File size: 38,710 Bytes
6dc8c30 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 |
#!/usr/bin/env python3
"""
Advanced Magnus Model Backend Integration
Loads and serves the latest trained advanced Magnus model for FastAPI
"""
import sys
import pickle
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
from pathlib import Path
from typing import Dict, List, Tuple, Optional, Any
from collections import Counter
import chess
import chess.pgn
import yaml
import json
import warnings
warnings.filterwarnings("ignore")
# Add project root to path
project_root = Path(__file__).parent.parent.parent
sys.path.append(str(project_root))
class AdvancedChessFeatureExtractor:
"""Extract advanced chess features for better move prediction"""
def __init__(self):
self.piece_values = {
"p": 1,
"n": 3,
"b": 3,
"r": 5,
"q": 9,
"k": 0,
"P": 1,
"N": 3,
"B": 3,
"R": 5,
"Q": 9,
"K": 0,
}
def extract_features(self, position_data):
"""Extract comprehensive position features"""
features = []
# Basic piece counts and material balance
white_material = sum(
self.piece_values.get(p, 0) for p in str(position_data) if p.isupper()
)
black_material = sum(
self.piece_values.get(p, 0) for p in str(position_data) if p.islower()
)
material_balance = white_material - black_material
# Feature vector
features.extend(
[
white_material / 39.0, # Normalized material (max = Q+2R+2B+2N+8P)
black_material / 39.0,
material_balance / 39.0,
abs(material_balance) / 39.0, # Material imbalance magnitude
]
)
# Game phase estimation (opening/middlegame/endgame)
total_material = white_material + black_material
game_phase = total_material / 78.0 # 0 = endgame, 1 = opening
features.extend(
[
game_phase,
1 - game_phase, # Endgame indicator
min(game_phase * 2, 1), # Opening indicator
max(0, min((game_phase - 0.3) * 2, 1)), # Middlegame indicator
]
)
return np.array(features, dtype=np.float32)
class MultiHeadAttention(nn.Module):
"""Multi-head attention mechanism for position encoding"""
def __init__(self, d_model, num_heads):
super().__init__()
self.d_model = d_model
self.num_heads = num_heads
self.d_k = d_model // num_heads
self.W_q = nn.Linear(d_model, d_model)
self.W_k = nn.Linear(d_model, d_model)
self.W_v = nn.Linear(d_model, d_model)
self.W_o = nn.Linear(d_model, d_model)
def forward(self, x):
batch_size = x.size(0)
# Linear transformations
Q = self.W_q(x).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
K = self.W_k(x).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
V = self.W_v(x).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
# Attention
scores = torch.matmul(Q, K.transpose(-2, -1)) / np.sqrt(self.d_k)
attn = F.softmax(scores, dim=-1)
context = torch.matmul(attn, V)
# Concatenate heads
context = (
context.transpose(1, 2).contiguous().view(batch_size, -1, self.d_model)
)
output = self.W_o(context)
return output.mean(dim=1) # Global average pooling
class AdvancedMagnusModel(nn.Module):
"""Advanced Magnus model architecture matching the trained model"""
def __init__(self, vocab_size: int, feature_dim: int = 8):
super().__init__()
self.vocab_size = vocab_size
# Advanced board encoder with residual connections
self.board_encoder = nn.Sequential(
nn.Linear(768, 1024),
nn.BatchNorm1d(1024),
nn.ReLU(),
nn.Dropout(0.2),
nn.Linear(1024, 512),
nn.BatchNorm1d(512),
nn.ReLU(),
nn.Dropout(0.2),
nn.Linear(512, 256),
