def best_play(board: chess.Board, player, profondeur=5):
"""
:param profondeur: profondeur de l'algorithme
:param board: chess.Board
:param player: boolean
:return: chess.Move
"""
results = []
id = 0
moves = []
pool = mp.Pool(processes=16)
for move in board.legal_moves:
moves.append(move)
clone = board.copy(stack=False)
clone.push(move)
if player:
results.append((pool.apply(_min, args=(clone, profondeur)), id))
else:
results.append((pool.apply(_max, args=(clone, profondeur)), id))
id += 1
if player:
m = max(results)
else:
m = min(results)
return moves[m[1]]
def without_opponent_pieces(board: chess.Board) -> chess.Board:
"""Returns a copy of `board` with the opponent's pieces removed."""
b = board.copy()
b.ep_square = None
for piece_type in chess.PIECE_TYPES:
for sq in b.pieces(piece_type, not board.turn):
b.remove_piece_at(sq)
return b
def __init__(self,
board: chess.Board = None,
main_UI=None,
*args,
**kwargs):
super().__init__(*args, **kwargs)
self.board = board.copy()
self.main_UI = main_UI
def expand_state_chess(state: chess.Board) -> Dict[str, chess.Board]:
children = {}
for move in state.legal_moves:
# TODO: can we push/pop to aviod copies?
board = state.copy()
board.push(move)
children[move.uci()] = board
return children
def get_defending_pieces(board: chess.Board, defending_color: chess.Color, square: chess.Square) -> [chess.PieceType]:
cloned_board = board.copy()
defending_pieces = get_attacking_pieces(cloned_board, defending_color, square)
del cloned_board
return defending_pieces
def make_move(self, board: chess.Board, is_white: Boolean):
t_start = time.time()
self.reset()
#root = ChessNode(board, 0)
#print('The current system have {} CPUs of which {} are usable'.format(mp.cpu_count(),len(os.sched_getaffinity(0))))
max_num_processes = 3
num_legal_moves = board.legal_moves.count()
legal_moves_per_process = math.ceil(num_legal_moves /
max_num_processes)
processes = []
root_parts: List[ChessNode] = []
out_queue = mp.Queue()
print("Starting make_move")
# When there are few moves left, they might be all distributed among less than all processes.
# Eg. with max_num_processes==3 and num_legal_moves==4, we get legal_moves_per_process==2
# so we'll only use 2 processes.
num_processes = math.ceil(num_legal_moves / legal_moves_per_process)
for i in range(0, num_processes):
# Only create node if there are legal moves left to use
if num_legal_moves > legal_moves_per_process * i:
root_parts.append(
ChessNode(board.copy(), 0, i * legal_moves_per_process,
legal_moves_per_process))
processes.append(
mp.Process(target=self.run_it,
args=(board.copy(), is_white, root_parts[-1],
t_start, i + 1, out_queue)))
processes[-1].start()
for p in processes:
p.join()
max_val, max_val_move = -math.inf, None
for i in range(0, num_processes):
cur_max_move, cur_max_val = out_queue.get()
print(cur_max_move, cur_max_val)
if cur_max_val > max_val:
max_val = cur_max_val
max_val_move = cur_max_move
print("Done with all the jobs")
move = max_val_move
print(
'Chose move {} with value/visits={}. Node count: {} Max depth: {}'.
