ChessGPT / extract_moves.py
jcandane's picture
Upload extract_moves.py with huggingface_hub
5842de6 verified
Raw
History Blame Contribute Delete
7.68 kB
import chess.pgn
import chess
import numpy as np
from tqdm.notebook import tqdm
def extract_moves_from_pgn(pgn_path:str, max_games=None) -> np.ndarray:
all_moves = []
def is_improper_piece(piece_type):
return piece_type in {chess.PAWN, chess.ROOK, chess.KING}
def make_starting_position_move(board):
setup = ['X', '->']
parts = []
for square in chess.SQUARES:
piece = board.piece_at(square)
if piece:
color = 'w' if piece.color == chess.WHITE else 'b'
symbol = piece.symbol().upper()
square_name = chess.square_name(square)
if is_improper_piece(piece.piece_type):
piece_id = f"{symbol.lower()}{color}"
else:
piece_id = f"{symbol}{color}"
parts.extend([piece_id, square_name])
for i in range(0, len(parts), 2):
if i > 0:
setup.append('&')
setup.extend(parts[i:i+2])
return setup
def get_piece_id(piece, color, square, moved_from):
symbol = piece.symbol().upper()
if square not in moved_from and is_improper_piece(piece.piece_type):
return f"{symbol.lower()}{color}"
return f"{symbol}{color}"
with open(pgn_path) as pgn:
progress = tqdm(desc="Processing Games", unit="games", dynamic_ncols=True, leave=True)
game_count = 0
while True:
game = chess.pgn.read_game(pgn)
if game is None or (max_games is not None and game_count >= max_games):
break
game_count += 1
fen = game.headers.get("FEN", chess.STARTING_FEN)
board = chess.Board(fen)
moved_from = set()
game_moves = []
game_moves.append(['<S>'])
game_moves.append(make_starting_position_move(board))
for move in game.mainline_moves():
from_sq = move.from_square
to_sq = move.to_square
from_sq_name = chess.square_name(from_sq)
to_sq_name = chess.square_name(to_sq)
moving_piece = board.piece_at(from_sq)
color = 'w' if moving_piece.color == chess.WHITE else 'b'
piece_id_before = get_piece_id(moving_piece, color, from_sq_name, moved_from)
# fallback castling detection if castling_rook_square() is not available
if board.is_castling(move):
king_move = [piece_id_before, from_sq_name, '->', piece_id_before, to_sq_name]
fen_before = board.fen()
board_before = chess.Board(fen_before)
rooks_before = set(sq for sq in board_before.pieces(chess.ROOK, moving_piece.color))
board.push(move)
rooks_after = set(sq for sq in board.pieces(chess.ROOK, moving_piece.color))
rook_from_sq = list(rooks_before - rooks_after)
rook_to_sq = list(rooks_after - rooks_before)
if not rook_from_sq and not rook_to_sq:
# ✅ Rook didn't move (legal in Chess960), no second action needed
moved_from.add(from_sq_name)
game_moves.append(king_move)
continue
if len(rook_from_sq) == 1 and len(rook_to_sq) == 1:
rook_from = chess.square_name(rook_from_sq[0])
rook_to = chess.square_name(rook_to_sq[0])
rook_piece = board.piece_at(rook_to_sq[0])
rook_id = get_piece_id(rook_piece, color, rook_from, moved_from)
rook_move = [rook_id, rook_from, '->', rook_id, rook_to]
action = king_move + ['&'] + rook_move
moved_from.update([from_sq_name, rook_from])
game_moves.append(action)
continue
# Still couldn't determine rook move
print(f"\n⚠️ Warning: Could not determine rook move during castling at {from_sq_name}->{to_sq_name}")
print("→ PGN headers:")
for key, val in game.headers.items():
print(f"[{key} \"{val}\"]")
print("→ Current FEN before move:")
print(fen_before)
print("→ Move UCI:", move.uci())
print("→ Rooks before:", [chess.square_name(sq) for sq in rooks_before])
print("→ Rooks after:", [chess.square_name(sq) for sq in rooks_after])
print("→ Difference:")
print(" Disappeared:", [chess.square_name(sq) for sq in rooks_before - rooks_after])
print(" Appeared:", [chess.square_name(sq) for sq in rooks_after - rooks_before])
print()
game_moves.append(king_move)
continue
captured_piece = board.piece_at(to_sq)
board.push(move)
new_piece = board.piece_at(to_sq)
piece_id_after = f"{new_piece.symbol().upper()}{color}"
action = [piece_id_before, from_sq_name, '->', piece_id_after, to_sq_name]
moved_from.add(from_sq_name)
if board.is_en_passant(move):
capture_sq = chess.square_name(to_sq + (8 if color == 'b' else -8))
action += ['&', f"P{'w' if color == 'b' else 'b'}", capture_sq, '->', 'X']
elif captured_piece:
captured_id = f"{captured_piece.symbol().upper()}{'w' if captured_piece.color == chess.WHITE else 'b'}"
action += ['&', captured_id, to_sq_name, '->', 'X']
game_moves.append(action)
result = game.headers.get("Result", "*")
game_moves.append([result])
game_moves.append(['<E>'])
all_moves.extend(game_moves)
progress.update(1)
progress.close()
# 🔽 Final flattening step to yield List[str]
flattened = []
for move in all_moves:
flattened.extend(move)
flattened.append(',') # comma delimiter
if flattened:
flattened.pop() # Remove last comma
# Step 2: Token dictionary
NORMALIZE_MAP = {
'½-½': '1/2-1/2',
'1-0': '1-0',
'0-1': '0-1',
# Just in case someone uses space around dashes
'1 - 0': '1-0',
'0 - 1': '0-1',
'1/2 - 1/2': '1/2-1/2',
'½ - ½': '1/2-1/2',
}
tokens = ['<S>', '<E>', '->', '&', ',', 'X']
files = 'abcdefgh'
ranks = '12345678'
squares = [f + r for r in ranks for f in files]
tokens += squares
pieces = ['P', 'p', 'R', 'r', 'N', 'B', 'Q', 'K', 'k']
colors = ['w', 'b']
tokens += [p + c for p in pieces for c in colors]
annotations = ['!!', '!', '!?', '?!', '?', '??', '+', '#', '+−', '−+']
results = ['1-0', '0-1', '1/2-1/2']
tokens += annotations + results
str_2_int = {token: idx for idx, token in enumerate(tokens)}
# Step 3: Convert to integer list
int_tokens = []
for token in flattened:
token = NORMALIZE_MAP.get(token, token) # normalize if needed
if token in str_2_int:
int_tokens.append(str_2_int[token])
else:
raise ValueError(f"Unknown token: {token}")
# Step 4: Convert to numpy array
arr = np.array(int_tokens, dtype=np.int8)
return arr