Move files into chesspp module

chesspp/classic_mcts.py

@@ -0,0 +1,109 @@
+import chess
+import random
+import eval
+import util
+import numpy as np
+
+
+class ClassicMcts:
+
+    def __init__(self, board: chess.Board, color: chess.Color, parent=None, move: chess.Move | None = None,
+                 random_state: int | None = None):
+        self.random = random.Random(random_state)
+        self.board = board
+        self.color = color
+        self.parent = parent
+        self.move = move
+        self.children = []
+        self.visits = 0
+        self.legal_moves = list(board.legal_moves)
+        self.untried_actions = self.legal_moves
+        self.score = 0
+
+    def _expand(self) -> 'ClassicMcts':
+        """
+        Expands the node, i.e., choose an action and apply it to the board
+        :return:
+        """
+        move = self.random.choice(self.untried_actions)
+        self.untried_actions.remove(move)
+        next_board = self.board.copy()
+        next_board.push(move)
+        child_node = ClassicMcts(next_board, color=self.color, parent=self, move=move)
+        self.children.append(child_node)
+        return child_node
+
+    def _rollout(self, rollout_depth: int = 20) -> int:
+        """
+        Rolls out the node by simulating a game for a given depth.
+        Sometimes this step is called 'simulation' or 'playout'.
+        :return: the score of the rolled out game
+        """
+        copied_board = self.board.copy()
+        steps = 1
+        for i in range(rollout_depth):
+            if copied_board.is_game_over():
+                break
+
+            m = util.pick_move(copied_board)
+            copied_board.push(m)
+            steps += 1
+
+        return eval.score_manual(copied_board) // steps
+
+    def _backpropagate(self, score: float) -> None:
+        """
+        Backpropagates the results of the rollout
+        :param score:
+        :return:
+        """
+        self.visits += 1
+        # TODO: maybe use score + num of moves together (a win in 1 move is better than a win in 20 moves)
+        self.score += score
+        if self.parent:
+            self.parent._backpropagate(score)
+
+    def is_fully_expanded(self) -> bool:
+        return len(self.untried_actions) == 0
+
+    def _best_child(self) -> 'ClassicMcts':
+        """
+        Picks the best child according to our policy
+        :return: the best child
+        """
+        # NOTE: maybe clamp the score between [-1, +1] instead of [-inf, +inf]
+        choices_weights = [(c.score / c.visits) + np.sqrt(((2 * np.log(self.visits)) / c.visits))
+                           for c in self.children]
+        best_child_index = np.argmax(choices_weights) if self.color == chess.WHITE else np.argmin(choices_weights)
+        return self.children[best_child_index]
+
+    def _select_leaf(self) -> 'ClassicMcts':
+        """
+        Selects a leaf node.
+        If the node is not expanded is will be expanded.
+        :return: Leaf node
+        """
+        current_node = self
+        while not current_node.board.is_game_over():
+            if not current_node.is_fully_expanded():
+                return current_node._expand()
+            else:
+                current_node = current_node._best_child()
+
+        return current_node
+
+    def build_tree(self, samples: int = 1000) -> 'ClassicMcts':
+        """
+        Runs the MCTS with the given number of samples
+        :param samples: number of simulations
+        :return: best node containing the best move
+        """
+        for i in range(samples):
+            # selection & expansion
+            # rollout
+            # backpropagate score
+            node = self._select_leaf()
+            score = node._rollout()
+            node._backpropagate(score)
+
+        return self._best_child()