format(max_val_move, max_val, self.no_nodes, self.max_level))
return move
def quiesce(fen, alpha, beta, flag, options):
# flag check
if not flag.is_set():
return 0, 0
# heuristic
if options.heuristic == 'NeuralNetwork':
stand_pat = heuristic.nn_heuristic(fen, options, options.model)
elif options.heuristic == 'Random':
stand_pat = heuristic.random_heuristic()
else:
stand_pat = heuristic.heuristic(fen, options)
nodes = 1
if (stand_pat >= beta):
return beta, nodes
board = Board(fen)
if len(board.piece_map()) > 8:
delta = True
else:
delta = False
if delta:
# full delta pruning
if (stand_pat < alpha - 1000):
return alpha, nodes
if (stand_pat > alpha):
alpha = stand_pat
# expansion and search
legal = board.legal_moves
for move in legal:
board.push(move)
new = board.copy()
board.pop()
if board.is_capture(move) or new.is_check():
# delta pruning
if delta and board.is_capture(move):
value = value_captured_piece(
board.piece_type_at(move.to_square)) + 200
if (stand_pat + value < alpha):
continue
score, count = quiesce(new.fen(), -beta, -alpha, flag, options)
score = -score
nodes += count
if (score >= beta):
return beta, nodes
if (score > alpha):
alpha = score
return alpha, nodes
def alphabeta(board: chess.Board, depth: int, alpha: chess.Board, beta: chess.Board, player: chess.Color):
if depth == 0:
return eval(board, alpha, beta, player)
if player == chess.WHITE:
for move in board.generate_legal_moves():
new_board = board.copy()
new_board.push(move)
alpha = alphabeta(new_board, depth-1, alpha, beta, chess.BLACK)
if model.predict_mlp(utils.bitify(beta.fen()), utils.bitify(alpha.fen()))[0] == 1:
break
return alpha
else:
for move in board.generate_legal_moves():
new_board = board.copy()
new_board.push(move)
beta = alphabeta(new_board, depth-1, alpha, beta, chess.WHITE)
if model.predict_mlp(utils.bitify(beta.fen()), utils.bitify(alpha.fen()))[0] == 1:
break
return beta
def get_move(self, board: chess.Board) -> chess.Move:
# TODO: Implement Minimax to explore more than a single move ahead.
possible_moves = []
board = board.copy()
for move in board.legal_moves:
board.push(move)
possible_moves.append((self.evaluate(board), move))
board.pop()
sorted_moves = sorted(possible_moves, key=lambda x: x[0], reverse=board.turn)
for move in sorted_moves[:3]:
print(f'Move: {move[1]}, evaluated at: {move[0]}')
return sorted_moves[0][1]
def is_defended(board: Board, piece: Piece, square: Square) -> bool:
if board.attackers(piece.color, square):
return True
# ray defense https://lichess.org/editor/6k1/3q1pbp/2b1p1p1/1BPp4/rp1PnP2/4PRNP/4Q1P1/4B1K1_w_-_-_0_1
for attacker in board.attackers(not piece.color, square):
if board.piece_at(attacker).piece_type in ray_piece_types:
bc = board.copy(stack = False)
bc.remove_piece_at(attacker)
if bc.attackers(piece.color, square):
return True
return False
def generate_ply_info_list_for_game(game):
"Returns a dict which contains the list of tasks to be run"
board = Board()
ply_no = 0
game_tasks = []
for ply in game.mainline_moves():
board.push(ply)
ply_info = {"ply_no": ply_no, "board": board.copy()}
game_tasks.append(ply_info)
ply_no = ply_no + 1
return {"game_tasks": game_tasks, "ply_count": ply_no}
def rollout_move(board: Board,
move: str,
simulations: int = 10) -> Tuple[str, Number[Any]]:
rollouts_scores = list()
for _ in range(simulations):
child_node = board.copy()
child_node.push(move)
while not child_node.is_game_over():
rand_move = get_random_move(child_node)
child_node.push_san(rand_move)
result = parse_result(child_node.result())
rollouts_scores.append(result)
return (move, np.mean(rollouts_scores))
def create_image(self, board: chess.Board):
fen = board.copy().fen().split(' ')[0].split('/')
obervations = np.empty(shape=(8, 8), dtype=str) # make empty container
for i, row in enumerate(fen):
j = 0
for square in row:
if square.isnumeric():
for _ in range(j, int(square) + j):
obervations[i][j] = '.'