nn.BatchNorm1d(256),
nn.ReLU(),
)
# Multi-head attention mechanism for board understanding
self.board_attention = MultiHeadAttention(256, 8)
# Advanced feature encoder
self.feature_encoder = nn.Sequential(
nn.Linear(feature_dim, 64),
nn.BatchNorm1d(64),
nn.ReLU(),
nn.Dropout(0.1),
nn.Linear(64, 32),
nn.ReLU(),
)
# Combined feature processing
combined_dim = 256 + 32
self.feature_combiner = nn.Sequential(
nn.Linear(combined_dim, 512),
nn.BatchNorm1d(512),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(512, 256),
nn.BatchNorm1d(256),
nn.ReLU(),
nn.Dropout(0.2),
)
# Move prediction with multiple paths
self.move_predictor = nn.Sequential(
nn.Linear(256, 512),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(512, vocab_size),
)
# Evaluation head
self.eval_predictor = nn.Sequential(
nn.Linear(256, 128),
nn.ReLU(),
nn.Dropout(0.2),
nn.Linear(128, 64),
nn.ReLU(),
nn.Linear(64, 1),
nn.Tanh(),
)
def forward(self, position, features):
# Process board position
board_enc = self.board_encoder(position)
# Apply attention (reshape for attention if needed)
if len(board_enc.shape) == 2:
board_enc_reshaped = board_enc.unsqueeze(1) # Add sequence dimension
board_att = self.board_attention(board_enc_reshaped)
else:
board_att = self.board_attention(board_enc)
# Process additional features
feature_enc = self.feature_encoder(features)
# Combine features
combined = torch.cat([board_att, feature_enc], dim=1)
combined = self.feature_combiner(combined)
# Predictions
move_logits = self.move_predictor(combined)
eval_pred = self.eval_predictor(combined)
return move_logits, eval_pred
class AdvancedMagnusPredictor:
"""Advanced Magnus model predictor for FastAPI backend"""
def __init__(self, model_path: Optional[str] = None):
self.device = self._get_device()
self.model = None
self.move_to_idx = {}
self.idx_to_move = {}
self.vocab_size = 0
self.model_config = {}
self.feature_extractor = AdvancedChessFeatureExtractor()
# Default to latest MLflow model if no path provided
if model_path is None:
model_path = self._get_latest_mlflow_model()
if model_path and Path(model_path).exists():
self.load_model(model_path)
else:
print(f"⚠️ Model path not found: {model_path}")
def _get_device(self):
"""Get the best available device"""
if torch.backends.mps.is_available():
return torch.device("mps")
elif torch.cuda.is_available():
return torch.device("cuda")
else:
return torch.device("cpu")
def _get_latest_mlflow_model(self):
"""Get the latest MLflow model path"""
# Try multiple possible paths
possible_paths = [
project_root
/ "mlruns"
/ "427589957554434254"
/ "cbb3fccf10b64db5a8985add8bcac5ef"
/ "artifacts"
/ "model_artifacts",
Path(__file__).parent.parent
/ "mlruns"
/ "427589957554434254"
/ "cbb3fccf10b64db5a8985add8bcac5ef"
/ "artifacts"
/ "model_artifacts",
Path(
"/Users/levandalbashvili/Documents/GitHub/What-Would---DO/mlruns/427589957554434254/cbb3fccf10b64db5a8985add8bcac5ef/artifacts/model_artifacts"
),
]
for path in possible_paths:
if path.exists():
print(f"✅ Found model at: {path}")
return str(path)
print(f"❌ Model not found in any of these paths:")
for path in possible_paths:
print(f" - {path}")
return None
def load_model(self, model_path: str):
"""Load the trained model"""
try:
model_path = Path(model_path)
# Load configuration
config_file = model_path / "config.yaml"
if config_file.exists():
