Skip to content

Instantly share code, notes, and snippets.

@antoniSea

antoniSea/chess-ai.ipynb

Last active January 1, 2023 21:53
Show Gist options
  • Save antoniSea/628ec04e1486df6ccccf5a9e8674e23f to your computer and use it in GitHub Desktop.
Save antoniSea/628ec04e1486df6ccccf5a9e8674e23f to your computer and use it in GitHub Desktop.
Download ZIP
chess-ai.ipynb
Display the source blob
Display the rendered blob
Raw
chess-ai.ipynb
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"provenance": [],
"private_outputs": true,
"include_colab_link": true
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"accelerator": "GPU"
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
},
"source": [
"<a href=\"https://colab.research.google.com/gist/antoniSea/628ec04e1486df6ccccf5a9e8674e23f/chess-ai.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"cell_type": "code",
"metadata": {
"id": "fn0sCWoHYsbB"
},
"source": [
"!pip install python-chess==0.31.3"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "QauvWk2MkddY"
},
"source": [
"import chess\n",
"import chess.engine\n",
"import random\n",
"import numpy\n",
"\n",
"\n",
"# this function will create our x (board)\n",
"def random_board(max_depth=200):\n",
" board = chess.Board()\n",
" depth = random.randrange(0, max_depth)\n",
"\n",
" for _ in range(depth):\n",
" all_moves = list(board.legal_moves)\n",
" random_move = random.choice(all_moves)\n",
" board.push(random_move)\n",
" if board.is_game_over():\n",
" break\n",
"\n",
" return board\n",
"\n",
"\n",
"# this function will create our f(x) (score)\n",
"def stockfish(board, depth):\n",
" with chess.engine.SimpleEngine.popen_uci('/content/stockfish') as sf:\n",
" result = sf.analyse(board, chess.engine.Limit(depth=depth))\n",
" score = result['score'].white().score()\n",
" return score"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "ULOEWyyfYqtq"
},
"source": [
"board = random_board()\n",
"board"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "QtZy4cR8ZMhq"
},
"source": [
"print(stockfish(board, 10))"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "Rdo64dA7dhBE"
},
"source": [
"squares_index = {\n",
" 'a': 0,\n",
" 'b': 1,\n",
" 'c': 2,\n",
" 'd': 3,\n",
" 'e': 4,\n",
" 'f': 5,\n",
" 'g': 6,\n",
" 'h': 7\n",
"}\n",
"\n",
"\n",
"# example: h3 -> 17\n",
"def square_to_index(square):\n",
" letter = chess.square_name(square)\n",
" return 8 - int(letter[1]), squares_index[letter[0]]\n",
"\n",
"\n",
"def split_dims(board):\n",
" # this is the 3d matrix\n",
" board3d = numpy.zeros((14, 8, 8), dtype=numpy.int8)\n",
"\n",
" # here we add the pieces's view on the matrix\n",
" for piece in chess.PIECE_TYPES:\n",
" for square in board.pieces(piece, chess.WHITE):\n",
" idx = numpy.unravel_index(square, (8, 8))\n",
" board3d[piece - 1][7 - idx[0]][idx[1]] = 1\n",
" for square in board.pieces(piece, chess.BLACK):\n",
" idx = numpy.unravel_index(square, (8, 8))\n",
" board3d[piece + 5][7 - idx[0]][idx[1]] = 1\n",
"\n",
" # add attacks and valid moves too\n",
" # so the network knows what is being attacked\n",
" aux = board.turn\n",
" board.turn = chess.WHITE\n",
" for move in board.legal_moves:\n",
" i, j = square_to_index(move.to_square)\n",
" board3d[12][i][j] = 1\n",
" board.turn = chess.BLACK\n",
" for move in board.legal_moves:\n",
" i, j = square_to_index(move.to_square)\n",
" board3d[13][i][j] = 1\n",
" board.turn = aux\n",
"\n",
" return board3d"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "gHONl_M1hG9i"
},
"source": [
"split_dims(board)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "6S7QNZqwmBOP"
},
"source": [
"import tensorflow.keras.models as models\n",
"import tensorflow.keras.layers as layers\n",
"import tensorflow.keras.utils as utils\n",
"import tensorflow.keras.optimizers as optimizers\n",
"\n",
"\n",
"def build_model(conv_size, conv_depth):\n",
" board3d = layers.Input(shape=(14, 8, 8))\n",
"\n",
" # adding the convolutional layers\n",
" x = board3d\n",
" for _ in range(conv_depth):\n",
" x = layers.Conv2D(filters=conv_size, kernel_size=3, padding='same', activation='relu', data_format='channels_first')(x)\n",
" x = layers.Flatten()(x)\n",
" x = layers.Dense(64, 'relu')(x)\n",
" x = layers.Dense(1, 'sigmoid')(x)\n",
"\n",
" return models.Model(inputs=board3d, outputs=x)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "3IjDiS3Bmo5m"
},