j += 1
continue
obervations[i][j] = square
j += 1
return obervations.tolist()
def erase_random_piece(board: chess.Board):
removable_piece_squares = []
for square, piece in board.piece_map().items():
if piece.piece_type != chess.KING:
removable_piece_squares.append(square)
for _ in range(0, 10):
square = random.choice(removable_piece_squares)
new_board = board.copy()
new_board.remove_piece_at(square)
if new_board.is_valid():
return new_board
return None
def mcts_player(board: chess.Board, evaluation: EvaluationFunction, iterations: int, maxdepth: int) -> chess.Move:
for move_choice in board.legal_moves:
copy = board.copy()
copy.push(move_choice)
if copy.is_game_over():
board.push(move_choice)
return move_choice
root = MCTSNode(
ChessState(board, board.turn, maxdepth),
evaluation,
move_to_get_here=None,
)
move = root.best_action(iterations)
# move = UCT(board, itermax, depthmax, evaluation)
board.push(move)
return move
def get_syzygy(self, board: chess.Board) -> tuple[int, chess.Move]:
"""
Get a move from Syzygy tablebases.
:param chess.Board board: Board to get best move and evaluation.
:return: The evaluation from Syzygy tablebases and the best move.
:rtype: tuple[int, chess.Move]
"""
# Generate DTZ list
wdl: dict[int, dict[int, chess.Move]] = {
2: {},
1: {},
0: {},
-1: {},
-2: {}
}
for move in board.legal_moves:
test_board: chess.Board = board.copy()
test_board.push(move)
wdl[self.syzygy_tb.probe_wdl(test_board)][self.syzygy_tb.probe_dtz(
test_board)] = move
# Get best WDL
best_wdl: int = -2
while wdl[best_wdl] == {}:
best_wdl += 1
# Get best move
best_dtz: int = 0
best_dtz = max(wdl[best_wdl].items(), key=lambda key: key[0])[0]
# Evaluation
print(wdl)
syzygy_evaluation: int = 0
if best_wdl < 0:
if board.turn:
syzygy_evaluation = 10000
else:
syzygy_evaluation = -10000
elif best_wdl == 0:
syzygy_evaluation = 0
else:
if board.turn:
syzygy_evaluation = -10000
else:
syzygy_evaluation = 10000
# Return
return syzygy_evaluation, wdl[best_wdl][best_dtz]
def _evaluate_move(turn: chess.Color, board: chess.Board,
move: chess.Move) -> float:
# sum up the total weighted pieces for each player and return a score
board = board.copy()
board.push(move)
total = 0
if board.is_game_over():
return -1000 if board.turn == turn else 1000
for p in (chess.PAWN, chess.KNIGHT, chess.BISHOP, chess.ROOK, chess.QUEEN,
chess.KING):
for c in board.pieces(p, turn):
total += piece[p] + pst[p][c]
return total
def run(col, depth):
refer = get_ref()
if col == True:
old_bo = online_minmax(col, Board(), refer, depth)
else:
old_bo = Board()
while True:
see_bo = read2(ref=refer, flip=col)
print(old_bo, '\n_______________')
print(see_bo, '\n_______________')
if str(old_bo) != str(see_bo):
new_bo = old_bo.copy()
new_bo.push(find_move(old_bo, see_bo))
move_bo = online_minmax(col, new_bo, refer, depth)
read_bo = read2(ref=refer, flip=col)
if str(move_bo) == str(read_bo):
old_bo = move_bo
def predict_best_move(self, curr_board: chess.Board) -> list:
turn = curr_board.turn
eval_moves = []
mate_moves = []
opponent_mates = []
for move in curr_board.legal_moves:
new_board = curr_board.copy()
new_board.push(move)
# analyze position and remember the best one, analyze is not yet implemented
prediction = self.analyze(new_board)
evaluation = prediction[0].item()
moves_til_mate = prediction[1].item()
is_mate = prediction[2].item()
if turn:
if is_mate > 0.5 and moves_til_mate >= 0:
mate_moves.append((curr_board.san(move), moves_til_mate))
elif is_mate > 0.5:
opponent_mates.append((curr_board.san(move), moves_til_mate))
else:
eval_moves.append((curr_board.san(move), evaluation))
else:
if is_mate > 0.5 and moves_til_mate <= 0:
mate_moves.append((curr_board.san(move), moves_til_mate))
elif is_mate > 0.5:
opponent_mates.append((curr_board.san(move), moves_til_mate))
else:
eval_moves.append((curr_board.san(move), evaluation))
if turn:
eval_moves.sort(key=lambda x: x[1], reverse=True)
mate_moves.sort(key=lambda x: x[1])
opponent_mates.sort(key=lambda x: x[1])
else:
eval_moves.sort(key=lambda x: x[1])
mate_moves.sort(key=lambda x: x[1], reverse=True)
opponent_mates.sort(key=lambda x: x[1], reverse=True)
return eval_moves, mate_moves, opponent_mates
def lookahead1_move(original: Board, verbose: bool = False) -> Move:
"""Looks ahead one move
"""
best_score = None
best_moves = None
def is_better(reference_score, new_score) -> bool:
if reference_score is None:
return True
if original.turn == chess.WHITE:
return new_score > reference_score
else:
return reference_score > new_score
my_moves = list(original.legal_moves)
for my_move in my_moves:
b = original.copy()
b.push(my_move)
if b.is_game_over():
score = score_board(b)
else:
their_move = pick_move(b)
b.push(their_move)
score = score_board(b)
if verbose:
print(f"{my_move} then {their_move} gives {score}.")
