand names will also have a strong international presence and the placement capability that is required for companies that need to have strong international presence;
* local small and medium-sized banks are generally well-connected with local institutional investors which is a key success criterium for small IPOs. These banks may also be important players in large IPOs where, in many cases, they fit into a syndication usually led by large banks.
Beyond these general rules, selection of a banking partner(s) depends on a number of criteria, and the weight to apply to each of the criteria may be different depending on the segment the company belongs to. The reader is invited to read the detailed selection criteria in Section 3.
Finally, Table 2 below summarises the two segmentation axis developed above, specifying the appropriate IPO expectations and likely outcome depending on such segmentation.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
12 Section 1. Background on Preparing an IPO
1.3. The Investment Bank: a central and multiple-headed actor in the IPO journey
In the large majority of IPO cases, the IPO process is led by an investment bank with three internal departments involved in the process. These have different roles (described hereafter) and interfaces with the company’s management and its shareholders (in a few cases – generally in some small IPOs – those roles are assumed by different parties working with one another).
1. The corporate finance head and his team are the company’s central interface. They coordinate the overall process, produce the prospectus, in conjunction with other advisers (lawyers, accountants, etc.) and drive the marketing (road show).
2. The analyst provides input into the market in writing research notes and news reports used by institutional investors in their investment decision (for primary and for post IPO). For the company, it is crucial to get both quality (reputation of the analyst(s)) and quantity (number of analysts) of research coverage to achieve proper market valuation and stock liquidity. It is also important for the management to understand firstly the independence of the analyst from the corporate finance team (the Chinese Wall strictly followed by Anglo-Saxon banks where, for example, the corporate finance person is restricted from speaking directly to the analyst during the IPO process) and secondly, that the analyst works for the buyer (the market) when presenting a company’s business plan to him or her. Such a presentation should be carefully planned both in terms of format and level of details given (any data provided to the analyst will be tracked).
3. The trading floor provides the corporate team with market input for setting pricing and sells the company shares to institutional investors. They act as gatekeepers to the investors (their customers).
Beyond the specific organisation of a bank’s teams to serve clients preparing an IPO, investment banks are very diverse in their reputation and market access, but also on their overall service offering. Indeed, post IPO, the company is likely to need other types of support from the banks, such as M&A advisory, debt and convertible instruments, treasury management, asset-backed financing and many more services which are generally offered by mid to large-sized banks (small investment boutiques are more focused on traditional equity financing and M&A advisory or even just specialised on public placement for small capitalisation companies.
Thus, when considering an investment bank, the management should first evaluate the breadth of the services required in the mid to long-term and secondly, decide whether it is more appropriate to select a relationship-driven one-stop shop approach (that will provide the whole breadth of service required) or best-of-breed partner (focusing on the IPO).
Whilst this paper is not by any means directed towards AIM listings, it should be noted that it is not uncommon for AIM candidate companies to select a different firm to act as investment banker (nomad) to that performing the broking (sales/marketing) function.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Section 2. Sanity Checks before Launching
an IPO Process 13
Before embarking on the long IPO process, some basic rules to increase the IPO’s chances of success and optimise the Return on Investment (ROI) for historical shareholders, including venture capitalists, should be followed:
* check the alignment of investors and management regarding the objectives and expectations of the IPO and ensure that the IPO process does not distract management form their day job of running the business – if necessary consider hiring an interim project management team to support the endeavour;
* be clear on the role of each party post IPO and make sure that the management is up to the job.
2.1. Aligning the company and its stakeholders among themselves
The general experience of venture capitalists, bankers and management is that a misalignment between stakeholders on the objectives of the IPO and a misunderstanding of each other’s expectations on the outcome contributes to a high risk of failure of the IPO.
Firstly, all the parties need to have a common understanding of the strategy of the company and the best development scenario. They need to have done their homework on exit scenarios and be aware of each of the scenario’s pros and cons. For instance, they need to have weighed a trade sale against an IPO. They also must be alert to of the fact that the IPO is only a step in company development and, save for very large IPOs, is in most cases only an intermediary step on the road to full liquidity for venture capitalists. They also need to be aware of the constraints generated by the public market, their role post-IPO, and have a common understanding of the venture capitalists’ path to full liquidity and how to best manage the after market and secondary offerings.
Secondly, they need to be aligned on how their own individual objectives fit in with the company’s IPO objective. This implies an alignment on the offering size, the ratio of old and new shares to be sold, the expectations on valuation and the venture capitalists liquidity beyond IPO. Those expectations also need to be in line with what the market can accept. For instance, new investors’ investment decisions will be highly sensitive to the use of proceeds (providing historic investors liquidity versus cash for company’s development).
This may all seem easy to achieve, but the stronger/deeper this alignment, the safer the IPO route.
2.2. Understanding stakeholders’ new roles and ensuring that management is fit for his/her new environment
The roles of both the management and the current investors (the venture capitalists) need to evolve as the company prepares to become public.
First, after the IPO, the management adds the supervision of institutional investors (as well as incumbent venture capitalists), and reporting to the analysts and the public market to its job of driving strategy and operations. The pressure increases on management with higher visibility, increased level of expectations and accountability demanded by new investors. This new role starts during the IPO process with analyst and investors presentations, often a new experience for management.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Section 2. Sanity Checks before Launching
14 an IPO Process
In addition to this post IPO new job requirements, it appears critical that the CEO be strong enough to be the central driver of the IPO process, including such areas as bank and syndication selection, size of offering, pricing and secondary issues. Indeed, as a large number of actors are involved in these decisions, including venture capitalists, management, banking partners, and advisers, a strong CEO deciding on those areas is best for protecting the company’s interests. However, managing the process and having the management remain focused on managing the company’s business is a difficult task. Thus, as mentioned earlier, it might be a good idea to hire an interim project manager to support the process internally.
Having a management team up to facing these challenges is a pre-requisite for engaging in an IPO process with a reasonable amount of risk. In many technology company IPO cases, venture capitalists and management acknowledge the pre-IPO management lack of experience in managing a public company, and a new CEO and/or CFO often comes on board six to twelve months ahead of the IPO to fill this gap.
Second, after their portfolio company becomes public, the venture capitalists are less active (in company development) and leave part of their pre-IPO duty to the management and new investors. This new role is also difficult for venture capitalists since they have to give away a significant amount of the control they had over the company while it was private and have to accept that the public market will determine the ultimate outcome of their lengthy and complex support of the company’ pre-IPO development. However, accepting and living by those new conditions are absolutely critical for overall IPO success.
Once these three acid tests are passed, the management and investors are ready to select their IPO banking adviser. At this point, the company often wonders how to best select this important partner.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Section 3. Going Beyond the Surface in Selecting
(a) Banking Partner(s) for the IPO 15
The investment bank is a central and key contributor to the IPO process and its success. Thus, it is extremely important to select a skilled, loyal and high-performing partner. The selection is generally done through a two-step process. First, the company board or management pre-selects a few candidates (three to five) using their own industry contacts and referrals. Second, a structured beauty contest takes place to choose the best candidate, in the circumstances.
Because of the complexity of the IPO process and the natural objective of the bank candidates to present themselves to their best advantage, it is not easy to conduct a robust and professional beauty contest.
This section aims to provide advice on organising efficient beauty contests with a structured list of questions to be asked to the candidates to ensure that:
* no critical aspect of the IPO process is left uncovered; and
* the comparison between candidates is streamlined and rigorous.
3.1. Building a bank syndicate: a valuable objective paired with
economics-driven constraints
Overall, most IPOs do not have syndicates. However, a strong syndicate can offer many advantages to increase chances of success of the IPO. Using a syndicate, indeed, will bring the company many advantages: more placement power and thus a greater possibility of success and potentially better pricing, more research at the IPO and post-IPO period, and potentially (but not obviously) emulation between the investments banks party to the syndicate.
However, building a syndication comes with complications, due to economic constraint, as investment banks require minimum fees per deal and a small size offering might not produce enough fees to motivate more than one player. We estimate that above $1m fees or $20m offering, the fees should not be a barrier to build a syndicate. This leaves in principle room for syndication in a significant portion of IPOs of venture capital-backed companies.
Syndication also has some risks. The first one is that, even if the offering is theoretically large enough to allow enough fees for few banks, the company might end up being bottom of the list of the bank’s priorities. The second risk is to have a non-functioning syndicate when the banks do not work well together (either for culture compatibility reasons, or as a result of a priori unwillingness to syndicate).
To conclude, given the balance of pros and cons listed above, we would encourage companies and stakeholders, when economic constraints permit, not to give up on building a syndicate without having tried really hard.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Section 3. Going Beyond the Surface in Selecting
16 (a) Banking Partner(s) for the IPO
3.2. Running an efficient beauty contest
In preparing a beauty contest, one should focus on the number of pre-selected participants and the depth of information asked.
Three to four banks is usually a good number to invite. If a company invites more, the process could significantly burden the management’s workload and might discourage bankers from putting together quality work for the beauty parade.
It is essential to ask your candidate concrete, pertinent questions. Questions relative to its track record should call for measurable and provable answers. A detailed list of those questions is included in the Annex 2. of this report; it might be useful to extract the questions that are the most relevant to your case to simplify the process. Because the professional preparation of beauty contests is heavy and costly for the bank, too many and not directly useful questions might discourage the banks from doing a high-quality job in answering them.
Using a well-connected and informed adviser might be very useful in getting the best output from the beauty contest. Such a person might be an ex-investment banker who is exceptionally knowledgeable about the different teams and the fee negotiations.
Finally, it is especially important to conduct reference calls to validate the performance of the candidate along the components described below.
3.3. Assessing investment banks through a structured list of questions
3.3.1. Overall fit of the bank with the company’s case
In this area, the objective is to focus on two key issues: the likely fit of the bank structure with the company, and the level of commitment of the bank to deliver throughout the process.