with open(config_file, "r") as f:
self.model_config = yaml.safe_load(f)
print(f"✅ Loaded model config: {config_file}")
# Load version info
version_file = model_path / "version.json"
if version_file.exists():
with open(version_file, "r") as f:
version_info = json.load(f)
print(f"✅ Model version: {version_info.get('model_id', 'unknown')}")
# Load the model state dict
model_file = model_path / "model.pth"
if not model_file.exists():
raise FileNotFoundError(f"Model file not found: {model_file}")
checkpoint = torch.load(model_file, map_location=self.device)
# Extract model components
if "model_state_dict" in checkpoint:
model_state = checkpoint["model_state_dict"]
self.move_to_idx = checkpoint.get("move_to_idx", {})
self.idx_to_move = checkpoint.get("idx_to_move", {})
self.vocab_size = checkpoint.get("vocab_size", len(self.move_to_idx))
else:
# Handle direct state dict
model_state = checkpoint
# Try to load vocabulary from config
vocab_size = self.model_config.get("vocab_size", 2000) # Default
self.vocab_size = vocab_size
# Check if vocabulary is missing and create it
if not self.move_to_idx or len(self.move_to_idx) == 0:
print(
"⚠️ Move vocabulary not found in checkpoint, creating from chess games"
)
# Get vocab size from model architecture
vocab_size = self.model_config.get("data", {}).get("vocab_size", 945)
self.vocab_size = vocab_size
self._create_vocabulary_from_games()
# Initialize model
vocab_size = self.model_config.get("data", {}).get("vocab_size", 945)
feature_dim = 8 # The saved model was trained with 8 features
self.vocab_size = vocab_size
self.model = AdvancedMagnusModel(self.vocab_size, feature_dim).to(
self.device
)
# Load state dict
self.model.load_state_dict(model_state)
self.model.eval()
total_params = sum(p.numel() for p in self.model.parameters())
print(f"✅ Advanced Magnus model loaded successfully!")
print(f" Device: {self.device}")
print(f" Parameters: {total_params:,}")
print(f" Vocabulary size: {self.vocab_size}")
print(f" Model path: {model_path}")
except Exception as e:
print(f"❌ Error loading model: {e}")
self.model = None
def _create_vocabulary_from_games(self):
"""Create vocabulary from actual chess games (like the training data)"""
print("🔧 Creating vocabulary from Magnus Carlsen games...")
moves = set()
# Try to load moves from available PGN files
pgn_paths = [
Path(__file__).parent / "data_processing" / "carlsen-games-quarter.pgn",
Path(__file__).parent / "data_processing" / "carlsen-games.pgn",
]
games_processed = 0
for pgn_path in pgn_paths:
if pgn_path.exists():
print(f"📖 Reading moves from {pgn_path.name}...")
try:
with open(pgn_path, "r") as f:
while True:
game = chess.pgn.read_game(f)
if game is None:
break
# Extract all moves from the game
board = game.board()
for move in game.mainline_moves():
moves.add(move.uci())
board.push(move)
games_processed += 1
if games_processed % 100 == 0:
print(
f" Processed {games_processed} games, {len(moves)} unique moves"
)
# Limit games processed to avoid too long loading
if games_processed >= 500:
break
if moves:
break # We have enough moves from this file
except Exception as e:
print(f" ⚠️ Error reading {pgn_path}: {e}")
continue
# If we couldn't read from PGN files, fall back to comprehensive UCI generation
if not moves:
print("📝 Falling back to comprehensive UCI move generation...")