if is_better(best_score, score):
best_score = score
best_moves = [my_move]
elif score == best_score:
best_moves.append(my_move)
if verbose:
print(f"\nBest moves are {best_moves} scoring {best_score}")
if best_moves:
return random.choice(best_moves)
else:
print("No valid moves to pick from")
return None
def getGameEnded(self, board: chess.Board, player):
"""
Input:
board: current board
player: current player (1 or -1)
Returns:
r: 0 if game has not ended. 1 if player won, -1 if player lost,
small non-zero value for draw.
"""
new_board = board.copy()
if new_board.is_game_over(claim_draw=True):
if new_board.result() == "1-0":
return 1
elif new_board.result() == "1/2-1/2":
return -0.5
else:
return -1
return 0
def search(self, board: chess.Board):
"""
Recursevly search a tree using UCT and the NNet to intellegently select the proper tree to search.
"""
if board.is_game_over():
result = board.result()
return -end_states[result]
board_hash = polyglot.zobrist_hash(board)
if board_hash not in self.Policy_vectors:
# Get nnet prediction of the current board for use in the
# Determine which branch to explore next.
# If black to play, mirror the board, unmirror at end. NNet always sees current player as white.
# Make a copy of the board as to preserve our move stack.
temp = None
if not board.turn:
# If black to play, flip and mirror board.
temp = board.transform(chess.flip_vertical)
temp.apply_mirror()
else:
temp = board.copy()
cannonical = adapter.get_cannonical(temp)
policy_vector, nnet_value = self.network.predict(cannonical)
# Mask out invalid moves
valids = adapter.moves_to_policy_mask(list(board.legal_moves))
policy_vector *= valids
# Normalize vector, add valid moves if needed.
if np.sum(policy_vector > 0):
policy_vector /= np.linalg.norm(policy_vector)
self.Policy_vectors[board_hash] = policy_vector
else:
print(
"All valid moves were masked. Adding valids. Warning if lots of these messages."
)
policy_vector += valids
policy_vector /= np.linalg.norm(policy_vector)
self.N_vals[board_hash] = 0
del temp
# Return the esimate until we actually reach the end.
return -nnet_value
# Iterate over legal moves and get the probability of making that move, according to the nnet.
action_heuristic_dict = {}
curr_move_policy = self.Policy_vectors[board_hash]
for move in list(board.legal_moves):
move_prob = adapter.get_move_prob(curr_move_policy, move)
action_heuristic_dict[move] = self.get_ucb(board_hash, move,
move_prob * 2.0)
# Pick move with max value, make it bigger
move = max(action_heuristic_dict, key=action_heuristic_dict.get)
# action_heuristic_dict[max_move] *= 50.0
# # Normalize
# values = np.array(list(action_heuristic_dict.values()))
# values += np.abs(values.min())
# values /= np.linalg.norm(values)
# p = self.fix_p(values)
# move = np.random.choice(list(action_heuristic_dict.keys()), p=p)
board.push(move)
value = self.search(board)
board.pop()
