def _parse_game(game: Game) -> List[str]:
positions: List[str] = []
board = game.board()
for move in game.mainline_moves():
board.push(move)
positions.append(board.fen())
return positions
def start_new_game(temporary=False):
base = 'tmp' if temporary else str(int(time.time()))
fn = os.path.extsep.join([base, 'pgn'])
path = games_path(fn)
_logger().info("Starting new game '{}'...".format(path))
game = Game()
game.headers["Date"] = time.strftime("%Y.%m.%d")
game.headers["White"] = "Human"
game.headers["Black"] = "Raspberry Turk"
_save_game(game, path)
enter_game(fn)
def mate_in(game: Game) -> Optional[TagKind]:
if not game.end().board().is_checkmate():
return None
moves_to_mate = int(len(list(game.mainline_moves())) / 2)
if moves_to_mate == 2:
return "mateIn2"
elif moves_to_mate == 2:
return "mateIn2"
elif moves_to_mate == 3:
return "mateIn3"
elif moves_to_mate == 4:
return "mateIn4"
elif moves_to_mate == 5:
return "mateIn5"
return "mateIn6+"
def get_puzzle(self, fen: str, prev_score: Score, move: str, current_score: Score, moves: str) -> None:
board = Board(fen)
game = Game.from_board(board)
node = game.add_main_variation(Move.from_uci(move))
current_eval = PovScore(current_score, not board.turn)
result = self.gen.analyze_position(node, prev_score, current_eval, tier=10)
self.assert_is_puzzle_with_moves(result, [Move.from_uci(x) for x in moves.split()])
def learn(self, iters=100, c=10):
"""
Run the Q-learning algorithm. Play greedy on the final iter
Args:
iters: int
amount of games to train
c: int
update the network every c games
Returns: pgn (str)
pgn string describing final game
"""
for k in range(iters):
if k % c == 0:
print("iter", k)
self.agent.fix_model()
greedy = True if k == iters - 1 else False
self.env.reset()
self.play_game(k, greedy=greedy)
pgn = Game.from_board(self.env.board)
reward_smooth = pd.DataFrame(self.reward_trace)
reward_smooth.rolling(window=10, min_periods=0).mean().plot()
return pgn
def analyze_game(server: Server, engine: SimpleEngine, game: Game,
args: argparse.Namespace) -> Optional[Puzzle]:
logger.debug("Analyzing game {}...".format(game.headers.get("Site")))
prev_score: Score = Cp(20)
for node in game.mainline():
current_eval = node.eval()
if not current_eval:
logger.debug("Skipping game without eval on ply {}".format(
node.ply()))
return None
result = analyze_position(server, engine, node, prev_score,
current_eval, args)
if isinstance(result, Puzzle):
return result
prev_score = -result
logger.debug("Found nothing from {}".format(game.headers.get("Site")))
return None
def read(doc) -> Puzzle:
board = Board(doc["fen"])
node = Game.from_board(board)
for uci in doc["moves"]:
move = Move.from_uci(uci)
node = node.add_main_variation(move)
return Puzzle(doc["_id"], node.game())
def not_puzzle(self, fen: str, prev_score: Score, move: str, current_score: Score) -> None:
board = Board(fen)
game = Game.from_board(board)
node = game.add_main_variation(Move.from_uci(move))
current_eval = PovScore(current_score, not board.turn)
result = self.gen.analyze_position( node, prev_score, current_eval, tier=10)
self.assertIsInstance(result, Score)
def read(doc) -> Puzzle:
board = Board(doc["fen"])
node: GameNode = Game.from_board(board)
for uci in (doc["line"].split(' ') if "line" in doc else doc["moves"]):
move = Move.from_uci(uci)
node = node.add_main_variation(move)
return Puzzle(doc["_id"], node.game(), int(doc["cp"]))
def clean(game: Game) -> None:
for node in game.mainline():
for variation in reversed(node.variations):
if not variation.is_mainline():
node.remove_variation(variation.move)
node.nags = set()
node.comment = ""
def analyze_game(self, game: Game, tier: int) -> Optional[Puzzle]:
logger.debug(f'Analyzing tier {tier} {game.headers.get("Site")}...')