First, one should understand the banking capability and track record in relation to your case. In particular, its typical size of offerings, sector expertise, local and international coverage and other banking services offered.
In the case of small and mid-capitalisation offerings, it is crucial that the candidate bank has a dedicated team (for those small- and mid-size markets) both on the investment and the trading side.
Second, the motivation of the bank for a particular case should be examined: why are they interested in working with the company, and how high will the company be on their priority list? Their work load and their involvement with competing IPOs and, more generally, their relationship with competitors (which impacts their ability to work in syndicate) should also be looked into.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
17
3.3.2. Probing credibility on driving a solid process in syndicate
The company and its shareholders want to be comfortable with the timing of the offering committed by the bank and the bank’s ability to drive a timely process. They also want to explore how the bank might work in a syndicate if needed.
With regards to the IPO’s timing, the bank should offer its view on market liquidity, the market’s likely appetite
for the company’s IPO, and the best timing in respect to other – possibly competing – IPO offerings.
The bank should lay out a detailed process and timetable with specific deliverables, and the resources required to support it.
It is useful to ask for a list of the recent IPOs led by the bank and withdrawn (potentially on a no-names basis),
the reasons why these IPOs were dropped and the level of the bank’s involvement in the withdrawal.
Finally, the candidate should also present its vision on the possible syndication, its role in such syndication and its preferred parties. This should provide the company with more information to decide whether syndication is best for the stakeholders and how to make it work efficiently.
3.3.3. Assessing the investment bank’s team: measuring individuals’ and team skills, their ability to understand and communicate the company’s story, the dedication of, and the fit with management
In this section, the company and shareholders want to ensure the competency of the proposed team, their availability and commitment, and the likely chemistry with the management over the duration of the process.
Probing the level of skills is first achieved by mapping a few key indicators of past IPOs involving this team (e.g. number, sizes, valuations, stock price evolution, etc.). This will help in understanding the team’s track record, their experience working together (both corporate finance and trading side), their capacity to deliver results and the dedication and involvement of senior management.
The level of skills can also be analysed by examining the team’s personal track records, former prospectus and presentations, and references. During this process, meetings should be face-to-face not only on the corporate finance side but also, even more critically, among lead team members on the trading/sales side.
The second area to probe, through direct questions, is the bank team’s understanding of the company and its market, and the team’s ability to formulate and sell the company’s story to institutional investors and to the public market.
Lastly, the chemistry between the management and the investment bank’s team needs to be robust since they will work very closely for a long period through the IPO process and beyond (in the case of AIM-listed companies the bank (nomad) is the company’s regulator as well as its adviser). This is best assessed through the experience of the meetings conducted prior to and during the selection process. Quite often, corporate finance team leader (of the candidate) and management have regular contact up to twelve months before the beauty contest, which clearly helps evaluate the chemistry and capabilities of each other and facilitate the selection process.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Section 3. Going Beyond the Surface in Selecting
18 (a) Banking Partner(s) for the IPO
3.3.4. Research strength: a requirement increasing with IPO’s size
During the assessment process, focus should be put on understanding the quality and reputation of the analyst by reading his/her coverage and analysis of the company’s sector and comparables, meeting him/her and making reference calls to institutional investors (please note: as a general feedback from our contributors, ratings are of lesser importance).
Beyond reviewing previous notes written on comparable companies, the face-to-face meeting with the analyst to sense his or her understanding and enthusiasm of the company and its market is of crucial importance. Indeed, due to the Chinese wall between the corporate finance and the research side, the analyst of the lead bank can formulate any recommendation – including a negative one.
When meeting research analysts it is important to assess the time and energy the research analyst is prepared to commit to providing research for a company. If the company has a large free float, there will be obvious commercial reasons for tracking the company. If the company has a small free float, it may be difficult for the analyst to justify his or her time in covering a stock. However, an interesting company that fits into the analyst’s overall coverage strategy may get the appropriate time and focus. Timely and insightful commentary drives stock liquidity.
Finally, it is also a good practice to analyse the history of research coverage (by the relevant analysts of other banks) of IPOed companies recently led by the candidate. This gives a sense, beyond the quality of the first research note, of the research support one can expect over time, which can be a valuable support for liquidity and secondary events.
3.3.5. Assessing the ability to size, value and price the offering
The ratio of old and new shares sold and the pricing are usually highly discussed topics during the process as parties may have different objectives: the bank wants to have a successful IPO but also please its client while the venture capitalist might want to increase its liquidity on the primary, and the management wants a successful IPO with a steadily increasing share price.
However, whatever the objectives of the parties are, the fair price is set by the market and a competent bank should be (and generally is) able to advise on the optimum price the market will accept, keeping in mind financial market dynamics and the specifics of your case.
In this section, the company and shareholders will examine the ability of the bank to evaluate the introduction price range and access the bank candidate’s track record of participating in IPOs with positive price evolution post introduction.
The candidate should detail its view of the price range and its rationale. This should enable a judgment on the bank team’s ability in understanding the company’s positioning, in using the appropriate comparables, and in justifying the premium or discount over comparables it suggests.
The bank should also provide price data regarding its last relevant IPOs, in particular the change of introduction price range during the IPO preparation process (from the pitch day to the IPO day), and the share price evolution (relative to market) in the next 12 months post IPO.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
19
3.3.6. Assessing the ability to market and distribute efficiently
In this section, company and shareholders evaluate the investment bank’s understanding of investors’ target base and its level of access to this base.
First of all, these investors have to be classified (e.g. by the size of their investment, the market sector and where they are active, as well as their behaviour – whether they are long or hedging investors and thus participate in the stability or liquidity of your stock), and the bank should provide a clear picture of the relevant segments for the IPO, and emphasise its credible access to these segments.
The evaluation of the bank’s adequate investor accessed base can be done by:
* analysing the bank’s sales force organisation structure; and
* examining the relevance of the investors (to the company’s IPO) that participated in precedent IPOs led by the candidate.
In addition, it can be useful to qualify the soft skills of the sales force, i.e. the level of trust it has developed over time with its customers using independent ranking surveys on the bank sales force (especially regarding the level of perceived professionalism and trust) and possibly using reference calls of institutional investors.
3.3.7. Assessing the ability to manage lock-up and provide after market support
Stock stability and liquidity post-IPO should be key in the bank’s choice and one should rely on the banking partner to provide significant support post-IPO.
Thus, the candidates’ view on post-IPO management – their services, experience, and commitment during that period – needs to be evaluated. As for many of the topics above, this can be achieved by probing the bank’s proposals for the planned IPO, and analysing the track record. Please note that legacy investors in companies on AIM with a limited trading record are required under the AIM admission rules to commit to a 12 month lock-up and most banks (nomads) would also insist upon such a lock-up even for those with more established trading records.
The candidate proposal should include:
* the timing of lock up;
* the necessity or otherwise for liquidity contract and orderly management sales contracts and the main characteristics of such arrangements;
* the timing and its involvement on secondary offerings; and
* its planned activity in market trading.
The candidate should provide data examples of lock up management, liquidity contracts, organisation of secondary offerings of similar IPOs, and its involvement in after-market trading (i.e. the percentage of stock traded by the bank post IPO).
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Section 3. Going Beyond the Surface in Selecting
20 (a) Banking Partner(s) for the IPO
3.3.8. Fees structure: probing for optimisation and staying focused on a good deal
Lastly, the fees and their structure proposed by the bank should be evaluated, keeping in mind that the banking partners need to be motivated to deliver. The major portion of the total fees is a success fee based on a percentage of the amount raised. As an indicator, the range of value of this percentage is from 2% to 6%+, with medians around 4,5% (main fees around 5%) across regulated and non regulated markets, for IPOs with significant amount raised (between €50m and €100m). For smaller amounts raised, one should likely expect an extra 1% to 2% on this fees level.
Benchmarking these fees against market databases could be useful. As an example, the subscription website www.dealmonitor.co.uk provides useful information on all fees and commissions (and warrant terms) charged by banks and brokers on AIM IPOs and secondary issues.
3.4. Synthesis: a proposed RFP for beauty contest organisation and investment bank(s) selection
As presented in Section 3.3., the efficient approach for conducting a thorough assessment of the candidate is performed by analysing the candidate’s answers to focused questions and cross examining its track record.
Thus, this report ends with two structured documents which can be utilised to drive the selection process:
* a questionnaire including the key questions, by topic, to send to the banks for the preparation of the beauty contest meetings; and
* an “IPO track record map”, including the key indicators of past IPOs led by candidates, and “hard” measures on the role and performance of the candidates in these operations.
These documents (in Annexes 2 and 3) are intended to provide a structured and comprehensive grid of the issues covered in this report on the topic of selecting banking partners for a sizeable venture capital-backed company IPO. Companies and venture capitalists can adapt them to their individual needs and approach.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Acknowledgement 21
On behalf of EVCA, we would like to thank the main contributors to this study (mentioned in Annex 1.) who have generously provided their time and knowledge for this document. We hope it will be useful to the venture capital community at large and will help the venture capital community make Europe an even better place to invest.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
22 Annexes
Annex 1. Main contributors
Interviewees
Renee Aguiar Lucander 3i
Richard Anton Amadeus Capital
Anne Valery Bach Banexi Venture
Markus Boser JP Morgan
Charles Cameron Jefferies
Jacques Chatain Auriga Partners
Simon Cornwell Amadeus Capital
Eric Favier Casenove
Eric Forest Oddo Corporate Finance
Nick Garrett JP Morgan Cazenove
Rupert Hume-Kendall Merrill Lynch
Henry Moulard Truffle Venture
Edward Snow Amadeus Capital
Jean Tardy Joubert Merill Lynch
Benedict Tompkins independent
About the authors
Francois Lainée is an IT partner with Auriga Partners, an early-stage French venture capital firm managing about €330m. Prior to joining Auriga Partners, in 2002, he was a manager of development and product lines with Thomson Broadcast, and a high-tech consultant with McKinsey in France and the US. He is a member of the EVCA Venture Capital Committee. Contact: lainee@aurigapartners.com, Web: www.aurigapartners.com
Eric Plantier is the founder and president of Blueprint Consulting, a financial advisory firm focused on private placement and M&A for venture-backed IT companies. Prior to Blueprint Consulting, Eric was Investment Director at Viventures (€500m under management), Head of Strategy for Ericsson Mobile Phone and Sales Director at Schneider Electric. Contact: eric.plantier@blueprintconsulting.fr, Web: www.blueprintconsulting.fr
Acknowledgements from the authors to
Lawrence M Levy, Charles Crosthwaite and James Bedar from the London office of international law firm Brown Rudnick LLP for their dedication in giving legal advice before the publication of this document.