# We've done our search, now we're back-propigating values for the next search.
self.update_values(board_hash, move, value)
self.N_vals[board_hash] += 1
return -value
def make_child(self, board: chess.Board, move: str) -> chess.Board:
child = board.copy()
child.push(move)
return child
def get_mobility(board: Board, mobility_coeff: float = 1) -> float:
player_legal_moves = len(list(board.legal_moves))
_board = board.copy()
_board.turn = not board.turn
oponent_legal_moves = len(list(_board.legal_moves))
return mobility_coeff * (player_legal_moves - oponent_legal_moves)
def successors(board: chess.Board) -> Iterator[chess.Board]:
for move in board.legal_moves:
board_after = board.copy(stack=False)
board_after.push(move)
yield board_after
def search(self, board: chess.Board, depth, side, alpha=-2, beta=2):
"""
Performs a minimax search with alpha-beta pruning.
Alpha, beta are set to -2, 2 respectively; because the suppport of the CNN is [0,1].
Args:
`board`: Chess board state at current node.
`depth`: Search depth.
`side`: White (True), Black (False). White is the maximising player.
`alpha`: Current minimum score White is assured of.
`beta`: Current maximum score Black is assured of.
Returns:
(evaluation, uci-encoded move that `side` should play)
"""
if depth == 0:
input = szr.Serializer(board).serialize()
return (self.model.predict(np.expand_dims(input, 0)), board.pop())
if board.is_game_over():
res = board.result()
if res == '0-1':
return 0
elif res == '1-0':
return 1
else:
return 0.5
if side:
max_score = -2
best_move = ''
for move in board.legal_moves:
new_board = board.copy()
new_board.push(move)
score = self.search(new_board, depth - 1, alpha, beta,
False)[0]
if score > max_score:
max_score = score
best_move = move
alpha = max(alpha, score)
if alpha >= beta:
# black has a better option available earlier on in the tree.
break
return (max_score, best_move)
else:
min_score = 2
best_move = ''
for move in board.legal_moves:
new_board = board.copy()
new_board.push(move)
score = self.search(new_board, depth - 1, alpha, beta, True)[0]
min_score = min(score, min_score)
if score < min_score:
min_score = score
best_move = move
beta = min(beta, score)
if beta <= alpha:
# white has a better option avilable earlier on in the tree.
break
return (min_score, best_move)
class Viridithas():
def __init__(
self,
human: bool = False,
fen: str = 'rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1',
pgn: str = '',
time_limit: float = 15,
fun: bool = False,
contempt: int = 3000,
book: bool = True,
advancedTC: list = [],
):
if pgn == '':
self.node = Board(fen)
else:
self.node = Board()
for move in pgn.split():
try:
self.node.push_san(move)
except Exception:
continue
self.time_limit = time_limit
if advancedTC:
self.endpoint = time.time()+advancedTC[0]*60
self.increment = advancedTC[1]
else:
self.endpoint = 0
self.increment = 0
self.fun = fun
self.contempt = contempt
self.human = human
self.nodes = 0
self.advancedTC = advancedTC
self.hashtable: dict[Hashable, TTEntry] = dict()
self.inbook = book
def set_position(self, fen):
self.node = Board(fen)
def __repr__(self) -> str:
return str(self.node) + '\n' + self.__class__.__name__+"-engine at position " + str(self.node.fen())
def __str__(self) -> str:
return self.__class__.__name__
def user_setup(self):
if input("Do you want to configure the bot? (Y/N) ").upper() != 'Y':
return
myname = self.__class__.__name__.upper()
print(f"BEGINNING USER CONFIGURATION OF {myname}-BOT")
datadict = get_engine_parameters()
self.__init__(
human=datadict["human"],
fen=datadict["fen"],
pgn=datadict["pgn"],
time_limit=datadict["time_limit"],
fun=datadict["fun"],
contempt=datadict["contempt"],
book=datadict["book"],
advancedTC=datadict["advancedTC"]
)
def gameover_check_info(self):
checkmate = self.node.is_checkmate()