moves = self._generate_comprehensive_uci_moves()
# Convert to sorted list and limit to vocab_size
moves_list = sorted(list(moves))
if len(moves_list) > self.vocab_size:
# Keep the most common/basic moves first
basic_moves = []
promotion_moves = []
other_moves = []
for move in moves_list:
if len(move) == 5 and move[4] in "qrbn": # Promotion
promotion_moves.append(move)
elif len(move) == 4: # Basic move
basic_moves.append(move)
else:
other_moves.append(move)
# Prioritize basic moves, then promotions, then others
moves_list = (basic_moves + promotion_moves + other_moves)[
: self.vocab_size
]
# Pad if needed
while len(moves_list) < self.vocab_size:
moves_list.append(f"null_move_{len(moves_list)}")
self.move_to_idx = {move: idx for idx, move in enumerate(moves_list)}
self.idx_to_move = {idx: move for move, idx in self.move_to_idx.items()}
print(
f"✅ Created vocabulary with {len(self.move_to_idx)} moves from {games_processed} games"
)
print(f" Sample moves: {moves_list[:10]}")
print(f" Last moves: {moves_list[-10:]}")
def _generate_comprehensive_uci_moves(self):
"""Generate comprehensive UCI moves as fallback"""
moves = set()
files = "abcdefgh"
ranks = "12345678"
# All possible square-to-square moves
for from_file in files:
for from_rank in ranks:
for to_file in files:
for to_rank in ranks:
from_sq = from_file + from_rank
to_sq = to_file + to_rank
if from_sq != to_sq:
moves.add(from_sq + to_sq)
# Pawn promotions
promotion_pieces = ["q", "r", "b", "n"]
for from_file in files:
for to_file in files:
# White promotions (rank 7 to 8)
for piece in promotion_pieces:
moves.add(f"{from_file}7{to_file}8{piece}")
# Black promotions (rank 2 to 1)
for piece in promotion_pieces:
moves.add(f"{from_file}2{to_file}1{piece}")
return moves
def board_to_tensor(self, board: chess.Board) -> torch.Tensor:
"""Convert chess board to tensor representation"""
# Create 768-dimensional board representation (8x8x12)
board_tensor = np.zeros((8, 8, 12), dtype=np.float32)
piece_map = {
chess.PAWN: 0,
chess.ROOK: 1,
chess.KNIGHT: 2,
chess.BISHOP: 3,
chess.QUEEN: 4,
chess.KING: 5,
}
for square in chess.SQUARES:
piece = board.piece_at(square)
if piece is not None:
rank, file = divmod(square, 8)
piece_type = piece_map[piece.piece_type]
color_offset = 0 if piece.color == chess.WHITE else 6
board_tensor[rank, file, piece_type + color_offset] = 1.0
return torch.FloatTensor(board_tensor.flatten())
def extract_features(self, board: chess.Board) -> torch.Tensor:
"""Extract advanced features from the board position"""
# Get FEN string for the feature extractor
fen = board.fen()
# Use the advanced feature extractor
features = self.feature_extractor.extract_features(fen)
return torch.FloatTensor(features)
def predict_moves(self, board: chess.Board, top_k: int = 5) -> List[Dict[str, Any]]:
"""Predict top-k moves prioritizing best moves with Magnus style flavor"""
if self.model is None:
return [{"move": "e2e4", "confidence": 0.5, "evaluation": 0.0}]
try:
# Get legal moves first
legal_moves = list(board.legal_moves)
if not legal_moves:
return []
# Strategy: Start with chess engine quality, then add Magnus flavor
predictions = []
# Get quick engine analysis for all legal moves
try:
import chess.engine
with chess.engine.SimpleEngine.popen_uci(
"/opt/homebrew/bin/stockfish"
) as engine:
# Analyze current position briefly
main_info = engine.analyse(board, chess.engine.Limit(time=0.1))
for legal_move in legal_moves:
# Make the move and evaluate
board_copy = board.copy()
board_copy.push(legal_move)
try:
# Quick evaluation
move_info = engine.analyse(
board_copy, chess.engine.Limit(time=0.03)
)
move_score = move_info.get(
"score",