“All information for this study was gathered through interviews and panel discussions. In this respect, it concisely and professionally reflects the findings and interpretations of the authors and EVCA representatives but the document cannot be considered as comprehensive in such a way that in the occurring case further professional and specific advise should not be acquired or could be replaced by it.”
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
23
Annex 2. Suggested RFP for investment bank(s) selection for an IPO
Questions with an * at the end can be covered by specific portions of the “IPO track record map”. They might therefore be skipped from interview guides sent to the bank if IPO maps are provided instead.
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
24 Annexes
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
Selecting (an) Investment Bank(s) for the IPO of a VC-backed Company - An EVCA Venture Capital Committee Paper - October 2006
26
Annexes
Name of person involved
Current presence in the bank (Y/N)
Philip & Jeune Kirmser International Student Scholarship Guidelines To apply for this scholarship, please read the following information thoroughly.
By completing this scholarship application form, you will be considered for the following scholarship:
Philip and Jeune Kirmser International Student Scholarship: $750.
The recipient of this scholarship will be a non-resident undergraduate student properly enrolled in any curriculum at Kansas State University with financial need. First preference will be a student of African origin and be able to provide proof of citizenship from any country on the continent of Africa.
Deadline to apply: Monday, March 6, 2017 by noon.
No incomplete or late applications will be accepted. Recommendation letter must also be submitted by the deadline. NO EXCEPTIONS!
Eligibility Requirements for the Philip & Jeune Kirmser International Student Scholarship:
* You must be a full-time undergraduate student (12 credit hours) at Kansas State University of African origin and must provide proof of citizenship from any country on the continent of Africa (passport copy).
* You must not have received this scholarship during the previous academic year.
* You must currently be in the U.S. on a nonimmigrant visa.
* You must have been enrolled full time at K-State during the Fall 2016 semester or before.
* You must not currently be receiving any scholarship assistance.
* You must be continuing as a full time student at K-State during the Fall 2017 semester, since the scholarship money will be awarded at that time to cover tuition fees. (Students graduating prior to Fall 2017, are not qualified to receive this scholarship).
Step 1: Complete the online application form. You can save the form and return to it later in OrgSync by using your login information with your K-State eID and password if you login to OrgSync before beginning the scholarship application form. If you do not login, then you cannot save the form to return to it later.
Click here to go to the online scholarship application form:
https://orgsync.com/106581/forms/243466
Step 2: Submit only one Recommendation letter—you will need to contact a K-State professor, advisor, etc. to ask if he/she will write a recommendation letter for you for the international student scholarship. Here is the information that they should include in their recommendation letter:
* For how long and in what capacity have you known the applicant?
* Please comment specifically on the applicant in terms of why the applicant should receive a scholarship. (Please include details about academic abilities, involvement in campus and community organizations and any other information that would be helpful for the scholarship committee to know about this applicant).
You will need to deliver the one recommendation letter in a SEALED ENVELOPE to International Student and Scholar Services by the deadline of Monday, March 6, 2017 by noon or the person completing your recommendation letter can mail the letter to:
Ruth Newton
104 International Student Center
1414 Mid-Campus Drive North
Manhattan, KS 66506
*Recommendation letters received after noon on March 6, 2017, will not be considered.
Step 3: Step 3: You will receive an automatically generated email from orgsync after your application form has been successfully submitted online. Once the recommendation letter is physically received, please allow 2 business days to receive another email from orgsync indicating that your application is complete.
* Your application will be reviewed by the scholarship committee after March 6, 2017. All students will be notified by the beginning of April if they receive a scholarship.
Employee Instruction Manual
How to access the Performance Management System
Employees, Supervisors and Reviewing Officers will work on and view the Performance requirements and evaluation through the Cardinal Talent Employee Portal.
To enter the Employee Portal click on the Cardinal Talent Employee Portal link on the far right.
Note: For those employees with multiple users, make sure the user type is changed to “employee” to complete your
self-evaluation.
The Cardinal Talent Employee Portal is set to a green background. You can exit the Employee Portal by clicking on
Employee completes self-evaluation
* In the Home page the employee will see all actions that require his/her attention.
* To open the Self-evaluation click on the far right.
* In the Self-Evaluation you will be able to complete the following actions:
o View your job description and add any comments on your performance as it relates to your Job Duties.
o Opportunity to comment on your goals/activities from last year.
o Summarize tasks and responsibilities you performed this past year, as well as, any goals you would like to establish for next year.
* Click when sections are done.
* A green bar should appear at the top of the page as confirmation.
* The system will automatically send the self-appraisal form to your Supervisor for review.
Employee Acknowledges Evaluation
* Once the Supervisor and Reviewing Officer finalizes the Performance Evaluation form the employee and direct supervisor will need to schedule the Evaluation Review Meeting.
* After review meeting the employee will have an opportunity to comment on his/her evaluation and Acknowledge that he/she received the evaluation.
Employee Acknowledges the Goals/Activities for Next year
* Supervisor will create Goals/Activities for the upcoming year.
* Employee will need to acknowledge the goals/activities for the year.
EAST TEXAS BAPTIST UNIVERSITY
(Revised November 11, 2011)
GUIDELINES FOR HANDLING FORMAL SEXUAL HARASSMENT COMPLAINTS
DESCRIPTION OF HARASSMENT – East Texas Baptist University prohibits harassment and intimidation on the basis of one’s sex, race, color, national origin, religion, age and disability. Examples of conduct prohibited by these policies include but are not limited to repeated insults, humor, jokes and/or anecdotes that belittle or demean an individual or group because of sex, race, color, national origin, religion, age and disability, and physical conduct or verbal innuendo which, because of one of the above characteristics creates an intimidating hostile or offensive environment.
Examples of conduct prohibited by the policy against sexual harassment include, but are not limited to:
a. Persistent unwelcome flirtation advances and/or propositions of a sexual nature;
b. Repeated unwelcome comments of a sexual nature about an individual’s body or clothing;
c. Unwarranted displays of sexually suggestive objects or pictures;
d. Unnecessary touching, such as patting, pinching, hugging, or repeated brushing against an individual’s body;
e. Suggestions that submission to or rejection of sexual advances will affect decisions regarding such matters as an individual’s employment, work assignments or status, salary, academic standing, grades, receipt of financial aid, or letters of recommendation; and
f. Sexual assault.
YOUR RESPONSIBILITIES - As one of the designated persons to whom complaints regarding harassment are to be brought, the University places the following responsibilities upon you:
1. To inform the complainant of the University’s harassment complaint procedures so that the individual knows what he/she needs to do to initiate University action if that is desired.
2. To provide a skilled and sympathetic ear to a person with a harassment problem.
3. To receive information from the complainant and separate facts and observations from assumptions and conclusions and otherwise focus on relevant information.
1
4. To assess the situation to determine the next step. A part of the assessment should be whether interim steps are necessary to prevent further harm before the investigation can be completed. For example, it may be desirable to eliminate contact between the complainant and the accused. Any burden on the complainant should be minimized.
5. To provide accurate information on confidentiality of complaints and an explanation of the situations in which it might be impossible to maintain confidentiality, i.e. when other University personnel have a need-to-know. Also, it is difficult to confront the accused without specific information and the accused usually knows who he or she has harassed or approached inappropriately.
THE INTERVIEWS - When you are interviewing the person who has alleged that they have been harassed, it is important to do the following:
1. Set aside a sufficient amount of time for the initial interview.
2. Find out what happened. Get specifics.
3. Take notes of the meeting. Notes of interviews and meetings should be prepared contemporaneously and dated. If at all possible, two persons (preferably a male and female) should interview both parties to the complaint and any witnesses, coworkers or persons having knowledge of the situation. In the event the information obtained may be disputed later, the second set of notes becomes invaluable.
4. Find out what effect the alleged harassment has on the complaining party. Was the conduct perceived as a joke, was it really unwelcome, did it embarrass, frighten or humiliate?
5. Ask the complainant to identify any witnesses which support her/his allegations. Ask the person filing the complaint not to discuss the investigation with others.
6. Determine what the complainant wants to result from her/his complaint. However, avoid making premature commitments on behalf of the University to the person making the complaint.
7. Outline the timetable of the University’s investigation (suggested maximum of 30 days for completion, 60 days before a determination) and assure the person complaining of the importance of the issue to the institution. Assure the person filing the complaint that all matters will remain confidential to the degree practical and that the information will be shared only on a need-to-know basis. Contact between the complainant and the accused will be eliminated, or continue to be minimized, or avoided during the investigation period.
8. Make it clear that the University takes its responsibility to investigate the matter seriously.
2
9. Take care that you conduct yourself so that you cannot be accused of any bias. Make no comments about the character, job performance or the family life of either the complainant or the accused.
10. Ask the person to put his or her complaint in writing, sign it and submit it to you; or, if the investigator takes the person’s statement, ask the complainant to sign and date the statement. The statement should contain as a minimum the specifics of the alleged behavior or action, the date(s) or approximate date and time, name of witnesses or people complainant has told and desired resolution.
