AI for the Fanorona game
A neural network in order to solve the fanorona game.
- Firstly using a multithreaded Alpha-Beta tree
- Then using a pre-trained neural network, trained from the root of the game tree
- 2 weeks of training
- 3 layers (45, 7, 1)
- Activation function : Sigmoid
- Loss function : MSR
- Learning rate : 10e-5