chesspp/engine.py

@@ -0,0 +1,60 @@
+from abc import ABC, abstractmethod
+import chess
+import chess.engine
+from classic_mcts import ClassicMcts
+import random
+
+
+class Engine(ABC):
+    color: chess.Color
+    """The side the engine plays (``chess.WHITE`` or ``chess.BLACK``)."""
+
+    def __init__(self, color: chess.Color):
+        self.color = color
+
+    @abstractmethod
+    def play(self, board: chess.Board) -> chess.engine.PlayResult:
+        """
+        Return the next action the engine chooses based on the given board
+        :param board: the chess board
+        :return: the engine's PlayResult
+        """
+        pass
+
+    @staticmethod
+    @abstractmethod
+    def get_name() -> str:
+        """
+        Return the engine's name
+        :return: the engine's name
+        """
+        pass
+
+
+class ClassicMctsEngine(Engine):
+    def __init__(self, color: chess.Color):
+        super().__init__(color)
+
+    @staticmethod
+    def get_name() -> str:
+        return "ClassicMctsEngine"
+
+    def play(self, board: chess.Board) -> chess.engine.PlayResult:
+        mcts_root = ClassicMcts(board, self.color)
+        mcts_root.build_tree()
+        best_move = max(mcts_root.children, key=lambda x: x.score).move if board.turn == chess.WHITE else (
+            min(mcts_root.children, key=lambda x: x.score).move)
+        return chess.engine.PlayResult(move=best_move, ponder=None)
+
+
+class RandomEngine(Engine):
+    def __init__(self, color: chess.Color):
+        super().__init__(color)
+
+    @staticmethod
+    def get_name() -> str:
+        return "Random"
+
+    def play(self, board: chess.Board) -> chess.engine.PlayResult:
+        move = random.choice(list(board.legal_moves))
+        return chess.engine.PlayResult(move=move, ponder=None)