draw = self.node.is_stalemate() or \
self.node.is_insufficient_material( ) or \
self.node.is_repetition(2) or self.node.is_seventyfive_moves() or not any(self.node.generate_legal_moves())
return checkmate or draw, checkmate, draw
# @profile
def evaluate(self, depth: float, checkmate: bool, draw: bool) -> float:
self.nodes += 1
if checkmate:
return MATE_VALUE * int(max(depth+1, 1)) * (1 if self.node.turn else -1)
if draw:
return -self.contempt * (1 if self.node.turn else -1)
rating: float = 0
rating += pst_eval(self.node)
# rating += see_eval(self.node)
# rating += mobility(self.node) * MOBILITY_FACTOR
# rating += piece_attack_counts(self.node) * ATTACK_FACTOR
# rating += king_safety(self.node) * KING_SAFETY_FACTOR
# rating += space(self.node) * SPACE_FACTOR
return rating
def single_hash_iterator(self, best):
yield best
def captures_piece(self, p): # concentrate on MVV, then LVA
return itertools.chain(
self.node.generate_pseudo_legal_moves(self.node.pawns, p),
self.node.generate_pseudo_legal_moves(self.node.knights, p),
self.node.generate_pseudo_legal_moves(self.node.bishops, p),
self.node.generate_pseudo_legal_moves(self.node.rooks, p),
self.node.generate_pseudo_legal_moves(self.node.queens, p),
self.node.generate_pseudo_legal_moves(self.node.kings, p),
)
#@profile
def captures(self): # (MVV/LVA)
return (m for m in itertools.chain(
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.queens),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.rooks),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.bishops),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.knights),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.pawns),
) if self.node.is_legal(m))
def winning_captures(self): # (MVV/LVA)
target_all = self.node.occupied_co[not self.node.turn]
target_3 = target_all & ~self.node.pawns
target_5 = target_3 & (~self.node.bishops | ~self.node.knights)
target_9 = target_5 & ~self.node.rooks
return itertools.chain(
self.node.generate_pseudo_legal_moves(self.node.pawns, target_all),
self.node.generate_pseudo_legal_moves(self.node.knights, target_3),
self.node.generate_pseudo_legal_moves(self.node.bishops, target_3),
self.node.generate_pseudo_legal_moves(self.node.rooks, target_5),
self.node.generate_pseudo_legal_moves(self.node.queens, target_9),
self.node.generate_pseudo_legal_moves(self.node.kings, target_9),
)
def losing_captures(self): # (MVV/LVA)
target_pawns = self.node.pawns
target_pnb = target_pawns | self.node.bishops | self.node.knights
target_pnbr = target_pnb | self.node.rooks
return itertools.chain(
self.node.generate_pseudo_legal_moves(self.node.knights, target_pawns),
self.node.generate_pseudo_legal_moves(self.node.bishops, target_pawns),
self.node.generate_pseudo_legal_moves(self.node.rooks, target_pnb),
self.node.generate_pseudo_legal_moves(self.node.queens, target_pnbr),
self.node.generate_pseudo_legal_moves(self.node.kings, target_pnbr),
)
def ordered_moves(self):
return (m for m in itertools.chain(
# self.winning_captures(),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.queens),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.rooks),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.bishops),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.knights),
self.captures_piece(
self.node.occupied_co[not self.node.turn] & self.node.pawns),
self.node.generate_pseudo_legal_moves(
0xffff_ffff_ffff_ffff, ~self.node.occupied_co[not self.node.turn]),
# self.losing_captures()
) if self.node.is_legal(m))
def pass_turn(self) -> None:
self.node.push(Move.from_uci("0000"))
#@profile
def qsearch(self, alpha: float, beta: float, depth: float, colour: int, gameover: bool, checkmate: bool, draw: bool) -> float:
val = self.evaluate(1, checkmate, draw) * colour
if gameover:
return val
