Question 1

Consider the feature-based reinforcement learner at: http://artint.info/demos/rl/sGameFA.html.

1. What happens if the gradient descent step size is too large? (What values are too large?)

2. What happens if the gradient descent step size is too small? (What values are too small?)

3. What gradient descent step size gives the best performance? (Tell us how you measure performance, and how you found this step size).