Mancala Learning via Temporal Difference

Greg Bickerman

Project Description

For my final project I will be training a neural network to play mancala. This project was inspired by TD-Gammon, the neural Backgammon player constructed by Gerald Tesauro.

Related Code

My project was coded in java. All associated files can be found here (mostly uncommented) The important classes are as follows:
Mancala -- The mancala board itself
MancalaGame -- Handles actual game play
MancalaNetwork -- The learning network
Test -- Testing class

Presentations

Presentation: Reinforcement Learning in Board Games
Final Presentation: Bickermancala