11. Assure the complainant and any witnesses that the university in the strongest terms prohibits retaliation against a person who brings a complaint for harassment or provides information about such a complaint. If the person believes that retaliation in any form is occurring, the matter should be reported to you immediately so that it can be dealt with.
When interviewing the person accused of sexual harassment, consider the following:
1. Inform the accused of the details of the allegations against him/her.*
2. Obtain a statement from the accused which recounts his/her side of the story. Reduce it to writing and have him/her sign it.
3. Ask the accused to identify any witnesses that support his/her allegations.
4. Take notes of the meeting. Notes of interviews and meetings should be prepared contemporaneously and dated. If at all possible, the same persons who interviewed the complainant should interview the accused.
5. Clarify the professional relationship between him/her and the accused, i.e. does he/she teach her/him or supervise her/him in anyway; have they worked together in a professional capacity?
6. How long have the parties known each other? Was there any prior social relationship between the two? Is there a history of individual or group socializing?
7. You may expect the accused to deny the charges. Observe his/her reaction. Note whether or not there is surprise, anger or disbelief. Observe not only what he/she says but his/her body language as well. Can you make any observations about his/her credibility?
8. Assuming he/she denies the allegations, probe further to determine with the accused the background, reasons and motivation that could possibly trigger the allegations.
9. Inform the accused that the University has a policy against sexual harassment and will vigorously enforce it. Remind him/her what sexual harassment is.
3
10. Inform the accused of the process from this point on. Namely, you will complete your investigation and make recommendations to the President who will make a decision as to what if any action to take.
11. Do not be drawn into a discussion of any collateral issues. You are only there to discuss the allegations against the accused.
12. Make sure the accused understands that he/she is to take no retaliatory action against the persons making the complaint. It is best for him/her to keep his/her distance. Any contact which could be perceived as retaliation would automatically constitute sexual harassment.
13. Sometimes the accused will want to or offer to apologize; recommend this not be done until your investigator talks with complainant because it can be perceived other than intended. An apology can sometimes be a letter through the investigator or meeting with a third party present. In cases involving allegations of sexual assault, mediation with the complainant and the accused is not appropriate even on a voluntary basis.
14. If the allegation is rape or attempted rape or physical assault, it should be immediately reported to the police.
INVESTIGATION AND RESOLUTION - During the investigation, keep the following principles in mind:
1. Keep the University’s legal counsel informed of the status of the complaint and the progress of the investigation through the Office of the Vice President for Administration and Finance, who will inform the President.
2. Maintain confidentiality during the investigation process.
3. When interviewing witnesses and coworkers, ask broad, open-ended questions at first. Do not unnecessarily disclose information to witnesses. For example, instead of asking, “Did Paul touch Joan?” ask “Have you seen anyone display inappropriate behavior around Joan at work in a way that made her uncomfortable?” The purpose of the investigation is to gather facts not disseminate allegations. As the interview progresses, gradually ask for more specific information.
4. Interview the supervisors of the parties. Ask whether there are any relevant discipline problems. Are there any prior similar incidents involving the accused? Is there a history of any prior similar incidents involving other employees within the office or department?
5. Assure witnesses that they will not be retaliated against from cooperating in an investigation.
6. Carefully document all interviews, telephone calls or any other administrative activities responsive to the complaint.
4
7. If it is necessary to have the assistance of clerical staff during the investigation, use one such staff person who can be trusted to handle all clerical tasks, i.e. typing of memos and correspondence, setting up appointments with witnesses/coworkers.
8. Emphasize need for confidentiality to everyone involved in investigation.
9. Make note of all evidence regardless of whether it supports the merits of the complaint.
10. Evaluate the credibility of the parties and witnesses interviewed.
11. Complete the investigation and submit a written report or take other steps as called for under University policy. The written report should include a conclusion as to whether the University’s policy on harassment has been violated and specific reasons for the conclusion. If warranted the report should recommend reasonable, prompt, and effective action to end the harassment and remedy its effects on the victim. The report may recommend appropriate disciplinary sanctions.
12. Keep the complainant and accused advised of the progress of the investigation. Inform him or her of an anticipated completion date. Do not share substantive information with him or her.
13. The President will report the determination on the disposition of the matter to both parties.
14. Maintain confidentiality of accused even after investigation is complete and any necessary disciplinary actions are taken.
* It is appropriate to inform the accused of the name of the person who has made the allegations against him/her. Unless this is done, the accused cannot adequately respond to the allegations and your investigation will not be complete. Otherwise, the name of the accused and the details of the investigation should be kept strictly confidential and disclosed only on a need-to-know basis.
Any questions concerning these guidelines should be directed to the Vice President of Administration and Finance.
5
FM40432
INTERNATIONAL LIFT EQUIPMENT LTD.
London Office Midland Office
Units 1&2, Highams Park Ind Estate Wanlip Road
Larkshall Road Syston
London. E4 7HS Leicester, England LE7 1PD
Telephone (020) 85279669 Telephone 0116 2690900
Fax (020) 85270936 Fax 0116 2690939
Website: www.ileweb.com email: service@ilem.co.uk
TECHNICAL MANUAL
FOR THE
SHAFTENCODER
PROGRAMMING
TOOL
DATE: 16/09/2005 ISSUE NO. 1
Doc Ref: - Shaftencoder Programming Tool Manual.doc J.W.B
Issue details
1) Initial Launch.
WE RESERVE THE RIGHT TO ALTER WITHOUT GIVING PRIOR NOTICE TECHNICAL
DATA DIMENSIONS AND WEIGHTS DESCRIBED IN THIS MANUAL.
Contents
1 Introduction 3
2 Shaftencoder Programming Tool board layout / connections 4
3 Shaftencoder Programming 5
3.1 Display General Information 5
3.2 Set Pulse Per MM 6
3.3 PPM Learn 6
3.4 Set Floor levels 8
3.5 Set Slowing Distances 9
3.6 Set PX Lengths 10
3.7 Learning Run 6
3.8 Handwinding Mode 11
4 Technical Data 12
2
1 Introduction
The ILE Shaftencoder Programming Tool has been developed to allow the user on site to programme, interrogate, and alter parameters on the ILE Shaft encoder. The unit also incorporates extra features when used in conjunction with the ILE MRL System. The individual interfaces for each of the external devices are detailed below:
Shaft Encoder:
The Shaftencoder Programming Tool communicates with the Shaft encoder via the RS232 Serial link connection utilising a DB9 type Serial CABLE. This allows the user to amend, set, and view shaft encoder parameters. The options available are as follows:
Display General Information
Set Pulse Per Millimetre
Set Floor Levels
Set Slowing Distance
Set Px Lengths
Learn Pulses Per Millimetre
Perform Learning Run
These functions are described in detail in section 3 of this manual.
Hand winding Mode:
The Shaftencoder Programming Tool includes a hand winding positional display, with an audible floor level output. To operate in this mode a connection with an encoder is required and floor level signals from the shaftencoder. The hand winding display gives a description of the lifts position via the interface LCD, for example “LIFT JUST ABOVE ” “LIFT AT FLR LEVEL” etc.
The Shaftencoder Programming Tool has been developed to operate in conjunction with the ILE Emerald System, when used with the Emerald system extra features are available. In addition to giving a position display and audible floor level output, the unit also limits the speed of the lift during hand winding via the brake contact. This feature monitors the lift speed via the encoder to ensure the lift speed does not exceed 0.25m/s during handwind operation. This feature requires the correct setting of the “PPM” Pulse per Millimetre parameter.
3
2 Shaftencoder Programming Tool board layout / connections
The Shaftencoder Programming Tool inputs /outputs are as follows:
TA ENCODER CHANNEL A (CONNECTED TO SHAFTENCODER)
TB ENCODER CHANNEL B ( “ “ )
OVT ENCODER SUPPLY ( “ “ )
HWD HANDWINDING DOWN INPUT ( “ “ )
HRS HANDWINDING RESET INPUT ( “ “ )
HWR HANDWIND RETURN ( “ “ )
HW1 HANDWINDING INPUT (HANDWIND CONTACT IN PANEL)
HW2 HANDWINDING INPUT ( “ “ )
B- BATTERY CONNECTION TERMINAL (-)
B+ BATTERY CONNECTION TERMINAL (+)
AC1 9-30V AC/DC SUPPLY TERMINALS
AC2 ( “ “ )
1 BRAKE CONTACT TERMINALS
2 ( “ “ )
RS232 D TYPE 232 CONNECTOR
4
3 Shaftencoder Programming
Upon initialisation the screen will display the following
SHAFTENCODER
PROGRAMMING TOOL
PRESS ‘P’ FOR MENU
The menu is accessed by pressing the parameter button (labeled P), and is as follows:
DISPLAY GEN INFO SET PULSE/PER/MM SET FLOOR LEVELS SET SLOW DISTANCE SET PX LENGTHS LEARN PULSE/PER/MM START LEARNING RUN
The first four items are seen when the menu is accessed and the other three elements are accessed by scrolling the display. Each item in the menu is selected by placing the flashing cursor over the relevant line and pressing the return key.
The directions in which you can scroll will be signified by arrows displayed in the top left and bottom left of the screen.
The buttons that are active when in this function are:
? = move cursor up a line
? = move cursor down a line
? = select the option
3.1 Display General Information
This option on the menu retrieves the general information from the shaft encoder via the RS232 port. It is the equivalent of doing an ‘ecb’ command via a terminal window. Note the values for the floor vane length / px length / px gap, and slowing distances are displayed in millimetres.
Meas PPM =
Init stp dst=
Flrs seen=
Flr vane len=
Sp Pxlg PxGp Slw
0
1
4
5
The first four items appear when the option is accessed from the menu. Scrolling the display accesses the rest of the information.
The directions in which you can scroll will be signified by arrows displayed in the top left and bottom left of the screen.