chess.engine.PovScore(
chess.engine.Cp(0), board_copy.turn
),
)
# Calculate move quality based on engine
if move_score.is_mate():
if move_score.mate() > 0:
engine_quality = 0.95
else:
engine_quality = 0.05
else:
# Get centipawn evaluation from the side to move's perspective
cp_score = move_score.white().score(mate_score=10000)
if not board.turn: # Black to move
cp_score = -cp_score
# Convert to quality score (0.1 to 0.9)
engine_quality = max(
0.1, min(0.9, 0.5 + cp_score / 300)
)
except:
engine_quality = 0.5 # Neutral if evaluation fails
# Add Magnus style bonus (small influence)
magnus_bonus = 0.0
move_uci = legal_move.uci()
# Check if move is in Magnus's vocabulary
if move_uci in self.move_to_idx:
try:
with torch.no_grad():
position_tensor = (
self.board_to_tensor(board)
.unsqueeze(0)
.to(self.device)
)
features_tensor = (
self.extract_features(board)
.unsqueeze(0)
.to(self.device)
)
move_logits, _ = self.model(
position_tensor, features_tensor
)
move_probs = F.softmax(move_logits, dim=1)
idx = self.move_to_idx[move_uci]
magnus_style_score = float(
move_probs[0, idx].item()
)
magnus_bonus = (
magnus_style_score * 0.1
) # Only 10% influence
except:
magnus_bonus = 0.0
# Apply chess heuristics
heuristic_bonus = self._calculate_heuristic_bonus(
board, legal_move
)
# Final score: 80% engine quality, 10% Magnus style, 10% heuristics
final_confidence = (
0.8 * engine_quality
+ 0.1 * magnus_bonus
+ 0.1 * heuristic_bonus
)
predictions.append(
{
"move": move_uci,
"confidence": final_confidence,
"evaluation": (
cp_score if "cp_score" in locals() else 0.0
),
"engine_quality": engine_quality,
"magnus_bonus": magnus_bonus,
"heuristic_bonus": heuristic_bonus,
"is_legal": True,
"approach": "engine_primary",
}
)
except Exception as e:
print(f"Engine analysis failed, using heuristics only: {e}")
# Fallback to heuristics-based approach
for legal_move in legal_moves:
move_uci = legal_move.uci()
# Base quality from heuristics
heuristic_score = self._calculate_comprehensive_heuristic_score(
board, legal_move
)
# Small Magnus style influence
magnus_bonus = 0.0
if move_uci in self.move_to_idx:
try:
with torch.no_grad():
position_tensor = (
self.board_to_tensor(board)
.unsqueeze(0)
.to(self.device)
)
features_tensor = (
self.extract_features(board)
.unsqueeze(0)
.to(self.device)
)
move_logits, _ = self.model(
position_tensor, features_tensor
)
move_probs = F.softmax(move_logits, dim=1)
idx = self.move_to_idx[move_uci]
magnus_style_score = float(move_probs[0, idx].item())
magnus_bonus = (
magnus_style_score * 0.2
) # Slightly higher without engine
except:
magnus_bonus = 0.0
final_confidence = 0.8 * heuristic_score + 0.2 * magnus_bonus
predictions.append(
{
"move": move_uci,
"confidence": final_confidence,
"evaluation": 0.0,
"heuristic_score": heuristic_score,
"magnus_bonus": magnus_bonus,
"is_legal": True,
"approach": "heuristic_primary",
}
)
# Sort by confidence and return top-k
predictions.sort(key=lambda x: x["confidence"], reverse=True)
return predictions[:top_k]
except Exception as e:
print(f"❌ Prediction error: {e}")
# Return safe defaults with legal moves
legal_moves = list(board.legal_moves)
if legal_moves:
return [
{
"move": legal_moves[i % len(legal_moves)].uci(),
"confidence": max(0.15 - i * 0.02, 0.05),
"evaluation": 0.0,
"error": str(e),
"approach": "fallback",
}
for i in range(min(top_k, len(legal_moves)))
]
else:
return [
{
"move": "e2e4",
"confidence": 0.1,
"evaluation": 0.0,
"error": str(e),
}
]
def _calculate_heuristic_bonus(self, board: chess.Board, move: chess.Move) -> float:
"""Calculate a small heuristic bonus for the move"""
bonus = 0.0
piece = board.piece_at(move.from_square)
