Description Backgammon involves as large as 10^20 states. Thus it is impossible to have a table based Reinforcement Learning. I have instead used a Neural Network with hand-crafted features to represent the model. The weights are updated by the process of Temporal Difference Learning. To run the TD-Learning algorithm run: #TD_Gammon The final learned parameter is also provided in case you only wish to run the game.
Running Game To play the game with the computer, run the line given below on the command line. #GameSimulator The output of the codes is farely self explanatory which one can easily understand if one goes through the codes, specially TD_Gammon code. All the codes are well commented, so i strongly feel that it will be a smooth journey.