chesspp/eval.py

@@ -0,0 +1,183 @@
+import chess
+import chess.engine
+import sys
+
+# Eval constants for scoring chess boards
+# Evaluation metric inspired by Tomasz Michniewski: https://www.chessprogramming.org/Simplified_Evaluation_Function
+
+PIECE_VALUES = {
+    chess.PAWN: 100,
+    chess.KNIGHT: 320,
+    chess.BISHOP: 330,
+    chess.ROOK: 500,
+    chess.QUEEN: 900,
+    chess.KING: 20000
+}
+
+pawn_eval = [
+    0, 0, 0, 0, 0, 0, 0, 0,
+    5, 10, 10, -20, -20, 10, 10, 5,
+    5, -5, -10, 0, 0, -10, -5, 5,
+    0, 0, 0, 20, 20, 0, 0, 0,
+    5, 5, 10, 25, 25, 10, 5, 5,
+    10, 10, 20, 30, 30, 20, 10, 10,
+    50, 50, 50, 50, 50, 50, 50, 50,
+    0, 0, 0, 0, 0, 0, 0, 0
+]
+knight_eval = [
+    -50, -40, -30, -30, -30, -30, -40, -50,
+    -40, -20, 0, 0, 0, 0, -20, -40,
+    -30, 0, 10, 15, 15, 10, 0, -30,
+    -30, 5, 15, 20, 20, 15, 5, -30,
+    -30, 0, 15, 20, 20, 15, 0, -30,
+    -30, 5, 10, 15, 15, 10, 5, -30,
+    -40, -20, 0, 5, 5, 0, -20, -40,
+    -50, -40, -30, -30, -30, -30, -40, -50
+]
+bishop_eval = [
+    -20, -10, -10, -10, -10, -10, -10, -20,
+    -10, 5, 0, 0, 0, 0, 5, -10,
+    -10, 10, 10, 10, 10, 10, 10, -10,
+    -10, 0, 10, 10, 10, 10, 0, -10,
+    -10, 5, 5, 10, 10, 5, 5, -10,
+    -10, 0, 5, 10, 10, 5, 0, -10,
+    -10, 0, 0, 0, 0, 0, 0, -10,
+    -20, -10, -10, -10, -10, -10, -10, -20
+]
+rook_eval = [
+    0, 0, 0, 5, 5, 0, 0, 0,
+    -5, 0, 0, 0, 0, 0, 0, -5,
+    -5, 0, 0, 0, 0, 0, 0, -5,
+    -5, 0, 0, 0, 0, 0, 0, -5,
+    -5, 0, 0, 0, 0, 0, 0, -5,
+    -5, 0, 0, 0, 0, 0, 0, -5,
+    5, 10, 10, 10, 10, 10, 10, 5,
+    0, 0, 0, 0, 0, 0, 0, 0
+]
+queen_eval = [
+    -20, -10, -10, -5, -5, -10, -10, -20,
+    -10, 0, 0, 0, 0, 0, 0, -10,
+    -10, 0, 5, 5, 5, 5, 0, -10,
+    -5, 0, 5, 5, 5, 5, 0, -5,
+    0, 0, 5, 5, 5, 5, 0, -5,
+    -10, 5, 5, 5, 5, 5, 0, -10,
+    -10, 0, 5, 0, 0, 0, 0, -10,
+    -20, -10, -10, -5, -5, -10, -10, -20
+]
+king_eval = [
+    20, 30, 10, 0, 0, 10, 30, 20,
+    20, 20, 0, 0, 0, 0, 20, 20,
+    -10, -20, -20, -20, -20, -20, -20, -10,
+    20, -30, -30, -40, -40, -30, -30, -20,
+    -30, -40, -40, -50, -50, -40, -40, -30,
+    -30, -40, -40, -50, -50, -40, -40, -30,
+    -30, -40, -40, -50, -50, -40, -40, -30,
+    -30, -40, -40, -50, -50, -40, -40, -30
+]
+king_endgame_eval = [
+    50, -30, -30, -30, -30, -30, -30, -50,
+    -30, -30,  0,  0,  0,  0, -30, -30,
+    -30, -10, 20, 30, 30, 20, -10, -30,
+    -30, -10, 30, 40, 40, 30, -10, -30,
+    -30, -10, 30, 40, 40, 30, -10, -30,
+    -30, -10, 20, 30, 30, 20, -10, -30,
+    -30, -20, -10,  0,  0, -10, -20, -30,
+    -50, -40, -30, -20, -20, -30, -40, -50
+]
+
+PIECE_TABLES = {
+    chess.WHITE: {
+        chess.PAWN: pawn_eval,
+        chess.KNIGHT: knight_eval,
+        chess.BISHOP: bishop_eval,
+        chess.ROOK: rook_eval,
+        chess.QUEEN: queen_eval,
+        chess.KING: king_eval,
+        'end_game_king': king_endgame_eval
+    },
+    chess.BLACK: {
+        chess.PAWN: list(reversed(pawn_eval)),
+        chess.KNIGHT: list(reversed(knight_eval)),
+        chess.BISHOP: list(reversed(bishop_eval)),
+        chess.ROOK: list(reversed(rook_eval)),
+        chess.QUEEN: list(reversed(queen_eval)),
+        chess.KING: list(reversed(king_eval)),
+        'end_game_king': list(reversed(king_endgame_eval))
+    }
+}
+
+
+def check_endgame(board: chess.Board) -> bool:
+    """
+    Endgame according to Tomasz Michniewski:
+        1. Both sides have no queens or
+        2. Every side which has a queen has additionally no other pieces or one minorpiece maximum.
+    """
+    queens_white = 0
+    minors_white = 0
+    queens_black = 0
+    minors_black = 0
+    for s in chess.SQUARES:
+        piece = board.piece_at(s)
+        if piece is None:
+            continue
+
+        if piece.piece_type == chess.QUEEN:
+            if piece.color == chess.WHITE:
+                queens_white += 1 
+            else:
+                queens_black += 1
+
+        if piece.piece_type == chess.BISHOP or piece.piece_type == chess.KNIGHT:
+            if piece.color == chess.WHITE:
+                minors_white += 1 
+            else:
+                minors_black += 1
+
+    return (queens_black == 0 and queens_white == 0) or ((queens_black >= 1 >= minors_black) or (
+                queens_white >= 1 >= minors_white))
+
+
+def score_manual(board: chess.Board) -> int:
+    """
+    Calculate the score of the given board regarding the given color
+    :param board: the chess board
+    :return: score metric
+    """
+    outcome = board.outcome()
+    if outcome is not None:
+        if outcome.termination == chess.Termination.CHECKMATE:
+            return sys.maxsize if outcome.winner == chess.WHITE else -sys.maxsize
+        else:  # draw
+            return 0
+
+    score = 0
+    for s in chess.SQUARES:
+        piece = board.piece_at(s)
+        if piece is None:
+            continue
+
+        if piece.color == chess.WHITE:
+            if piece.piece_type == chess.KING and check_endgame(board):
+                score += PIECE_VALUES[piece.piece_type] * PIECE_TABLES[chess.WHITE]['end_game_king'][s]
+            else:
+                score += PIECE_VALUES[piece.piece_type] * PIECE_TABLES[chess.WHITE][piece.piece_type][s]
+        else:
+            if piece.piece_type == chess.KING and check_endgame(board):
+                score -= PIECE_VALUES[piece.piece_type] * PIECE_TABLES[chess.BLACK]['end_game_king'][s]
+            else:
+                score -= PIECE_VALUES[piece.piece_type] * PIECE_TABLES[chess.BLACK][piece.piece_type][s]
+
+    return score
+
+
+def score_stockfish(board: chess.Board) -> chess.engine.PovScore:
+    """
+    Calculate the score of the given board using stockfish
+    :param board:
+    :return:
+    """
+    engine = chess.engine.SimpleEngine.popen_uci("./stockfish/stockfish-ubuntu-x86-64-avx2")
+    info = engine.analyse(board, chess.engine.Limit(depth=0))
+    engine.quit()
+    return info["score"]

chesspp/i_mcts.py

@@ -0,0 +1,51 @@
+import chess
+from abc import ABC, abstractmethod
+from i_strategy import IStrategy
+from typing import Dict
+
+
+class IMcts(ABC):
+
+    def __init__(self, board: chess.Board, strategy: IStrategy):
+        self.board = board
+
+    @abstractmethod
+    def sample(self, runs: int = 1000) -> None:
+        """
+        Run the MCTS simulation
+        :param runs: number of runs
+        :return:
+        """
+        pass
+
+    @abstractmethod
+    def apply_move(self, move: chess.Move) -> None:
+        """
+        Apply the move to the chess board
+        :param move: move to apply
+        :return:
+        """
+        pass
+
+    @abstractmethod
+    def get_children(self) -> list['IMcts']:
+        """
+        Return the immediate children of the root node
+        :return: list of immediate children of mcts root
+        """
+        pass
+
+    @abstractmethod
+    def get_moves(self) -> Dict[chess.Move, int]:
+        """
+        Return all legal moves from this node with respective scores
+        :return: dictionary with moves as key and scores as values
+        """
+        pass
+
+"""
+TODO: add score class:
+how many moves until the end of the game?
+score ranges?
+perspective of white/black
+"""

chesspp/i_strategy.py

@@ -0,0 +1,8 @@
+from abc import ABC, abstractmethod
+
+# TODO extend class
+class IStrategy(ABC):
+
+    @abstractmethod
+    def pick_next_move(self, ):
+        pass

chesspp/lichess-engine.py

@@ -0,0 +1,20 @@
+from lichess_bot.lib.engine_wrapper import MinimalEngine, MOVE
+import chess.engine
+import engine
+
+
+class ProbStockfish(MinimalEngine):
+    def search(self, board: chess.Board, time_limit: chess.engine.Limit, ponder: bool, draw_offered: bool,
+               root_moves: MOVE) -> chess.engine.PlayResult:
+        moves = {}
+        untried_moves = list(board.legal_moves)
+        for move in untried_moves:
+            mean, std = engine.simulate_game(board, move, 10)
+            moves[move] = (mean, std)
+
+        return self.get_best_move(moves)
+
+    def get_best_move(self, moves: dict) -> chess.engine.PlayResult:
+        best_avg = max(moves.items(), key=lambda m: m[1][0])
+        next_move = best_avg[0]
+        return chess.engine.PlayResult(next_move, None)

chesspp/simulation.py

@@ -0,0 +1,72 @@
+import multiprocessing as mp
+import random
+import chess
+import chess.pgn
+from typing import Tuple, List
+from enum import Enum
+from dataclasses import dataclass
+
+from engine import Engine
+
+
+class Winner(Enum):
+    Engine_A = 0
+    Engine_B = 1
+    Draw = 2
+
+
+@dataclass
+class EvaluationResult:
+    winner: Winner
+    game: chess.pgn.Game
+
+
+def simulate_game(white: Engine, black: Engine) -> chess.pgn.Game:
+    board = chess.Board()
+
+    is_white_playing = True
+    while not board.is_game_over():
+        play_result = white.play(board) if is_white_playing else black.play(board)
+        board.push(play_result.move)
+        is_white_playing = not is_white_playing
+
+    game = chess.pgn.Game.from_board(board)
+    game.headers['White'] = white.get_name()
+    game.headers['Black'] = black.get_name()
+    return game
+
+
+class Evaluation:
+    def __init__(self, engine_a: Engine.__class__, engine_b: Engine.__class__):
+        self.engine_a = engine_a
+        self.engine_b = engine_b
+
+    def run(self, n_games=100) -> List[EvaluationResult]:
+        with mp.Pool(mp.cpu_count()) as pool:
+            args = [(self.engine_a, self.engine_b) for i in range(n_games)]
+            return pool.map(Evaluation._test_simulate, args)
+
+    @staticmethod
+    def _test_simulate(arg: Tuple[Engine.__class__, Engine.__class__]) -> EvaluationResult:
+        engine_a, engine_b = arg
+        flip_engines = bool(random.getrandbits(1))
+        if flip_engines:
+            black, white = engine_a(chess.BLACK), engine_b(chess.WHITE)
+        else:
+            white, black = engine_a(chess.WHITE), engine_b(chess.BLACK)
+
+        game = simulate_game(white, black)
+        winner = game.end().board().outcome().winner
+
+        result = Winner.Draw
+        match (winner, flip_engines):
+            case (chess.WHITE, True):
+                result = Winner.Engine_B
+            case (chess.BLACK, True):
+                result = Winner.Engine_A
+            case (chess.WHITE, False):
+                result = Winner.Engine_A
+            case (chess.BLACK, False):
+                result = Winner.Engine_B
+
+        return EvaluationResult(result, game)

chesspp/util.py

@@ -0,0 +1,79 @@
+import chess
+import chess.engine
+from stockfish import Stockfish
+import numpy as np
+import random
+
+
+def pick_move(board: chess.Board) -> chess.Move | None:
+    """
+    Pick a random move
+    :param board: chess board
+    :return: a valid move or None if no valid move available
+    """
+    if len(list(board.legal_moves)) == 0:
+        return None
+    return random.choice(list(board.legal_moves))
+
+
+def simulate_game(board: chess.Board, move: chess.Move, depth: int):
+    """
+    Simulate a game starting with the given move
+    :param board: chess board
+    :param move: chosen move
+    :param depth: number of moves that should be simulated after playing the chosen move
+    :return: the score for the simulated game
+    """
+    engine = chess.engine.SimpleEngine.popen_uci("./stockfish/stockfish-ubuntu-x86-64-avx2")
+    board.push(move)
+    for i in range(depth):
+        if board.is_game_over():
+            engine.quit()
+            return
+        r = engine.play(board, chess.engine.Limit(depth=2))
+        board.push(r.move)
+
+    engine.quit()
+
+
+def simulate_stockfish_prob(board: chess.Board, move: chess.Move, games: int = 10, depth: int = 10) -> (float, float):
+    """
+    Simulate a game using
+    :param board: chess board
+    :param move: chosen move
+    :param games: number of games that should be simulated after playing the move
+    :param depth: simulation depth per game
+    :return:
+    """
+    board.push(move)
+    copied_board = board.copy()
+    scores = []
+
+    stockfish = Stockfish("./stockfish/stockfish-ubuntu-x86-64-avx2", depth=2, parameters={"Threads": 8, "Hash": 2048})
+    stockfish.set_elo_rating(1200)
+    stockfish.set_fen_position(board.fen())
+
+    def reset_game():
+        nonlocal scores, copied_board, board
+        score = eval.score_stockfish(copied_board).white().score(mate_score=100_000)
+        scores.append(score)
+        copied_board = board.copy()
+        stockfish.set_fen_position(board.fen())
+
+    for _ in range(games):
+        for d in range(depth):
+            if copied_board.is_game_over() or d == depth - 1:
+                reset_game()
+                break
+
+            if d == depth - 1:
+                reset_game()
+
+            top_moves = stockfish.get_top_moves(3)
+            chosen_move = random.choice(top_moves)['Move']
+            stockfish.make_moves_from_current_position([chosen_move])
+            copied_board.push(chess.Move.from_uci(chosen_move))
+
+    print(scores)
+    # TODO: return distribution here?
+    return np.array(scores).mean(), np.array(scores).std()