The buttons that are active when in this function are:
P = Go back to menu
? = Move up
? = Move down
3.2 Set Pulse Per MM
This is to set the pulses per millimetre parameter for the shaftencoder, if the pulse per millimetre of the attached encoder is known the value can be entered and saved directly to the shaftencoder, therefore a learn pulse per millimetre, will not be required by the operator. The pulse per millimetre value is also stored by the programming tool, so that the lift can be run in handwind mode. The screen will display the following.
SET PULSE/PER/MM
0.000 p/mm
??? UP & DN ACTIVE
P = EXIT RETURN = SAVE
The buttons that are active in this function are:
P = Go back to the menu without saving the value
? = Go to previous digit
? = Go to next digit
? = Store the pulses per millimetre
? = Increase digit
? = Decrease digit
3.3 PPM Learn
Two button magnets should be positioned in between a floor EXACTLY 150mm apart on the guide in order to operate the proximity sensor. Operate the lift on test control through both magnets. Once both magnets have been seen the value for the Pulses per millimeter is returned by the shaft encoder via the RS232 port and will be displayed on the LCD display, this value will also be stored for future use as the handwind pulses per millimeter.
Note: If a “ learn” has been selected by pressing “return” it must then be performed as described above, before the user has the option to exit to the menu.
6
3.4 Set Floor Levels
This is done by changing the stopping distance, (distance from second vane to the floor level) for all or each individual floor; both are done from this option in the menu.
It is the equivalent of using the ecm(floor)(sign)(mm)(direction) command using a terminal window. The floor parameter signifies the floor that the command will change, the sign signifies whether it is a positive or negative change, the ‘mm’ is the number of millimeters by which the parameter will change and the direction is either up or down.
If –1 is used as a floor value then all floors are affected by the changes made, and the up/down parameter cannot be set.
The sign parameter is along side the millimeters parameter. When traversing across the parameters and digits the order is as follows (left to right):
Floor (digit2)
Floor (digit1)
Sign
Millimetres (digit5)
Millimetres (digit4)
Millimetres (digit3)
Millimetres (digit2)
Millimetres (digit1)
Direction
SET FLOOR LEVELS
FLOOR ‘mm’ U/D
P = EXIT RETURN = SAVE
The buttons that are active in this function are:
P = Go back to menu
? = Go to previous parameter/digit
? = Go to next parameter/digit
? = Send the command to the shaft encoder
? = Increase parameter/digit
? = Decrease parameter/digit
With the floor and the millimetres parameter you can set each digit up to a maximum of the number of floors allowed, and the highest change allowed in the number of millimetres. The floor is a 2 digit number and the millimetres are a 5 digit number.
When changing the floor levels for the up direction if the lift car is higher than a floor level you need to use the + sign, and if the lift car is lower than the floor level you need to use the – sign. Conversely when travelling in the down direction if the lift car is higher than a
7
floor level you need to use the – sign, and if the lift car is lower than the floor level then you need to use the + sign.
Once the return button is pressed to send the command to the shaft encoder, the screen will display ‘PARAM CHNGED & SAVED’. You can then set another floor level, or return to the menu.
3.5 Set Slowing Distances
This is done by changing the slowing distance for all or each individual floor, both are done from this option in the menu.
It is the equivalent of using the ecx(floor)(speed)(sign)(mm)(direction) command using a terminal window. The floor parameter signifies the floor that the command will change, the speed is the speed that it will affect, the sign is whether it is a positive or negative change, the ‘mm’ is the number of millimetres by which the parameter will change and the direction is either up or down.
If –1 is used as a floor value then all floors are affected by the changes made, and the up/down parameter cannot be set.
The sign parameter is along side the millimetres parameter. When traversing across the parameters and digits the order is as follows (left to right):
Floor (digit2)
Floor (digit1)
Speed
Sign
Millimetres (digit5)
Millimetres (digit4)
Millimetres (digit3)
Millimetres (digit2)
Millimetres (digit1)
Direction
SET SLOWING DISTANCE
FLOOR SP ‘mm’ U/D
P = EXIT RETURN = SAVE
The buttons that are active in this function are:
P = Go back to menu
? = Go to previous parameter/digit
? = Go to next parameter/digit
? = Send the command to the shaft encoder
? = Increase parameter/digit
? = Decrease parameter/digit
8
With the floor speed and the millimetres parameter you can set each digit up to a maximum of the number of floors allowed, speeds allowed, and the highest change allowed in the number of pulses. The floor is a 2 digit number, the speed a 1 digit number, and the millimetres is a 5 digit number.
SPEED DIGIT: “0” = LEVELLING SPEED, “1” = MEDIUM SPEED, “4” = HIGH SPEED
Once the return button is pressed to send the command to the shaft encoder, the screen will display ‘PARAM CHNGED & SAVED’. You can then set another slowing distance, or return to the menu.
3.6 Set PX Lengths
This is done by changing the slowing vane lengths for individual speeds.
It is the equivalent of using the ecv(speed)(mm) command using a terminal window. The speed parameter signifies the speed that the command will change, the sign is whether it is a positive or negative change, the ‘mm’ is the number of millimetres by which the parameter will change.
The sign parameter is along side the millimetres parameter. When traversing across the parameters and digits the order is as follows (left to right):
Speed (digit1)
Sign
Millimetres (digit5) Millimetres (digit4) Millimetres (digit3) Millimetres (digit2) Millimetres (digit1)
SET PX LENGTHS
SPEED ‘mm’
P = EXIT RETURN = SAVE
The buttons that are active in this function are:
P = Go back to menu
? = Go to previous parameter/digit
? = Go to next parameter/digit
? = Send the command to the shaft encoder
? = Increase parameter/digit
? = Decrease parameter/digit
With the speed and the millimetres parameter you can set each digit up to a maximum of the number of speeds allowed, and the highest change allowed in the number of millimetres. The speed is a 1 digit number, and the millimetres are a 5 digit number.
Once the return button is pressed to send the command to the shaft encoder, the screen will display ‘PARAM CHNGED & SAVED’. You can then set another slowing vane length, or return to the menu.
3.7 Learning Run
Position the correction floor magnets on the guide so that the correction proximity is operated as near as possible to floor level.
All movement during a learning run is to be at reduced speed.
Prior to commencing the learning run the lift should be able to run between terminal floors at reduced speed, stopping on the terminal limits.
The lift should be positioned at the bottom floor on the down overtravel limit below the bottom floor correction point magnet and with the RSD limit switch activated. It will be necessary to short out the bottom terminal limit to place the lift on the overtravel limit.
Upon selecting the learning run option from the shaft encoder menu, the display will tell you when it is ready to run to the top of the shaft, by displaying ‘RUN TO TOP’ on the display. Run the lift in panel test mode, operate the override button to get the lift off the overtravel limit and continue to run the lift up onto the top overtravel limit, again it will be necessary to short out the top terminal limit.
The display will tell you when it is ready to run to the bottom of the shaft by displaying ‘RUN TO BOTTOM’ on the display. Run the lift onto the bottom overtravel limit, again shorting out the bottom terminal limit.
The display will inform you of the number of floors seen on the way up and the way down, these should be the same, if not the learning run needs to be repeated.
Note: If “learn” has been selected by pressing “return” a learning run must be performed as described above, before the user has the option to exit to the menu. Failure to perform an actual learning run once selected will result in loss of shaftencoder data.
10
3.8 Handwinding Mode
This gives feedback to the operator as to the position of the lift in the shaft. The screen will display the following:
! HANDWIND CONTROL !
0 PPS .00M/S
On the second and third lines the status of the lift will be displayed, on the second line any of the following will be displayed:
LIFT BELOW
LIFT ABOVE
LIFT BETWEEN
LIFT AT FLR LEVEL
POSITION LOST
On the third line the actual level will be displayed, and the brake status if tripped will be
displayed
on the fourth line.
So if the lift is travelling between floors 3 and 4, the screen would display the following:
! HANDWIND CONTROL !
LIFT BETWEEN
FLOORS 3 AND 4
500 PPS 0.22M/S
11
4 Technical Data
Power supply:
9-30V AC/DC
12V Battery Input
@9V current = 200mA
@24V current = 80mA
@30V current = 70mA
Power supply peak current:
90mA @24V D.C
Power supply standby current:
80mA @24V D.C
Interface Inputs:
HW1/2 Opto isolated inputs 9 to 30V D.C
HWU/HWD/HRS Opto isolated inputs 9 to 30V D.C
Interface Outputs:
Solid state NPN 4A power transistors for position outputs
Brake relays 10A @250V D.C
12
JAGS Version 4.3.0 installation manual
Martyn Plummer Bill Northcott Matt Denwood
10 July 2017
JAGS is distributed in binary format for Microsoft Windows, macOS, and most Linux distributions. The following instructions are for those who wish to build JAGS from source. The manual is divided into three sections with instructions for Linux/Unix, macOS, and Windows.
1 Linux and UNIX
JAGS follows the usual GNU convention of
./configure
make
make install
which is described in more detail in the file INSTALL in the top-level source directory. On some UNIX platforms, you may be required to use GNU make (gmake) instead of the native make command. On systems with multiple processors, you may use the option ‘-j’ to speed up compilation, e.g. for a quad-core PC you may use:
make -j4
If you have the cppunit library installed then you can test the build with
make check
WARNING. If you already have a copy of the jags library installed on your system then the test program created by make check will link against the installed library and not the one in your build directory. So if the test suite includes a regression test for a bug that was fixed in the version you are building but a previous version of JAGS is already installed then the unit tests will fail. Best practice is to run the tests after make install so the build and installed versions are the same.
1.1 Configure options
At configure time you also have the option of defining options such as: • The names of the C, C++, and Fortran compilers.
* Optimization flags for the compilers. JAGS is optimized by default if the GNU compiler (gcc) is used. If you are using another compiler then you may need to explicitly supply optimization flags.
* Installation directories. JAGS conforms to the GNU standards for where files are installed. You can control the installation directories in more detail using the flags that are listed when you type ./configure --help.