prev_score: Score = Cp(20)
seen_epds: Set[str] = set()
board = game.board()
skip_until_irreversible = False
for node in game.mainline():
if skip_until_irreversible:
if board.is_irreversible(node.move):
skip_until_irreversible = False
seen_epds.clear()
else:
board.push(node.move)
continue
current_eval = node.eval()
if not current_eval:
logger.debug("Skipping game without eval on ply {}".format(
node.ply()))
return None
board.push(node.move)
epd = board.epd()
if epd in seen_epds:
skip_until_irreversible = True
continue
seen_epds.add(epd)
if board.castling_rights != maximum_castling_rights(board):
continue
result = self.analyze_position(node, prev_score, current_eval,
tier)
if isinstance(result, Puzzle):
return result
prev_score = -result
logger.debug("Found nothing from {}".format(game.headers.get("Site")))
return None
def export(puzzle, include_first_move=True):
fen = puzzle.last_pos.fen()
board = chess.Board(fen)
game = Game().from_board(board)
result = PgnExporter.determine_result_tag(board)
moves = puzzle.positions.move_list()
if include_first_move:
first_move = puzzle.last_move
else:
# simulate the move (blunder)
board.push(puzzle.last_move)
board.clear_stack()
# take resulting board and create new game
game = Game().from_board(board)
first_move = Move.from_uci(moves.pop(0))
# start the line
node = game.add_main_variation(first_move)
# add the rest of the moves
for m in moves:
node = node.add_variation(Move.from_uci(m))
# copy headers from the original game and override result tag
for h in puzzle.game.headers:
game.headers[h] = puzzle.game.headers[h]
game.headers['Result'] = result
return str(game)
def export(self, pgn_headers=None) -> Game:
""" pgn_headers - PGN headers to include in the exported PGN
"""
fen = self.puzzle.initial_board.fen()
board = chess.Board(fen)
game = Game().from_board(board)
game_node = game
game_node.comment = "score: %s -> %s" % (_score_to_str(
self.puzzle.initial_score), _score_to_str(self.puzzle.final_score))
comment = self._candidate_moves_annotations(self.puzzle.analyzed_moves)
for position in self.puzzle.positions:
game_node = game_node.add_variation(
chess.Move.from_uci(position.initial_move.uci()))
if comment:
game_node.comment = comment
comment = self._candidate_moves_annotations(
position.candidate_moves)
if pgn_headers:
for h in pgn_headers:
if h == "FEN":
continue
game.headers[h] = pgn_headers[h]
game.headers['PuzzleCategory'] = self.puzzle.category()
puzzle_winner = self.puzzle.winner()
if puzzle_winner:
game.headers['PuzzleWinner'] = puzzle_winner
game.headers['PuzzleEngine'] = AnalysisEngine.name()
game.headers['PuzzleMakerVersion'] = __version__
return game
def pgn_game_to_serializable_game(pgn_game: pgn.Game) -> SerializableGame:
current_index = 1
all_positions: List[SerializablePosition] = []
def parse_position(pgn_position: pgn.ChildNode,
is_mainline: bool) -> SerializablePosition:
nonlocal current_index
nonlocal all_positions
index = current_index
current_index += 1
next_pos_index = None
if pgn_position.variations:
next_pos_index = parse_position(pgn_position.variations[0],
True).index
variations_indexes = []
if len(pgn_position.variations) > 1:
variations_indexes = [
parse_position(p, False).index
for i, p in enumerate(pgn_position.variations) if i > 0
]
position = SerializablePosition(
index=index,
next_position_index=next_pos_index,
variations_indexes=variations_indexes,
nags=list(pgn_position.nags),
fen=pgn_position.board().fen(),
comment=pgn_position.comment,
commentBefore=pgn_position.starting_comment,
san=pgn_position.san(),
is_mainline=is_mainline,
move=Move(from_square=square_name(pgn_position.move.from_square),
to=square_name(pgn_position.move.to_square),
promotion=pgn_position.move.promotion))
all_positions.insert(0, position)
return position
first_pos = SerializablePosition(index=0,
fen=pgn_game.board().fen(),
is_mainline=True,
nags=[],
san="",
variations_indexes=[],
next_position_index=1)
all_positions.append(first_pos)
for i, position in enumerate(pgn_game.variations):
if i != 0:
first_pos.variations_indexes.append(current_index)
parse_position(position, i == 0)
headers: Dict[str, str] = {}
for key in pgn_game.headers.keys():
headers[key] = pgn_game.headers[key]
game = SerializableGame(
headers=headers,
comment=pgn_game.starting_comment,
positions=all_positions,
)
return game
def advanced_pawn(game: Game) -> bool:
for node in game.mainline():
if is_pawn_move(node):
rank = square_rank(node.move.to_square)
if rank > 5 and node.turn() == BLACK:
return True
if rank < 3 and node.turn() == WHITE:
return True
return False
def test_minimax(self):
board = chess.Board()
brain = MiniMax(self.umka)
while not board.is_game_over():
move = brain.run(board, DEPTH)
board.push(move)
print(str(chess.pgn.Game().from_board(board)).split("\n\n")[1])
game = Game().from_board(board)
print(game)
def configure_pgn(board):
"""Return a PGN representation of a completed chess game."""
