Playing endgame chess with the Tsetlin Machine
MetadataShow full item record
Original versionOtterlei, S. J. & Reiersøl, J. A. H. (2020) Playing endgame chess with the Tsetlin Machine (Master's thesis). University of Agder, Grimstad
The report is about training an Artificial Intelligence(AI) that is able to play out endgames, using the existing solved endgame of chess to train the Tsetlin Machine on. This report describes the methods used to train and test a Tsetlin Ma-chine using both the convolutional and multiclass implementation. We have further tested out different methods to handle the data it trains on to investigate what methods work best. Where these methods are; to split the data for two machines for either white or black starting player, transforming the data to only be from one starting players perspective and one splitting based on results by first having one machine looking at win versus draw and loss, then a second machine for looking at draw versus loss. The results showed that some of the methods used, involving only looking at one players perspective, worked well for predicting if the board would lead to a win with perfect play. Since several om the methods achieved over90% accuracy in the testing, while the best achieves an accuracy of 95%. However the playing off the endgame was lacking, as the games mostly ended in draws even when the Tsetlin Machine should have been able to win. Such as only drawing against each other, and only drawing against Monte Carlo Tree Search.
Master's thesis in Information- and communication technology (IKT590)