1.1.1 Configuration for a 64-bit build
By default, JAGS will install all libraries into ‘/usr/lo cal/lib’. If you are building a 64-bit version of JAGS, this may not be appropriate for your system. On Fedora and other RPM-based distributions, for example, 64-bit libraries should be installed in ‘lib64’, and on
1
Solaris, 64-bit libraries are in a subdirectory of ‘lib’ (e.g. ‘lib/amd64’ if you are using a x86-64 processor), whereas on Debian, and other Linux distributions that conform to the FHS, the correct installation directory is ‘lib’.
To ensure that JAGS libraries are installed in the correct directory, you should supply the ‘--libdir’ argument to the configure script, e.g.:
./configure --libdir=/usr/local/lib64
It is important to get the installation directory right when using the rjags interface between R and JAGS, otherwise the rjags package will not be able to find the JAGS library.
1.1.2 Configuration for a private installation
If you do not have administrative privileges, you may wish to install JAGS in your home directory. This can be done with the following configuration options
export JAGS_HOME=$HOME/jags #or wherever you want it ./configure --prefix=$JAGS_HOME
For more detailed control over the installation directories type ./configure --help
and read the section “Fine-tuning of the installation directories.”
With a private installation, you need to modify your PATH environment variable to include ‘$JAGS HOME/bin’. You may also need to set LD LIBRARY PATH to include ‘$JAGS HOME/lib’ (On Linux this is not necessary as the location of libjags and libjrmath is hard-coded into the JAGS binary).
1.2 BLAS and LAPACK
BLAS (Basic Linear Algebra System) and LAPACK (Linear Algebra Pack) are two libraries of routines for linear algebra. They are used by the multivariate functions and distributions in the bugs module. Most unix-like operating system vendors supply shared libraries that provide the BLAS and LAPACK functions, although the libraries may not literally be called “blas” and “lapack”. During configuration, a default list of these libraries will be checked. If configure cannot find a suitable library, it will stop with an error message.
You may use alternative BLAS and LAPACK libraries using the configure options --with-blas and --with-lapack
./configure --with-blas="-lmyblas" --with-lapack="-lmylapack"
If the BLAS and LAPACK libraries are in a directory that is not on the default linker path, you must set the LDFLAGS environment variable to point to this directory at configure time:
LDFLAGS="-L/path/to/my/libs" ./configure ...
At runtime, if you have linked JAGS against BLAS or LAPACK in a non-standard location, you must supply this location with the environment variable LD LIBRARY PATH, e.g.
LD_LIBRARY_PATH="/path/to/my/libs:${LD_LIBRARY_PATH}"
2
Alternatively, you may hard-code the paths to the blas and lapack libraries at compile time. This is compiler and platform-specific, but is typically achieved with
LDFLAGS="-L/path/to/my/libs -R/path/to/my/libs
JAGS can also be linked to static BLAS and LAPACK if they have both been compiled with the -fPIC flag. You will probably need to do a custom build of BLAS and LAPACK if you require this. The configure options for JAGS are then:
./configure --with-blas="-L/path/to/my/libs -lmyblas -lgfortran -lquadmath" \ --with-lapack="-L/path/to/my/libs -lmylapack"
Note that with static linking you must add the dependencies of the BLAS library manually. The LAPACK library will pick up the same dependencies. Note also that libtool does not like linking directly to archive files. If you attempt a configuration of the form
--with-blas="/path/to/my/libs/myblas.a"
then this will pass at configure time but “make” will not correctly build the JAGS modules. 1.2.1 Multithreaded BLAS and LAPACK
Some high-performance computing libraries offer multi-threaded versions of the BLAS and LAPACK libraries. Although instructions for linking against some of these libraries are given below, this should not be taken as encouragement to use multithreaded BLAS. Testing shows that using multiple threads in BLAS can lead to significantly worse performance while using up substantially more computing resources.
1.3 GNU/Linux
GNU/Linux is the development platform for JAGS, and a variety of different build options have been explored, including the use of third-party compilers and linear algebra libraries.
1.3.1 Fortran compiler
The GNU FORTRAN compiler changed between gcc 3.x and gcc 4.x from g77 to gfortran. Code produced by the two compilers is binary incompatible. If your BLAS and LAPACK libraries are linked against libgfortran, then they were built with gfortran and you must also use this to compile JAGS.
Most recent GNU/Linux distributions have moved completely to gcc 4.x. However, some older systems may have both compilers installed. Unfortunately, if g77 is on your path then the configure script will find it first, and will attempt to use it to build JAGS. This results in a failure to recognize the installed BLAS and LAPACK libraries. In this event, set the F77 variable at configure time.
F77=gfortran ./configure
1.3.2 BLAS and LAPACK
The BLAS and LAPACK libraries from Netlib (http://www.netlib.org) should be provided as part of your Linux distribution. If your Linux distribution splits packages into “user” and “developer” versions, then you must install the developer package (e.g. blas-devel and lapack-devel).
On Red Hat Enterprise Linux (RHEL) you need to activate an optional repository in order to have access to BLAS and LAPACK. The repository is called is called rhel-<v>-<type>-optional-rpms, where <v> is the RHEL release version and <type> is the type of installation (server or workstation). Find the corresponding entry in the file ‘/etc/yum.repos.d/redhat.repo’ and change the line enabled = 0 to enabled = 1.
Suse Linux Enterprise Server (SLES) does not include BLAS and LAPACK in the main distribution. They are included in the SLES SDK, on a set of CD/DVD images which can be downloaded from https://download.suse.com/index.jsp (subscription and login required).
1.3.3 ATLAS
On Fedora Linux, pre-compiled atlas libraries are available via the atlas and atlas-devel RPMs. These RPMs install the atlas libraries in the non-standard directory /usr/lib/atlas (or /usr/lib64/atlas for 64-bit builds) to avoid conflicts with the standard blas and lapack RPMs. To use the atlas libraries, you must supply their location using the LDFLAGS variable (see section 1.2)
./configure LDFLAGS="-L/usr/lib/atlas"
Runtime linking to the correct libraries is ensured by the automatic addition of /usr/lib/atlas to the linker path (see the directory /etc/ld.so.conf.d), so you do not need to set the environment variable LD LIBRARY PATH at run time.
1.3.4 AMD Core Math Library
The AMD Core Math Library (acml) provides optimized BLAS and LAPACK routines for AMD processors. To link JAGS with acml, you must supply the acml library as the argument to --with-blas. It is not necessary to set the --with-lapack argument as acml provides both sets of functions. See also section 1.2 for run-time instructions.
For example, to link to the 64-bit acml using gcc 4.0+:
LDFLAGS="-L/opt/acml4.3.0/gfortran64/lib" \
./configure --with-blas="-lacml -lacml_mv"
The acmv mv library is a vectorized math library that exists only for the 64-bit version and is omitted when linking against 32-bit acml.
On multi-core systems, you may wish to use the threaded acml library (See the warning in section 1.2.1 however). To do this, link to acml mp and add the compiler flag ‘-fopenmp’:
LDFLAGS="-L/opt/acml4.3.0/gfortran64_mp/lib" \
CXXFLAGS="-O2 -g -fopenmp" ./configure --with-blas="-lacml_mp -lacml_mv"
The number of threads used by multi-threaded acml may be controlled with the environment variable OMP NUM THREADS.
4
1.3.5 Intel Math Kernel Library
The Intel Math Kernel library (MKL) provides optimized BLAS and LAPACK routines for Intel processors. MKL is designed to be linked to executables, not shared libraries. This means that it can only be linked to a static version of JAGS, in which the JAGS library and modules are linked into the main executable. To build a static version of JAGS, use the configure option ‘--disable-shared’.
MKL version 10.0 and above uses a “pure layered” model for linking. The layered model gives the user fine-grained control over four different library layers: interface, threading, computation, and run-time. Some examples of linking to MKL using this layered model are given below. These examples are for GCC compilers on x86 64. The choice of interface layer is important on x86 64 since the Intel Fortran compiler returns complex values differently from the GNU Fortran compiler. You must therefore use the interface layer that matches your compiler (mkl intel* or mkl gf*).
For further guidance, consult the MKL Link Line advisor at http://software.intel. com/en-us/articles/intel-mkl-link-line-advisor.
Recent versions of MKL include a shell script that sets up the environment variables necessary to build an application with MKL.
source /opt/intel/composerxe-2011/mkl/bin/mklvars.sh intel64
After calling this script, you can link JAGS with a sequential version of MKL as follows:
./configure --disable-shared \
--with-blas="-lmkl_gf_lp64 -lmkl_sequential -lmkl_core -lpthread"
Note that libpthread is still required, even when linking to sequential MKL. Threaded MKL may be used with:
./configure --disable-shared \
--with-blas="-lmkl_gf_lp64 -lmkl_gnu_thread -lmkl_core -liomp5 -lpthread"
The default number of threads will be chosen by the OpenMP software, but can be controlled
by setting OMP NUM THREADS or MKL NUM THREADS. (See the warning in section 1.2.1 however).
1.3.6 Using Intel Compilers
JAGS has been successfully built with the Intel Composer XE compilers. To set up the environment for using these compilers call the ‘compilervars.sh’ shell script, e.g.
source /opt/intel/composerxe-2011/bin/compilervars.sh intel64
Then call the configure script with the Intel compilers:
CC=icc CXX=icpc F77=ifort ./configure
1.3.7 Using Clang
JAGS has been built with the clang compiler for C and C++ (version 3.1). The configuration was
LD="llvm-ld" CC="clang" CXX="clang++" ./configure
In this configuration, the gfortran compiler was used for Fortran and the C++ code was linked to the GNU standard C++ library (libstdc++) rather than the version supplied by the LLVM project (lib c++).
5
1.4 Solaris
JAGS has been successfully built and tested on the Intel x86 platform under Solaris 11.3 using the Oracle Developer Studio 12.6 compilers.