if piece:
# Center control
center_squares = [chess.E4, chess.E5, chess.D4, chess.D5]
if move.to_square in center_squares:
bonus += 0.05
# Piece development in opening
if (
piece.piece_type in [chess.KNIGHT, chess.BISHOP]
and board.fullmove_number <= 10
):
bonus += 0.03
# Captures
if board.is_capture(move):
captured = board.piece_at(move.to_square)
if captured:
piece_values = {
chess.PAWN: 1,
chess.KNIGHT: 3,
chess.BISHOP: 3,
chess.ROOK: 5,
chess.QUEEN: 9,
}
if piece_values.get(captured.piece_type, 0) >= piece_values.get(
piece.piece_type, 0
):
bonus += 0.04
# Checks
if board.gives_check(move):
bonus += 0.02
# Castling
if board.is_castling(move) and board.fullmove_number <= 15:
bonus += 0.06
return min(bonus, 0.15) # Cap the bonus
def _calculate_comprehensive_heuristic_score(
self, board: chess.Board, move: chess.Move
) -> float:
"""Calculate a comprehensive heuristic score for a move (used when engine is unavailable)"""
score = 0.5 # Base score
piece = board.piece_at(move.from_square)
if piece:
# Piece values and basic principles
piece_values = {
chess.PAWN: 1,
chess.KNIGHT: 3,
chess.BISHOP: 3,
chess.ROOK: 5,
chess.QUEEN: 9,
chess.KING: 0,
}
# Center control (major bonus)
center_squares = [chess.E4, chess.E5, chess.D4, chess.D5]
extended_center = [
chess.C3,
chess.C4,
chess.C5,
chess.C6,
chess.D3,
chess.D6,
chess.E3,
chess.E6,
chess.F3,
chess.F4,
chess.F5,
chess.F6,
]
if move.to_square in center_squares:
score += 0.15
elif move.to_square in extended_center:
score += 0.08
# Opening principles
if board.fullmove_number <= 10:
if piece.piece_type in [chess.KNIGHT, chess.BISHOP]:
score += 0.12 # Develop pieces
elif (
piece.piece_type == chess.PAWN and move.to_square in center_squares
):
score += 0.10 # Central pawns
# Captures (evaluate by material gain)
if board.is_capture(move):
captured = board.piece_at(move.to_square)
if captured:
material_gain = piece_values.get(
captured.piece_type, 0
) - piece_values.get(piece.piece_type, 0)
if material_gain >= 0:
score += min(0.2, 0.05 + material_gain * 0.02)
else:
score -= 0.1 # Bad capture
# Castling
if board.is_castling(move):
score += 0.15
# Checks (can be good or bad)
if board.gives_check(move):
score += 0.05 # Modest bonus for checks
# Avoid moving same piece twice in opening
if board.fullmove_number <= 8:
# Check if this piece has moved before
moves_history = list(board.move_stack)
piece_moved_before = any(
m.from_square == move.from_square for m in moves_history[-6:]
)
if piece_moved_before and piece.piece_type != chess.PAWN:
score -= 0.08
return max(0.1, min(0.9, score)) # Clamp between 0.1 and 0.9
def predict_moves_with_engine_guidance(
self,
board: chess.Board,
top_k: int = 5,
engine_path: str = "/opt/homebrew/bin/stockfish",
) -> List[Dict[str, Any]]:
"""Predict moves combining Magnus style with engine guidance for better quality"""
try:
import chess.engine
# Get Magnus-style predictions first
magnus_predictions = self.predict_moves(
board, top_k * 2
) # Get more candidates
# Analyze with engine
with chess.engine.SimpleEngine.popen_uci(engine_path) as engine:
# Get top engine moves
info = engine.analyse(
board, chess.engine.Limit(time=0.1), multipv=top_k
)
enhanced_predictions = []
for pred in magnus_predictions:
move_uci = pred["move"]
try:
move = chess.Move.from_uci(move_uci)
if move in board.legal_moves:
# Get engine evaluation of this move
board_copy = board.copy()
board_copy.push(move)
try:
eval_info = engine.analyse(
board_copy, chess.engine.Limit(time=0.05)
)
score = eval_info.get("score")
# Convert engine score to confidence adjustment
engine_confidence = 0.5 # Base
if score:
if score.is_mate():
