
Deep Q Learning is a topic that has become very popular of late in the field of Machine Learning (ML), and especially when it comes to games. AlphaGo Zero and AlphaZero are two ML agents that have been created by Google utilizing Deep Q Networks (DQN) that have had massive success. AlphaGo Zero became a master of the game go, while AlphaZero was able to master the games of: go, chess, and shogi. The games of go and chess are played between two players, each player has the same number of pieces to move, and the move sets of each player are equivalent.Instead of working with a game where both sides have equal pieces and move sets, I decided to work with a game that does not share those qualities. The game of fox and hounds is a simple game to both understand and play, but it deviates from chess and go by giving each player different pieces, move sets, and goals to accomplish. This makes the problem of creating ML agents more complex, as you now must create two different ML agents to play the same game instead of just one. I created a DQN agent for both the fox and hounds’ players to play the game. The results taught me how impactful different inputs, objectives, and outputs could be on the agents involved.
Read More