search

shanive / bgu-ailab-bandits

bandits expiriment
6 stars 0 forks source link

tune Cp for win-loss #43

Closed dtolpin closed 12 years ago

dtolpin commented 13 years ago

UCT is sensitive to the right value of Cp. Perform experiments with a few different values of Cp (like 0.125 * sqrt(2)^n) find the best value for UCT.

0.125, 0.125_sqrt(2), 0.125_2, 0.125_2_sqrt(2), .... until 2.00

shanive commented 13 years ago

added option in play.py. in order to experiment the value of cp, insert at command line: --initial-pc --step-cp (the actual step will be the square root of ). I hope this is what you wanted :)

dtolpin commented 13 years ago

http://www.arachnoid.com/linux/shell_programming.html

shanive commented 13 years ago

after improving some slow function I will run the experiment on server