checkers_game

Reinforcement learning 'American Checkers 6x6' project.

Main file

Training the agents

• To train the agents set the boolean 'training' variable to True.

training = True

• Adjust the amount of training episodes.

train(1000)

• Run 'main' and wait some time for training.

Playing agent vs agent

• To simulate playing of agent vs agent set 'training' variable to False and 'playWithGui' to False.

training = False
playWithGui = False

• Adjust the amount of playing episodes.

play_game(1000)

• Run 'main' and wait some time for agents to finish playing.

• Resulting graph of agent's performance will be shown. The graph is made for the agent making first move.

Playing agent vs human

• To play against the agent set 'training' variable to False and 'playWithGui' to True.

training = False
playWithGui = True

• Run 'main'.

• You will play until you close the window or stop the program from your IDE.

Reinforcement learning

Approach

Basic idea was to train the agents using Q-learning approach. After training the agent we moved forward and added Deep Q-learning to compare the performance of the agents in this simple game.

About the comparison you can read in more details here.

Results

With a proper setup of the agents and with the same amount of training episodes we see the following graph where DQL agents performs better than his opponent.