I experienced some difficulty with the libtool dynamic linker ltdl on Solaris. This is due to the fact that output from the solaris utility nm does not match what libtool expects. This can be overcome by exporting the environment variable NM:
export NM=gnm
to use the GNU version of nm.
The C++ library ‘libCstd’ is not supported. You must therefore add the option ‘-library=stlport4’
to ‘CXXFLAGS’ to use the alternative STLPort4 library,
export LEX=flex
CC=cc CXX="CC -std=sun03" F77=f95 ./configure \
CFLAGS="-O3 -xarch=sse2" \
CXXFLAGS="+w -O3 -xarch=sse2 -library=stlport4 -lCrun"
or ‘-library=-stdcpp’ for the libstdc++ library.
./configure CC=cc CXX="CC -std=c++03" F77=f95 \
CFLAGS="-O3 -xarch=sse2" \
CXXFLAGS="+w -O3 -xarch=sse2 -library=stdcpp"
The Sun Studio compiler is not optimized by default. Use the option ‘-xO3’ for optimization (NB This is the letter “O” not the number 0) In order to use the optimization flag ‘-xO3’ you must specify the architecture with the ‘-xarch’ flag. The options above are for an Intel processor with SSE2 instructions. This must be adapted to your own platform.
To compile a 64-bit version of JAGS, add the option ‘-m64’ to all the compiler flags. On So-laris, 64-bit files are generally installed in an architecture-specific sub-directory (e.g. ‘amd64’ on the x86 platform). If you wish to conform to this convention for 64-bit JAGS then you should set the configure options ‘--libdir’, ‘--libexecdir’, and ‘--bindir’ appropriately.
The configure script automatically detects the Sun Performance library, which implements the BLAS/LAPACK functions.
6
2 macOS
A binary distribution of JAGS is provided for Mac OS X versions 10.9 to 10.11 and macOS 10.12 onwards, which is compatible with the current binary distribution of R and the corresponding rjags and runjags packages that are provided on CRAN. These instructions are only for those users that want to install JAGS from source.
The recommended procedure is to build JAGS using clang and the lib c++ standard library, which have been the default since OS 10.9. This provides compatibility with all builds of R available on CRAN from version 3.3.0 onwards, as well as “Mavericks builds” of earlier versions of R. Users needing to build against the libstdc++ library and/or with a version of Mac OS X predating 10.9 (Mavericks) should refer to the installation instructions given in older versions of the JAGS manual.
2.1 Required tools
If you wish to build from a released source package i.e. ‘JAGS-4.3.0.tar.gz’, you will need to install command line compilers. The easiest way to do this is using the Terminal application from ‘/Applications/Utilities’ - opening the application gives you a terminal with a UNIX shell. Run the following command on the terminal and follow the instructions:
xcode-select --install
You will also need to install the gfortran package, which you can download from: https://gcc.gnu.org/wiki/GFortranBinaries#MacOS
This setup should be sufficient to build the JAGS sources and also source packages in R. All the necessary libraries such as BLAS and LAPACK are included within macOS. Additional tools are required to run the optional test suite (see section 2.3).
2.2 Basic installation
2.2.1 Prepare the source code
Move the downloaded ‘JAGS-4.3.0.tar.gz’ package to some suitable working space on your disk and double click the file. This will decompress the package to give a folder called ‘JAGS-4.3.0’. You now need to re-open the Terminal and change the working directory to the JAGS source code. In the Terminal window after the $ prompt type cd followed by a space. In the Finder drag the ‘JAGS-4.3.0’ folder into the Terminal window and hit return. If this worked for you, typing ls followed by a return will list the contents of the JAGS folder.
2.2.2 Set up the environment
Before configuring JAGS it is first necessary to set a linker flag to include the Accelerate framework (https://developer.apple.com/documentation/accelerate). This allows the JAGS installation to use Apple’s implementation of BLAS.
Copy and paste the following command into the Terminal window:
export LDFLAGS="-framework Accelerate"
Other compiler options such as optimisation flags can also be set at this stage if desired, for example:
7
export CFLAGS="-Os"
export CXXFLAGS="-Os"
export FFLAGS="-Os"
Note that JAGS is usually compiled using -O2 by default, but it may be necessary to specify this explicitly depending on the version of the compiler being used.
2.2.3 Configuration
To configure the package type:
./configure
This instruction should complete without reporting an error.
2.2.4 Compile
To compile the code type:
make -j 8
The number ‘8’ indicates the number of parallel build threads that should be used (this will speed up the build process). In general this is best as twice the number of CPU cores in the computer - you may want to change the number in the instruction to match your machine. Again, this instruction should complete without errors.
2.2.5 Install
Finally to install JAGS you need to be using an account with administrator privileges. Type: sudo make install
This will ask for your account password and install the code ready to run as described in the User Manual. You need to ensure that /usr/local/bin is in your PATH in order for the command jags to work from a shell prompt.
2.3 Running the test suite
2.3.1 Installing CppUnit
As of JAGS version 4, a test suite is included with the source code that can be run to ensure that the compiled code produces the expected results. To run this code on your installation, you will need to download the CppUnit framework either using Homebrew (see section 2.4.1) or from:
http://freedesktop.org/wiki/Software/cppunit/
For the latter, download the source code under “Release Versions” corresponding to the latest release (currently Cppunit 1.14.0), unarchive the file, and then navigate a terminal window to the working directory inside the resulting folder. Then follow the usual terminal commands (as given on the website) to install CppUnit.
8
2.3.2 Running the tests
The test suite is run following the instructions given in section 2.2.5, using the following additional command:
make check
If successful, a summary of the checks will be given. If compiler errors are encountered, you may need to add the following compiler flag (and subsequent reconfiguration) in order to force the compiler to build with C++11, as required by CppUnit 1.14.0 and later:
export CXXFLAGS="-std=c++11 $CXXFLAGS"
./configure
make check
Note that the configuration step may also need to be repeated if CppUnit was not installed the first time this was run. In this case, you may also need to clean the existing compiled code before running make check using:
make clean
2.4 Tips for developers and advanced users
2.4.1 Additional tools
Some additional tools are required to work with code from the JAGS repository. The easiest way of obtaining the necessary utilities is using Homebrew, which can be installed by following the instructions at http://brew.sh
The following instructions have been verified to work with both Mac OS X 10.9 (Mavericks) and macOS 10.12 (Sierra), and should also work with other supported versions of OS X / macOS. If problems are encountered with these instructions on OS X 10.8 or earlier, then an alternative method of installing the required tools using e.g. MacPorts (as given in version 4.1.0 of the JAGS manual) may be more successful.
2.4.2 Working with the development code
If you want to work on code from the JAGS repository, you will need to build and install the auxillary GNU tools (autoconf, automake and libtool), as well as mercurial, bison, and flex as follows:
brew install mercurial
brew install autoconf
brew install automake
brew install libtool
brew install pkg-config
brew install bison
brew install flex
Note that CppUnit can also be installed using the same method:
9
brew install cppunit
The following sequence should then retrieve a clone of the current development branch of JAGS, and prepare the source code:
hg clone -r release-4_patched http://hg.code.sf.net/p/mcmc-jags/code-0
cd code-0
autoreconf -fis
The following modification to the PATH is also currently required to find the Homebrew versions of bison and flex:
export PATH="/usr/local/opt/bison/bin:/usr/local/opt/flex/bin:$PATH"
For more information see brew info bison and brew info flex
Once these commands have been run successfully, you should be able to follow the con
figuration and installation instructions from section 2.2.2 onwards.
2.4.3 Using ATLAS
Rather than using the versions of BLAS and LAPACK provided within OS X, it is possible to use ATLAS, which is available from http://math-atlas.sourceforge.net This can be either be installed by following the instructions given at http://math-atlas.sourceforge. net/atlas_install/, or by using MacPorts (https://www.macports.org/) with the following Terminal command:
sudo port install atlas
Once ATLAS is successfully installed, the -framework Accelerate flag can be omitted from the instructions given in section 2.2.2.
10
3 Windows
These instructions use MinGW, the Minimalist GNU system for Windows. You need some familiarity with Unix in order to follow the build instructions but, once built, JAGS can be installed on any PC running windows, where it can be run from the Windows command prompt.
3.1 Preparing the build environment
You need to install the following packages
* The Rtools compiler suite for Windows
* MSYS
* NSIS, including the AccessControl plug-in
3.1.1 Rtools
Rtools is a set of compilers and utilities used for compiling R on Windows. Rtools can be downloaded from your nearest CRAN mirror (https://cran.r-project.org/bin/windows/ Rtools/). We only need the compilers, as we use the utilities provided by MSYS (See below). For this reason, we choose not to add Rtools to the Windows environment variable PATH when asked by the installer.
The JAGS binaries for Windows 4.0.0 and above are built with Rtools 3.3, which is based on gcc 4.6.3. We also successfully built JAGS with the TDM-GCC compilers (http: //tdm-gcc.tdragon.net) based on gcc 5.1.0. However, the resulting JAGS binary is not compatible with R. The rjags package can be successfully compiled and linked against JAGS built with TDM-GCC 5.1.0, and runs correctly on 64-bit R, but the package spontaneously crashes in 32-bit R.
3.1.2 MSYS
MSYS (the Minimal SYStem) is part of the MinGW project (Minimal GNU for Windows). It provides a bash shell for you to build Unix software. Download the MinGW installer ‘mingw-get-setup.exe’ from http://www.mingw.org. Run the installer and select msys-base (“A Basic MSYS Installation (meta)”) for installation and then select Apply Changes from the Installation menu. There is no need to install the developer toolkit (mingw-developer-toolkit) if you are working with a release tarball of JAGS. You should not install any of the compilers that come with MinGW as we shall be using the Rtools versions.
To make MSYS use the TDM compilers edit the file ‘c:/mingw/msys/1.0/etc/fstab’ to read
c:\Rtools\gcc-4.6.3\bin /mingw
This adds the Rtools compilers to your PATH inside the MSYS shell.
