chess-ai
Here’s a game where I (white) get smoked by my AI (black) :).
game.mov
Playing against the AI!
- Run
git clone https://github.com/samliu21/chess-ai. Navigate into the directory withcd chess-ai. - Create a virtual environment using
python -m venv .and activate it withsource bin/activate. - Install the necessary dependencies using
python -m pip install -r requirements.txt. - Call
python gui/main.pyto play!
GUI
The GUI was hand-made using the pygame and python-chess modules.
AI Architecture
This architecture was largely inspired by this Standford paper.
The AI uses two models. They both receive a board position as input and output an 8x8 matrix of softmax probabilities. The “from model” predicts the square to be moved out of and the “to model” predicts the square to be moved into.
The approach is best illustrated with an example. Consider the starting board position and the move: Nf3. The evaluation of this move is the product of the value at the g1 square of the from model and the value at the f3 square of the to model.
Among all legal moves, the largest product is the selected move.
The neural networks consist of five convolutional layers, followed by two affine layers and an output layer. A more detailed sketch of the architecture can be found below:
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 8, 8, 12)] 0 []
conv2d (Conv2D) (None, 8, 8, 32) 3488 ['input_1[0][0]']
batch_normalization (BatchNorm (None, 8, 8, 32) 128 ['conv2d[0][0]']
alization)
activation (Activation) (None, 8, 8, 32) 0 ['batch_normalization[0][0]']
conv2d_1 (Conv2D) (None, 8, 8, 64) 18496 ['activation[0][0]']
batch_normalization_1 (BatchNo (None, 8, 8, 64) 256 ['conv2d_1[0][0]']
rmalization)
activation_1 (Activation) (None, 8, 8, 64) 0 ['batch_normalization_1[0][0]']
conv2d_2 (Conv2D) (None, 8, 8, 256) 147712 ['activation_1[0][0]']
batch_normalization_2 (BatchNo (None, 8, 8, 256) 1024 ['conv2d_2[0][0]']
rmalization)
activation_2 (Activation) (None, 8, 8, 256) 0 ['batch_normalization_2[0][0]']
concatenate (Concatenate) (None, 8, 8, 512) 0 ['activation_2[0][0]',
'activation_2[0][0]']
conv2d_3 (Conv2D) (None, 8, 8, 256) 1179904 ['concatenate[0][0]']
batch_normalization_3 (BatchNo (None, 8, 8, 256) 1024 ['conv2d_3[0][0]']
rmalization)
activation_3 (Activation) (None, 8, 8, 256) 0 ['batch_normalization_3[0][0]']
concatenate_1 (Concatenate) (None, 8, 8, 320) 0 ['activation_3[0][0]',
'activation_1[0][0]']
conv2d_4 (Conv2D) (None, 8, 8, 256) 737536 ['concatenate_1[0][0]']
batch_normalization_4 (BatchNo (None, 8, 8, 256) 1024 ['conv2d_4[0][0]']
rmalization)
activation_4 (Activation) (None, 8, 8, 256) 0 ['batch_normalization_4[0][0]']
concatenate_2 (Concatenate) (None, 8, 8, 288) 0 ['activation_4[0][0]',
'activation[0][0]']
conv2d_5 (Conv2D) (None, 8, 8, 256) 663808 ['concatenate_2[0][0]']
batch_normalization_5 (BatchNo (None, 8, 8, 256) 1024 ['conv2d_5[0][0]']
rmalization)
activation_5 (Activation) (None, 8, 8, 256) 0 ['batch_normalization_5[0][0]']
dense (Dense) (None, 8, 8, 256) 65792 ['activation_5[0][0]']
batch_normalization_6 (BatchNo (None, 8, 8, 256) 1024 ['dense[0][0]']
rmalization)
dense_1 (Dense) (None, 8, 8, 64) 16448 ['batch_normalization_6[0][0]']
batch_normalization_7 (BatchNo (None, 8, 8, 64) 256 ['dense_1[0][0]']
rmalization)
dense_2 (Dense) (None, 8, 8, 1) 65 ['batch_normalization_7[0][0]']
batch_normalization_8 (BatchNo (None, 8, 8, 1) 4 ['dense_2[0][0]']
rmalization)
softmax (Softmax) (None, 8, 8, 1) 0 ['batch_normalization_8[0][0]']
==================================================================================================
Total params: 2,839,013
Trainable params: 2,836,131
Non-trainable params: 2,882
__________________________________________________________________________________________________
