M2ICAL Analyses HC-Gammon

Wee-Chong Oon, Martin Henz

We analyse Pollack and Blair's HC-Gammon backgammon program using a new technique that performs {M}onte Carlo simulations to derive a {M}arkov Chain model for {I}mperfect {C}omparison {AL}gorithms, called the M2ICAL method, which models the behavior of the algorithm using a Markov chain, each of whose states represents a class of players of similar strength. The Markov chain transition matrix is populated using Monte Carlo simulations. Once generated, the matrix allows fairly accurate predictions of the expected solution quality, standard deviation and time to convergence of the algorithm. This allows us to make some observations on the validity of Pollack and Blair's conclusions, and also shows the application of the M2ICAL method on a previously published work.

Subjects: 12. Machine Learning and Discovery; 1.8 Game Playing

Submitted: Apr 15, 2007


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.