MSYS creates a home directory for you in ‘c:/mingw/msys/1.0/home/username’, where username is your user name under Windows. You will need to copy and paste the source files for LAPACK and JAGS into this directory.
At the time of writing, the MinGW installer does not create a shortcut for MSYS on either the desktop or the start menu, even when these options are requested. Create your
11
own shortcut to ‘c:/MingGW/msys/1.0/msys.bat’ which launches the MSYS shell. For completeness, you may wish to use the icon ‘c:/MinGW/msys/1.0/msys.ico’ for your shortcut.
3.1.3 NSIS
The Nullsoft Scriptable Install System (http://nsis.sourceforge.net) allows you to create a self-extracting executable that installs JAGS on the target PC. These instructions were tested with NSIS 2.46. You must also install the AccessControl plug-in for NSIS, which is available from http://nsis.sourceforge.net/AccessControl_plug-in. The plug-in is distributed as a zip file which is unpacked into the installation directory of NSIS.
3.2 Building LAPACK
Download the LAPACK source file from http://www.netlib.org/lapack to your MSYS home directory. We used version 3.5.0.
You need to build LAPACK twice: once for 32-bit JAGS and once for 64-bit JAGS. The instructions below are for 32-bit JAGS. To build 64-bit versions, repeat the instructions with the flag ‘-m32’ replaced by ‘-m64’ and start in a clean build directory. Note that you cannot cross-build 64-bit BLAS and LAPACK on a 32-bit Windows system. This is because the build process must run some 64-bit test programs.
Launch MSYS (‘c:/MingW/msys/1.0/msys.bat’) and unpack the tarball.
tar xfvz lapack-3.5.0.tgz
cd lapack-3.5.0
Copy the file ‘INSTALL/make.inc.gfortran’ to ‘make.inc’ in the top level source directory. Then edit ‘make.inc’ replacing the following lines:
FORTRAN = gfortran -m32
LOADER = gfortran -m32
Type
make blaslib
make lapacklib
This will create two static libraries ‘librefblas.a’ and ‘liblapack.a’. These are insufficient for
building JAGS: you need to create dynamic link library (DLL) for each one.
First create a definition file ‘libblas.def’ that exports all the symbols from the BLAS library
dlltool -z libblas.def --export-all-symbols librefblas.a
Then link this with the static library to create a DLL (‘libblas.dll’) and an import library (‘libblas.dll.a’)
gcc -m32 -shared -o libblas.dll -Wl,--out-implib=libblas.dll.a \ libblas.def librefblas.a -lgfortran
Repeat the same steps for the LAPACK library, creating an import library (‘liblapack.dll.a’) and DLL (‘liblapack.dll’)
dlltool -z liblapack.def --export-all-symbols liblapack.a
gcc -m32 -shared -o liblapack.dll -Wl,--out-implib=liblapack.dll.a \ liblapack.def liblapack.a -L./ -lblas -lgfortran
12
3.3 Compiling JAGS
Unpack the JAGS source
tar xfvz JAGS-4.0.0.tar.gz
cd JAGS-4.0.0
and configure JAGS for a 32-bit build
CC="gcc -m32" CXX="g++ -m32 -std=c++98" F77="gfortran -m32" \ LDFLAGS="-L/path/to/import/libs/ -Wl,--enable-auto-import" \ ./configure
where ‘/path/to/imp ort/libs’ is a directory that contains the 32-bit import libraries (‘libblas.dll.a’ and ‘liblapack.dll.a’). This must be an absolute path name, and not relative to the JAGS build directory.
After the configure step, type
make win32-install
This will install JAGS into the subdirectory ‘win/inst32’. Note that you must go straight from the configure step to make win32-install without the usual step of typing make on its own. The win32-install target resets the installation prefix, and this will cause an error if the source is already compiled.
To install the 64-bit version, clean the build directory
make clean
reconfigure JAGS for a 64-bit build:
CC="gcc -m64" CXX="g++ -m64" F77="gfortran -m64" \ LDFLAGS="-L/path/to/import/libs/ -Wl,--enable-auto-import" \ ./configure
Then type
make win64-install
This will install JAGS into the subdirectory ‘win/inst64’.
With both 32-bit and 64-bit installations in place you can create the installer. Normally you will want to distribute the blas and lapack libraries with JAGS. In this case, put the 32-bit DLLs and import libraries in the sub-directory ‘win/runtime32’ and the 64-bit DLLs and import libraries in the sub-directory ‘win/runtime64’. They will be detected and included with the distribution.
Make sure that the file ‘makensis.exe’, provided by NSIS, is in your PATH. For a typical installation of NSIS, on 64-bit windows:
PATH=$PATH:/c/Program\ Files\ \(x86\)/NSIS
Then type
make installer
After the build process finishes, the self extracting archive will be in the subdirectory ‘win’.
13
3.4 Running the unit tests
In order to run the unit tests on Windows you must first install the cppunit library from source. Download the file ‘cppunit-1.12.1.tar.gz’ from Sourceforge (http://sourceforge. net/projects/cppunit/files/cppunit/1.12.1/) and unpack it:
tar xfvz cppunit-1.12.1.tar.gz
cd cppunit-1.12.1
Then compile and install as follows:
CXX="g++ -m32" ./configure --prefix=/opt32 --disable-shared
make
make install
The configure option --prefix=/opt32 installs the 32-bit library into ‘/opt32’ instead of the default location /usr/lo cal. Using this option allows you to do a parallel installation of the 64-bit version of the library, by rebuilding with configure options CXX=g++ -m64 and --prefix=/opt64. The two installations will not interfere with each other.
The configure option --disable-shared prevents the creation of the DLL ‘lib ccpunit.dll’ and builds only the static library ‘lib cppunit.a’. Without this option, the unit tests will fail. One of the major limitations of static linking to the C++ runtime is that you cannot throw exceptions across a DLL boundary. Linking the test programs against ‘lib cppunit.dll’ will result in uncaught exceptions and apparent failures for some of the tests, so it must be disabled.1.
To run the unit tests, add the option --with-cppunit-prefix=/optXX when configuring JAGS where XX is 32 or 64. Then run make check after make winXX-install.
3.5 Running the examples
The BUGS classic examples (file ‘classic-bugs.tar.gz’ from the JAGS Sourceforge site) can be run from the Windows command prompt using the make command provided by Rtools. This requires adding Rtools to the Windows search path if it is not currently there.
set PATH=c:\Rtools\bin;%PATH%
You must have R installed, along with the packages rjags and coda, both of which are available from CRAN (cran.r-project.org).
It is necessary to add R to the search path and to set the variable R_LIBS. Note that here we are using the 64-bit version of R. You may use the 32-bit version by substituting i386 for x64.
set PATH=c:\Program Files\R\R-3.2.2\bin\x64;%PATH%
set R_LIBS=c:\Users\username\Documents\R\win-library\3.2
where ‘username’ is your Windows user name. Then
1One of the attractions of the TDM-GCC compilers is that they do allow exceptions across DLL boundaries with static linking. However, we are not currently using TDM-GCC to build the JAGS binaries
14
tar xfvz classic-bugs.tar.gz
cd classic-bugs
cd vol1
make Rcheck
will check all examples in volume 1 using the rjags package. Repeat for ‘vol2’ to complete the checks.
You can also run checks using the command line interface of JAGS. This requires adding JAGS to the search path and overriding the default name of the JAGS executable.
set PATH=c:\Program Files\JAGS\JAGS-4.0.0\bin\x64;%PATH%
set JAGS=jags.bat
Then
make check
in directory ‘vol1’ or ‘vol2’ will run the checks using the command line interface. 3.6 Using TDM-GCC compilers
This section documents the use of TDM-GCC compilers to build JAGS. TDM-GCC was used to build Windows binaries for JAGS 3.x.y, but has been dropped in favour of Rtools for the 4.x.y release series. One reason for this is that the 32-bit version of JAGS built with TDM-GCC 5.1.0 causes the rjags package to spontaneously crash. The 64-bit version runs correctly.
TDM-GCC has a nice installer, available from Sourceforge (follow the links on the main TDM-GCC web site (http://tdm-gcc.tdragon.net). Ensure that you download the TDM64 MinGW-w64 edition as this is capable of producing both 32-bit and 64-bit binaries. We tested JAGS with ‘tdm64-gcc-5.1.0-2.exe’ based on gcc 5.1.0.
Select a “Recommended C/C++” installation and customize it by selecting the Fortran compiler, which is not installed by default. The installer gives you the option of adding TDM-GCC ‘bin’ folder to the windows PATH variable. We choose not to do this, but added the ‘bin’ to the PATH within the MSYS shell by editing ‘c:/mingw/msys/1.0/etc/fstab’ to read
c:\TDM-GCC-64 /mingw
After installation of TDM-GCC, to force the compiler to use static linking, delete any import libraries (files ending in ‘.dll.a’ in the TDM-GCC tree. If you do not do this then you will need to distribute runtime DLLs from TDM-GCC with JAGS. You can easily do this by copying the DLLs to ‘runtime32’ and ‘runtime64’ before building the installer, as described above. Nevertheless, it is often more convenient to use static linking.
Installation proceeds in the same way as for the Rtools build but with two differences. Firstly, when building the DLLs for blas and lapack, you need to add the linker flag -lquadmath after -lgfortran. Secondly, when configuring JAGS you should set the environment variable
CPPFLAGS=-D_GLIBCXX_USE_CXX11_ABI=0
This is necessary because gcc 5.1.0 introduced a new application binary interface (ABI) for the C++ standard library (See https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_
15
dual_abi.html. The old ABI is still supported and is used if you set the above flag. If you want to link JAGS with any software compiled with an earlier version of gcc then you need to use the old ABI. Failure to do so will result in error messages about undefined symbols from the linker.
16
Tidak ada komentar:
Posting Komentar