if score.mate() > 0:
engine_confidence = 0.95
else:
engine_confidence = 0.05
else:
cp_score = score.white().score(mate_score=10000)
if board.turn == chess.BLACK:
cp_score = -cp_score
# Convert centipawn to confidence (better moves get higher confidence)
engine_confidence = max(
0.1, min(0.9, 0.5 + cp_score / 500)
)
# Blend Magnus style with engine evaluation
magnus_weight = 0.6 # 60% Magnus style
engine_weight = 0.4 # 40% engine evaluation
blended_confidence = (
magnus_weight * pred["confidence"]
+ engine_weight * engine_confidence
)
enhanced_predictions.append(
{
"move": move_uci,
"confidence": blended_confidence,
"evaluation": pred.get("evaluation", 0.0),
"magnus_confidence": pred["confidence"],
"engine_confidence": engine_confidence,
"style": "magnus_engine_hybrid",
}
)
except:
# If engine analysis fails, use original prediction
enhanced_predictions.append(pred)
except:
continue
# Sort by blended confidence
enhanced_predictions.sort(key=lambda x: x["confidence"], reverse=True)
return enhanced_predictions[:top_k]
except Exception as e:
print(f"Engine guidance failed, falling back to Magnus-only: {e}")
return self.predict_moves(board, top_k)
def _apply_chess_heuristics(
self, board: chess.Board, predictions: List[Dict[str, Any]]
) -> List[Dict[str, Any]]:
"""Apply chess heuristics to improve prediction quality"""
for pred in predictions:
move_uci = pred["move"]
try:
move = chess.Move.from_uci(move_uci)
confidence_boost = 0.0
# Boost confidence for good chess principles
piece = board.piece_at(move.from_square)
if piece:
# Center control (e4, e5, d4, d5)
center_squares = [chess.E4, chess.E5, chess.D4, chess.D5]
if move.to_square in center_squares:
confidence_boost += 0.02
# Piece development (knights and bishops)
if piece.piece_type in [chess.KNIGHT, chess.BISHOP]:
if board.fullmove_number <= 10: # Opening phase
confidence_boost += 0.03
# Captures are generally good
if board.is_capture(move):
captured_piece = board.piece_at(move.to_square)
if captured_piece:
# Higher value captures get more boost
piece_values = {
chess.PAWN: 1,
chess.KNIGHT: 3,
chess.BISHOP: 3,
chess.ROOK: 5,
chess.QUEEN: 9,
}
capture_value = piece_values.get(
captured_piece.piece_type, 0
)
attacking_value = piece_values.get(piece.piece_type, 0)
if capture_value >= attacking_value: # Good trades
confidence_boost += 0.04
# Checks can be good (but not always)
if board.gives_check(move):
confidence_boost += 0.02
# Castling is usually good in opening/middlegame
if board.is_castling(move) and board.fullmove_number <= 15:
confidence_boost += 0.05
# Apply the boost
pred["confidence"] = min(0.95, pred["confidence"] + confidence_boost)
pred["heuristic_boost"] = confidence_boost
except Exception as e:
# If we can't analyze the move, keep original confidence
pred["heuristic_boost"] = 0.0
return predictions
def is_loaded(self) -> bool:
"""Check if the model is successfully loaded"""
return self.model is not None
# Global instance for FastAPI
_magnus_predictor = None
def get_magnus_predictor() -> AdvancedMagnusPredictor:
"""Get the global Magnus predictor instance"""
global _magnus_predictor
if _magnus_predictor is None:
_magnus_predictor = AdvancedMagnusPredictor()
return _magnus_predictor
def test_predictor():
"""Test the predictor with a simple position"""
predictor = AdvancedMagnusPredictor()
if predictor.is_loaded():
board = chess.Board()
predictions = predictor.predict_moves(board, top_k=3)
print("🧪 Test Predictions:")
for i, pred in enumerate(predictions, 1):
print(f" {i}. {pred['move']} (confidence: {pred['confidence']:.3f})")
else:
print("❌ Predictor not loaded")
if __name__ == "__main__":
test_predictor()
|