"source": [
"model = build_model(32, 4)\n",
"utils.plot_model(model, to_file='model_plot.png', show_shapes=True, show_layer_names=False)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "tAFSOFc8pJf8"
},
"source": [
"def build_model_residual(conv_size, conv_depth):\n",
" board3d = layers.Input(shape=(14, 8, 8))\n",
"\n",
" # adding the convolutional layers\n",
" x = layers.Conv2D(filters=conv_size, kernel_size=3, padding='same', data_format='channels_first')(board3d)\n",
" for _ in range(conv_depth):\n",
" previous = x\n",
" x = layers.Conv2D(filters=conv_size, kernel_size=3, padding='same', data_format='channels_first')(x)\n",
" x = layers.BatchNormalization()(x)\n",
" x = layers.Activation('relu')(x)\n",
" x = layers.Conv2D(filters=conv_size, kernel_size=3, padding='same', data_format='channels_first')(x)\n",
" x = layers.BatchNormalization()(x)\n",
" x = layers.Add()([x, previous])\n",
" x = layers.Activation('relu')(x)\n",
" x = layers.Flatten()(x)\n",
" x = layers.Dense(1, 'sigmoid')(x)\n",
"\n",
" return models.Model(inputs=board3d, outputs=x)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "CkOXxmoVyHdc",
"cellView": "both"
},
"source": [
"import tensorflow.keras.callbacks as callbacks\n",
"\n",
"\n",
"def get_dataset():\n",
"\tcontainer = numpy.load('dataset.npz')\n",
"\tb, v = container['b'], container['v']\n",
"\tv = numpy.asarray(v / abs(v).max() / 2 + 0.5, dtype=numpy.float32) # normalization (0 - 1)\n",
"\treturn b, v\n",
"\n",
"\n",
"x_train, y_train = get_dataset()\n",
"print(x_train.shape)\n",
"print(y_train.shape)"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "RyOYq9mv2ppC"
},
"source": [
"model.compile(optimizer=optimizers.Adam(5e-4), loss='mean_squared_error')\n",
"model.summary()\n",
"model.fit(x_train, y_train,\n",
" batch_size=2048,\n",
" epochs=1000,\n",
" verbose=1,\n",
" validation_split=0.1,\n",
" callbacks=[callbacks.ReduceLROnPlateau(monitor='loss', patience=10),\n",
" callbacks.EarlyStopping(monitor='loss', patience=15, min_delta=1e-4)])\n",
"\n",
"model.save('model.h5')"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "e4CfjcGorHzg"
},
"source": [
"# used for the minimax algorithm\n",
"def minimax_eval(board):\n",
" board3d = split_dims(board)\n",
" board3d = numpy.expand_dims(board3d, 0)\n",
" return model.predict(board3d)[0][0]\n",
"\n",
"\n",
"def minimax(board, depth, alpha, beta, maximizing_player):\n",
" if depth == 0 or board.is_game_over():\n",
" return minimax_eval(board)\n",
" \n",
" if maximizing_player:\n",
" max_eval = -numpy.inf\n",
" for move in board.legal_moves:\n",
" board.push(move)\n",
" eval = minimax(board, depth - 1, alpha, beta, False)\n",
" board.pop()\n",
" max_eval = max(max_eval, eval)\n",
" alpha = max(alpha, eval)\n",
" if beta <= alpha:\n",
" break\n",
" return max_eval\n",
" else:\n",
" min_eval = numpy.inf\n",
" for move in board.legal_moves:\n",
" board.push(move)\n",
" eval = minimax(board, depth - 1, alpha, beta, True)\n",
" board.pop()\n",
" min_eval = min(min_eval, eval)\n",
" beta = min(beta, eval)\n",
" if beta <= alpha:\n",
" break\n",
" return min_eval\n",
"\n",
"\n",
"# this is the actual function that gets the move from the neural network\n",
"def get_ai_move(board, depth):\n",
" max_move = None\n",
" max_eval = -numpy.inf\n",
"\n",
" for move in board.legal_moves:\n",
" board.push(move)\n",
" eval = minimax(board, depth - 1, -numpy.inf, numpy.inf, False)\n",
" board.pop()\n",
" if eval > max_eval:\n",
" max_eval = eval\n",
" max_move = move\n",
" \n",
" return max_move"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "C63ND0E_uffp"
},
"source": [
"board = chess.Board()\n",
"\n",
"while True:\n",
" move = get_ai_move(board, 1)\n",
" board.push(move)\n",
" print(f'\\n{board}')\n",
" if board.is_game_over():\n",
" break\n",
"\n",
" move = get_ai_move(board, 1)\n",
" board.push(move)\n",
" print(f'\\n{board}')\n",
" if board.is_game_over():\n",
" break"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "pyCtTjfI2aaW"
},
"source": [
"!cp \"/content/drive/My Drive/dataset.zip\" /content/dataset.zip\n",
"!unzip dataset.zip\n",
"!rm dataset.zip\n",
"!chmod +x stockfish"
],
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "O0sjAB1A2Y0U"
},
"source": [
"import random\n",
"random.seed(37)"
],
"execution_count": null,
"outputs": []
}
]
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment