Added basic bayes-mcts using beta distribution

main.py

@@ -1,7 +1,10 @@
+import random
 import chess
 import chess.engine
 import chess.pgn
 from src.chesspp.classic_mcts import ClassicMcts
+from src.chesspp.baysian_mcts import BayesianMcts
+from src.chesspp.random_strategy import RandomStrategy
 from src.chesspp import engine
 from src.chesspp import util
 from src.chesspp import simulation, eval
@@ -24,6 +27,18 @@ def test_mcts():
         print("move (mcts):", c.move, " with score:", c.score)
 
 
+def test_bayes_mcts():
+    global lookup_count
+    fools_mate = "rnbqkbnr/pppp1ppp/4p3/8/5PP1/8/PPPPP2P/RNBQKBNR b KQkq f3 0 2"
+    board = chess.Board(fools_mate)
+    seed = 1
+    stategy = RandomStrategy(random.Random(seed))
+    mcts = BayesianMcts(board, stategy, seed)
+    mcts.sample()
+    for c in mcts.get_children():
+        print("move (mcts):", c.move, " with score:", c.mu)
+
+
 def test_stockfish():
     fools_mate = "rnbqkbnr/pppp1ppp/4p3/8/5PP1/8/PPPPP2P/RNBQKBNR b KQkq f3 0 2"
     board = chess.Board(fools_mate)

requirements.txt

@@ -1,4 +1,6 @@
 chess==1.10.0
 numpy==1.26.3
 stockfish==3.28.0
+torch==2.1.2
+pytest
 aiohttp

src/chesspp/baysian_mcts.py

@@ -0,0 +1,145 @@
+import chess
+from src.chesspp.i_mcts import *
+from src.chesspp.i_strategy import IStrategy
+from src.chesspp.util_gaussian import gaussian_ucb1, max_gaussian, beta_std, beta_mean
+from src.chesspp.eval import *
+import numpy as np
+import math
+
+
+class BayesianMctsNode(IMctsNode):
+    def __init__(self, board: chess.Board, strategy: IStrategy, parent: Self | None, move: chess.Move | None,
+                 random_state: random.Random, inherit_results: list[int] | None = None):
+        super().__init__(board, strategy, parent, move, random_state)
+        self.visits = 0
+        self.results = inherit_results.copy() if inherit_results is not None else [1, 1]
+
+        self._set_mu_sigma()
+
+    def _create_child(self, move: chess.Move):
+        copied_board = self.board.copy()
+        copied_board.push(move)
+        return BayesianMctsNode(copied_board, self.strategy, self, move, self.random_state, inherit_results=self.results)
+
+    def _set_mu_sigma(self):
+        alpha = self.results[0]
+        beta = self.results[1]
+
+        self.mu = beta_mean(alpha, beta)
+        self.sigma = beta_std(alpha, beta)
+
+    def _select_child(self) -> IMctsNode:
+        # select child by modified UCB1
+        if self.board.is_game_over():
+            return self
+
+        best_child = self.random_state.choice(self.children)
+        best_val = gaussian_ucb1(best_child.mu, best_child.sigma, self.visits)
+        for c in self.children:
+            g = gaussian_ucb1(c.mu, c.sigma, self.visits)
+
+            if g > best_val:
+                best_val = g
+                best_child = c
+        return best_child
+
+    def select(self) -> IMctsNode:
+        if len(self.children) == 0:
+            return self
+        else:
+            return self._select_child().select()
+
+    def expand(self) -> IMctsNode:
+        if self.visits == 0:
+            return self
+
+        for move in self.legal_moves:
+            self.children.append(self._create_child(move))
+
+        return self._select_child()
+
+    def rollout(self, rollout_depth: int = 20) -> int:
+        copied_board = self.board.copy()
+        steps = 1
+        for i in range(rollout_depth):
+            if copied_board.is_game_over():
+                break
+
+            m = self.strategy.pick_next_move(copied_board)
+            if m is None:
+                break
+
+            copied_board.push(m)
+            steps += 1
+
+        score = eval.score_manual(copied_board) // steps
+        if score > 0:
+            self.results[1] += 1
+        else:
+            self.results[0] += abs(score) // 50_000
+        return score
+
+    def backpropagate(self, score: int | None = None) -> None:
+        self.visits += 1
+
+        if score is not None:
+            self.results.append(score)
+
+        if len(self.children) == 0:
+            # leaf node
+            self._set_mu_sigma()
+        else:
+            # interior node
+            shuffled_children = self.random_state.sample(self.children, len(self.children))
+            max_mu = shuffled_children[0].mu
+            max_sigma = shuffled_children[0].sigma
+            for c in shuffled_children[1:]:
+                max_mu, max_sigma = max_gaussian(max_mu, max_sigma, c.mu, c.sigma)
+
+            if max_sigma == 0:
+                max_sigma = 0.001
+            self.mu = max_mu
+            self.sigma = max_sigma
+
+        if self.parent:
+            self.parent.backpropagate()
+
+    def print(self, indent=0):
+        print("\t"*indent + f"visits={self.visits}, mu={self.mu}, sigma={self.sigma}")
+        for c in self.children:
+            c.print(indent+1)
+
+
+class BayesianMcts(IMcts):
+    def __init__(self, board: chess.Board, strategy: IStrategy, seed: int | None = None):
+        super().__init__(board, strategy, seed)
+        self.root = BayesianMctsNode(board, strategy, None, None, self.random_state)
+        self.root.visits += 1
+
+    def sample(self, runs: int = 1000) -> None:
+        for i in range(runs):
+            #print(f"sample {i}")
+            leaf_node = self.root.select().expand()
+            _ = leaf_node.rollout()
+            leaf_node.backpropagate()
+            #self.root.print()
+
+    def apply_move(self, move: chess.Move) -> None:
+        self.board.push(move)
+
+        # if a child node contains the move, set this child as new root
+        for child in self.get_children():
+            if child.move == move:
+                self.root = child
+                self.root.parent = None
+                return
+
+        # if no child node contains the move, initialize a new tree.
+        self.root = BayesianMctsNode(self.board, self.root.strategy, None, None, self.random_state)
+
+    def get_children(self) -> list[IMctsNode]:
+        return self.root.children
+
+    def print(self):
+        print("================================")
+        self.root.print()

src/chesspp/classic_mcts.py

@@ -3,8 +3,8 @@ import random
 import numpy as np
 
 
-from chesspp import eval
-from chesspp import util
+from src.chesspp import eval
+from src.chesspp import util
 
 
 class ClassicMcts:

src/chesspp/engine.py

@@ -3,7 +3,7 @@ import chess
 import chess.engine
 import random
 import time
-from chesspp.classic_mcts import ClassicMcts
+from src.chesspp.classic_mcts import ClassicMcts
 
 class Limit:
     """ Class to determine when to stop searching for moves """

src/chesspp/i_mcts.py

@@ -1,13 +1,61 @@
 import chess
+import random
 from abc import ABC, abstractmethod
-from typing import Dict
-from chesspp.i_strategy import IStrategy
+from typing import Dict, Self
+from src.chesspp.i_strategy import IStrategy
+
+
+class IMctsNode(ABC):
+    def __init__(self, board: chess.Board, strategy: IStrategy, parent: Self | None, move: chess.Move | None,
+                 random_state: random.Random):
+        self.board = board
+        self.strategy = strategy
+        self.parent = parent
+        self.children = []
+        self.move = move
+        self.legal_moves = list(board.legal_moves)
+        self.random_state = random_state
+
+    @abstractmethod
+    def select(self) -> Self:
+        """
+        Selects the next node leaf node in the tree
+        :return:
+        """
+        pass
+
+    @abstractmethod
+    def expand(self) -> Self:
+        """
+        Expands this node creating X child leaf nodes, i.e., choose an action and apply it to the board
+        :return:
+        """
+        pass
+
+    @abstractmethod
+    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
+        """
+        pass
+
+    @abstractmethod
+    def backpropagate(self, score: float) -> None:
+        """
+        Backpropagates the results of the rollout
+        :param score:
+        :return:
+        """
+        pass
 
 
 class IMcts(ABC):
-
-    def __init__(self, board: chess.Board, strategy: IStrategy):
+    def __init__(self, board: chess.Board, strategy: IStrategy, seed: int | None):
         self.board = board
+        self.strategy = strategy
+        self.random_state = random.Random(seed)
 
     @abstractmethod
     def sample(self, runs: int = 1000) -> None:
@@ -28,7 +76,7 @@ class IMcts(ABC):
         pass
 
     @abstractmethod
-    def get_children(self) -> list['IMcts']:
+    def get_children(self) -> list[IMctsNode]:
         """
         Return the immediate children of the root node
         :return: list of immediate children of mcts root

src/chesspp/i_strategy.py

@@ -1,8 +1,11 @@
 from abc import ABC, abstractmethod
 
+import chess
+
+
 # TODO extend class
 class IStrategy(ABC):
 
     @abstractmethod
-    def pick_next_move(self, ):
+    def pick_next_move(self, board: chess.Board) -> chess.Move:
         pass

src/chesspp/random_strategy.py

@@ -0,0 +1,13 @@
+import chess
+import random
+from src.chesspp.i_strategy import IStrategy
+
+
+class RandomStrategy(IStrategy):
+    def __init__(self, random_state: random.Random):
+        self.random_state = random_state
+
+    def pick_next_move(self, board: chess.Board) -> chess.Move | None:
+        if len(list(board.legal_moves)) == 0:
+            return None
+        return self.random_state.choice(list(board.legal_moves))

src/chesspp/simulation.py

@@ -6,7 +6,7 @@ from typing import Tuple, List
 from enum import Enum
 from dataclasses import dataclass
 
-from chesspp.engine import Engine, Limit
+from src.chesspp.engine import Engine, Limit
 
 
 class Winner(Enum):

src/chesspp/util_gaussian.py

@@ -0,0 +1,83 @@
+import math
+
+import torch
+import torch.distributions as dist
+from torch import exp
+
+F1: dict[float, float] = {}
+F2: dict[float, float] = {}
+CDF: dict[float, float] = {}
+lookup_count = 0
+
+
+def max_gaussian_numeric(mu1, sigma1, mu2, sigma2) -> (float, float):
+    pass
+
+
+def max_gaussian(mu1, sigma1, mu2, sigma2) -> (float, float):
+    global lookup_count
+    global F1
+    global F2
+    global CDF
+
+    """
+    Returns the combined max gaussian of two Gaussians represented by mu1, sigma1, mu2, simga2
+    :param mu1: mu of the first Gaussian
+    :param sigma1: sigma of the first Gaussian
+    :param mu2: mu of the second Gaussian
+    :param sigma2: sigma of the second Gaussian
+    """
+    # we assume independence of the two gaussians
+    try:
+        #print(mu1, sigma1, mu2, sigma2)
+        normal = dist.Normal(0, 1)
+        sigma_m = math.sqrt(sigma1 ** 2 + sigma2 ** 2)
+        alpha = (mu1 - mu2) / sigma_m
+
+        if alpha in CDF:
+            cdf_alpha = CDF[alpha]
+            lookup_count += 1
+        else:
+            cdf_alpha = normal.cdf(torch.tensor(alpha)).item()
+            CDF[alpha] = cdf_alpha
+
+        pdf_alpha = exp(normal.log_prob(torch.tensor(alpha))).item()
+
+        if alpha in F1:
+            f1_alpha = F1[alpha]
+            lookup_count += 1
+        else:
+            f1_alpha = alpha * cdf_alpha + pdf_alpha
+            F1[alpha] = f1_alpha
+
+        if alpha in F2:
+            f2_alpha = F2[alpha]
+            lookup_count += 1
+        else:
+            f2_alpha = alpha ** 2 * cdf_alpha * (1 - cdf_alpha) + (
+                        1 - 2 * cdf_alpha) * alpha * pdf_alpha - pdf_alpha ** 2
+            F2[alpha] = f2_alpha
+
+        mu = mu2 + sigma_m * f1_alpha
+        #sigma_old = sigma2 ** 2 + (sigma1 ** 2 - sigma2 ** 2) * cdf_alpha + sigma_m ** 2 * f2_alpha
+        sigma = math.sqrt((mu1**2 + sigma1**2) * cdf_alpha + (mu2**2 + sigma2**2) * (1 - cdf_alpha) + (mu1 + mu2) * sigma_m * pdf_alpha - mu**2)
+
+        return mu, sigma
+    except ValueError:
+        print(mu1, sigma1, mu2, sigma2)
+        exit(1)
+
+
+def beta_mean(alpha, beta):
+    return alpha / (alpha + beta)
+
+
+def beta_std(alpha, beta):
+    try:
+        return math.sqrt((alpha * beta) / ((alpha * beta)**2 * (alpha + beta + 1)))
+    except ZeroDivisionError:
+        print(alpha, beta)
+
+
+def gaussian_ucb1(mu, sigma, N) -> float:
+    return mu + math.sqrt(2 * math.log(N) * sigma)