AISpace2 Main Tools News Downloads Prototype Tools Customizable Applets Practice Exercises Help About AIspace

Experiment 1: SARSA, α=0.1, γ=0.5, Greedy Exploit = 80%, and initial Q-value = 0: 1152181 Experiment 6: SARSA, α=0.9, γ=0.5, Greedy Exploit = 80%, and initial Q-value = 0: 246605 Compare the results of the first experiment to the sixth experiment. The first experiment should get a significantly higher reward than the sixth experiment. The α parameter affects how much weight the last experience is given when updating the Q-value. When the weight is set too high (0.9 in experiment 6) this can allow the Q-value to diverge from its optimal value. For example, when performing the randomized 'up' action, a high α value could cause the Q-value to change significantly from its optimal value.

Main Tools: Graph Searching | Consistency for CSP | SLS for CSP | Deduction | Belief and Decision Networks | Decision Trees | Neural Networks | STRIPS to CSP