AI Methods

Coursework for - Konstanz (April 2001)

Introduction

  • Please include your name on each page of your assignment.
  • Please number each page
  • Your assignment must be typed and should be no longer than three A4 sides.
    This limit does not include any title page, nor the program listing, which you are required to hand in, as an appendix.

Note : This year, the coursework is not assessed. It is only been given out in case anybody is interested in having a go at it. I will be more than happy to mark and comment on your coursework – but it will not contribute to your overall mark for this course.

For this reason, this coursework is identical to last years.

Coursework

Develop a program that uses an evolutionary strategy. Your program should solve one of the problems outlined below.

In developing your program you should experiment with some of the following. It is recognised that you will not be able to test every scenario. Therefore, you should try to find the parameters that give reasonable results

  • Schemes that use ( + ) and (, )?
  • Try different values for  and ?
  • A strategy that uses a fixed standard deviation for its gaussian random variables
  • A strategy that changes the standard deviation using the “1/5” rule.
  • A strategy that changes the standard deviation value (that is, mutates this value, along with the other variables)

The deliverable for this coursework is a report that details the work you have done. The report should contain the following sections.

Section / % of Marks Awarded
A justification of the programming language you have chosen to use / 5
A description of your program design and how it has been implemented / 10
A discussion of the parameters you used and why you chose them / 20
A discussion on the results you achieved / 50
A conclusion of your work and what you think could be further done in order to improve the results further / 15

You should hand in your report and a listing of your source code. These can be sent to me via EMAIL if I have returned to the UK by the deadline date. If you can send a Word 97, that would be the most acceptable format.

The Problems

Blackjack

Your task is to develop an “agent” that learns how to play Blackjack (see the evolutionary strategy handout). You should define the rules you are trying to learn (using a simplified version, if you wish).

Your aim is to show that, over time, the agent gradually improves its play.

You might want to compare your agent against something how much it wins/loses once it has learnt how to play or against basic strategy (or both).

The Six Hump Problem

The six hump camelback function is defined in the handout and a simple implementation was shown in the lectures. One of the elements included in the lecture implementation was, instead of using a gaussian random number, a random number between zero and one was used as a mutation value.

If you decide to tackle this problem your task is to try and develop an evolutionary strategy that uses gaussian random numbers so that it out performs a method that simply uses random number in the interval zero to one.

Your Own Problem

You are welcome to come up with your own problem. It is probably best to use a problem where you are trying to optimise real valued variables – rather than discrete values.

If you choose this option, your report should define the function, state the global optimum (if known) or, failing that specify how you will measure the success of the evolutionary strategy.

D:\My Documents\Training & Courses\Lecture Courses\Konstanz\Konstanz Coursework.doc

Graham Kendall - 10/19/2018 - Page 1 of 2