pgn = Game.from_board(board)
pgn.headers['Event'] = 'Eternal Chess'
pgn.headers['Site'] = 'www.eternalchess.com'
pgn.headers['Date'] = datetime.now().strftime(DATE_FORMAT)
pgn.headers['Round'] = str(int(get_n_of_games()) + 1)
pgn.headers['White'] = 'Random'
pgn.headers['Black'] = 'Random'
return str(pgn)
def configure_pgn(board):
"""Return a PGN representation of a completed chess game."""
pgn = Game.from_board(board)
pgn.headers['Event'] = 'Eternal Chess'
pgn.headers['Site'] = 'www.eternalchess.com'
pgn.headers['Date'] = datetime.now().strftime(DATE_FORMAT)
pgn.headers['Round'] = str(int(get_n_of_games()) + 1)
pgn.headers['White'] = 'Random'
pgn.headers['Black'] = 'Random'
return str(pgn)
def play(brain):
board = chess.Board()
while not board.is_game_over():
move = brain.make_move(board, time_to_think=15 * 100)
print(move, brain.best_val, brain.best_move, brain.root_moves)
board.push(move)
pprint.pprint(str(board))
print(str(chess.pgn.Game().from_board(board)).split("\n\n")[1])
game = Game().from_board(board)
print(game)
def generate_pgn(self, game: Game):
"""
Gets PGN for given game
:param game: Current game
:type game: Game
:return: Bot message formatted with PGN
:rtype: str
"""
chess_game = ChessGame()
chess_game.headers["White"] = str(game.player1.name)
chess_game.headers["Black"] = str(game.player2.name)
chess_game.headers["Result"] = str(game.result)
last_node = chess_game
for move in game.board.move_stack:
last_node = last_node.add_variation(move)
result = f"```\n{str(chess_game)}\n```"
if game.id:
result += f'Game id: `{game.id}`'
return result
def classify(game: Game) -> int:
"""Attempt to classify the opening in a PGN game.
Args:
game (chess.pgn.Game): The game to classify.
Returns:
int: The ply count of the last book move.
If no opening is identified, returns 0.
"""
epds = list(
enumerate((node.board().epd() for node in game.mainline()), start=0))
with open("data/eco.tsv") as eco:
reader = csv.DictReader(eco, delimiter="\t")
for opening in reader:
for ply, epd in reversed(epds):
if epd == opening["fen"]:
game.headers["ECO"] = opening["eco"]
game.headers["Opening"] = opening["name"]
return ply
return 0
def parse_pgn_game(pgn: Game):
""" Parse a PGN game object and add it and its moves to DB """
print('Parsing PGN game...')
game_id = bson.ObjectId()
def is_int(var: str) -> bool:
""" Converter for ELO strings """
try:
int(var)
except ValueError:
return False
return True
db_game = {
'_id': game_id,
'date': pgn.headers['Date'] if 'Date' in pgn.headers else '???',
'white': pgn.headers['White'] if 'White' in pgn.headers else '???',
'white_elo': int(pgn.headers['WhiteElo']) if (
'WhiteElo' in pgn.headers
and is_int(pgn.headers['WhiteElo'])) else 0,
'black': pgn.headers['Black'] if 'Black' in pgn.headers else '???',
'black_elo': int(pgn.headers['BlackElo']) if (
'BlackElo' in pgn.headers
and is_int(pgn.headers['BlackElo'])) else 0,
'result': pgn.headers['Result'] if 'Result' in pgn.headers else '???'
