CSE 517A: Machine Learning

Spring 2007: Extra Credit Homework

Due 11:59pm, May 4, 2007

Purpose

Since we are not going to have a full fifth homework for the class, we're going to offer this set of questions as extra credit for those of you who want it. You don't have to do this homework, but if you want a few extra points, and you have the time, then here it is.

Assignment

Design and implement a genetic algorithm to solve the grid world problem from homework 4 part 7. Be careful to describe the representation you choose for your individuals, exactly how your three operators (selection, crossover, and mutation) modify these individuals, and your fitness function. Run a set of experiments to see how varying the population size, mutation rate, and other parameters affects the quality of the final solution, and the time taken to reach this solution. Report on these experiments, showing the learned policies, and compare them to those learned by Q-learning in homework 4. If you did not attempt this question in homework 4, you can use the results of another student as a comparison (but you cannot use any of their code in your GA implementation). [30 points].

What to Hand In

Write a report on your design, implementation, and experiments, and email it to the instructor by the deadline. Include an electronic copy of your code.
Page written by Bill Smart.