}
if db.games.find_one({k: v for k, v in db_game.items() if k != '_id'}):
print('SKIPPED!')
return
print('Inserting:', db_game)
db.games.insert_one(db_game)
white_theory = db_game['white_elo'] >= 2500
black_theory = db_game['black_elo'] >= 2500
turn = True
b = chess.Board()
for i, move in enumerate(pgn.mainline_moves()):
san = b.san(move)
b.push(move)
db_move = {
'leads_to': b.fen(),
'uci': move.uci(),
'san': san,
'score_diff': None
}
b.pop() # ugly, but we need the original position
if (turn and white_theory) or (not turn and black_theory):
idatabase.insert_board(b.fen(), [db_move], [], game_id if i > 5 else None)
else:
idatabase.insert_board(b.fen(), [], [db_move], game_id if i > 5 else None)
b.push(move) # so now we put it back
turn = not turn
if i > database.MAX_DEPTH:
break
def reportpgn(self):
rootboard = self.getrootboard()
game = Game()
game.setup(rootboard)
game.comment = self.rootnode().comment()
game = self.addmovesrecursive(self.rootnode(), game)
exporter = StringExporter(headers=True, variations=True, comments=True)
pgn = game.accept(exporter)
return pgn
def analyze_game(engine: SimpleEngine,
game: Game) -> Optional[Tuple[GameNode, List[Move], Kind]]:
"""
Evaluate the moves in a game looking for puzzles
"""
game_url = game.headers.get("Site", "?")
logger.debug("Analyzing game {}...".format(game_url))
prev_score: Score = Cp(0)
for node in game.mainline():
ev = node.eval()
if not ev:
logger.debug("Skipping game without eval on move {}".format(
node.board().fullmove_number))
return None
winner = node.board().turn
score = ev.pov(winner)
# was the opponent winning until their last move
if prev_score > Cp(-300) or not is_down_in_material(
node.board(), winner):
pass
elif mate_soon < score < Mate(1):
logger.info("Mate found on {}#{}. Probing...".format(
game_url, ply_of(node.board())))
solution = cook_mate(engine, node, winner)
if solution is not None:
return node, solution, Kind("mate")
elif score > juicy_advantage:
logger.info("Advantage found on {}#{}. Probing...".format(
game_url, ply_of(node.board())))
solution = cook_advantage(engine, node, winner)
if solution is not None:
return node, solution, Kind("mate")
else:
print(score)
prev_score = score
return None
def main() -> None:
sys.setrecursionlimit(10000) # else node.deepcopy() sometimes fails?
parser = argparse.ArgumentParser(prog='tagger.py', description='automatically tags lichess puzzles')
parser.add_argument("--verbose", "-v", help="increase verbosity", action="count")
args = parser.parse_args()
if args.verbose == 1:
logger.setLevel(logging.DEBUG)
mongo = pymongo.MongoClient()
db = mongo['puzzler']
puzzle_coll = db['puzzle2']
tag_coll = db['tag']
for puzzle in puzzle_coll.find():
# prev = tag_coll.find_one({"_id":puzzle._id})
board = Board(puzzle["fen"])
node = Game.from_board(board)
for uci in puzzle["moves"]:
move = Move.from_uci(uci)
node = node.add_main_variation(move)
puzzle = Puzzle(puzzle["_id"], node.game())
tags = cook.cook(puzzle)
for tag in tags:
tag_coll.update_one({"_id":puzzle.id},{"$addToSet":{tag: "lichess"}}, upsert = True)
def learn(self, iters=100, c=10):
"""
Run the Q-learning algorithm. Play greedy on the final iter
Args:
iters: int
amount of games to train
c: int
update the network every c games
Returns: pgn (str)
pgn string describing final game
"""
for k in range(iters):
self.env.reset()
states, actions, rewards, action_spaces = self.play_game(k)
self.reinforce_agent(states, actions, rewards, action_spaces)
pgn = Game.from_board(self.env.board)
reward_smooth = pd.DataFrame(self.reward_trace)
reward_smooth.rolling(window=10, min_periods=0).mean().plot()
return pgn
import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import matplotlib.pyplot as plt
import os
from RLC.real_chess import agent, environment, learn, tree
import chess
from chess.pgn import Game
opponent = agent.GreedyAgent()
env = environment.Board(opponent, FEN=None)
player = agent.Agent(lr=0.001, network='big')
player.fix_model()
learner = learn.TD_search(env, player, gamma=0.8, search_time=1.5)
node = tree.Node(learner.env.board, gamma=learner.gamma)
w_before = learner.agent.model.get_weights()
n_iters = 105
print(opponent.predict(np.expand_dims(env.layer_board, axis=0)))
learner.search_time = 60
learner.play_game(n_iters)
pgn = Game.from_board(learner.env.board)
with open("rlc_pgn","w") as log:
log.write(str(pgn))