if val >= beta:
return beta
if (val < alpha - QUEEN_VALUE):
return alpha
alpha = max(val, alpha)
for capture in self.captures():
self.node.push(capture)
gameover, checkmate, draw = self.gameover_check_info()
score = -self.qsearch(-beta, -alpha, depth - 1, -colour, gameover, checkmate, draw)
self.node.pop()
if score >= beta:
return score
alpha = max(score, alpha)
return alpha
def tt_lookup(self, board: Board) -> "TTEntry":
key = board._transposition_key()
return self.hashtable.get(key, TTEntry.default())
def tt_store(self, board: Board, entry: TTEntry):
key = board._transposition_key()
self.hashtable[key] = entry
#@profile
def negamax(self, depth: float, colour: int, alpha: float, beta: float) -> float:
initial_alpha = alpha
# (* Transposition Table Lookup; self.node is the lookup key for ttEntry *)
tt_entry = self.tt_lookup(self.node)
if not tt_entry.is_null() and tt_entry.depth >= depth:
if tt_entry.type == EXACT:
return tt_entry.value
elif tt_entry.type == LOWERBOUND:
alpha = max(alpha, tt_entry.value)
elif tt_entry.type == UPPERBOUND:
beta = min(beta, tt_entry.value)
if alpha >= beta:
return tt_entry.value
if self.node.is_legal(tt_entry.best):
moves = itertools.chain([tt_entry.best], filter(lambda x: x != tt_entry.best, self.ordered_moves()))
else:
moves = self.ordered_moves()
else:
moves = self.ordered_moves()
gameover, checkmate, draw = self.gameover_check_info()
if gameover:
return colour * self.evaluate(depth, checkmate, draw)
if depth < 1:
return self.qsearch(alpha, beta, depth, colour, gameover, checkmate, draw)
current_pos_is_check = self.node.is_check()
if not current_pos_is_check and depth >= 3 and abs(alpha) < MATE_VALUE and abs(beta) < MATE_VALUE:
# MAKE A NULL MOVE
self.node.push(Move.null())
# PERFORM A LIMITED SEARCH
value = - self.negamax(depth - 3, -colour, -beta, -beta + 1)
# UNMAKE NULL MOVE
self.node.pop()
if value >= beta:
return beta
do_prune = self.pruning(depth, colour, alpha, beta, current_pos_is_check)
best_move = Move.null()
search_pv = True
value = float("-inf")
for move_idx, move in enumerate(moves):
if move_idx == 0:
best_move = move
gives_check = self.node.gives_check(move)
is_capture = self.node.is_capture(move)
is_promo = bool(move.promotion)
depth_reduction = search_reduction_factor(
move_idx, current_pos_is_check, gives_check, is_capture, is_promo, depth)
if do_prune:
if not gives_check and not is_capture:
continue
self.node.push(move)
if search_pv:
r = -self.negamax(depth - depth_reduction, -colour, -beta, -alpha)
value = max(value, r)
else:
r = -self.negamax(depth - depth_reduction, -colour, -alpha-1, -alpha)
if (r > alpha): # // in fail-soft ... & & value < beta) is common
r = -self.negamax(depth - depth_reduction, -colour, -beta, -alpha) #// re-search
value = max(value, r)
self.node.pop()
if value > alpha:
alpha = value
best_move = move
alpha = max(alpha, value)
if alpha >= beta:
break
search_pv = False
# (* Transposition Table Store; self.node is the lookup key for ttEntry *)
# ttEntry = TTEntry()
tt_entry.value = value
if value <= initial_alpha:
tt_entry.type = UPPERBOUND
elif value >= beta:
tt_entry.type = LOWERBOUND
else:
tt_entry.type = EXACT
tt_entry.depth = depth
tt_entry.best = best_move
tt_entry.null = False
self.tt_store(self.node, tt_entry)
return value
def pruning(self, depth, colour, alpha, beta, current_pos_is_check):
if not (not current_pos_is_check and abs(
alpha) < MATE_VALUE / 2 and abs(beta) < MATE_VALUE / 2) or depth > 2:
return False
see = see_eval(self.node) * colour
DO_D1_PRUNING = depth <= 1 and see + FUTILITY_MARGIN < alpha
DO_D2_PRUNING = depth <= 2 and see + FUTILITY_MARGIN_2 < alpha
return DO_D1_PRUNING or DO_D2_PRUNING
def move_sort(self, moves: list, ratings: list):
pairs = zip(*sorted(zip(moves, ratings), key=operator.itemgetter(1)))
moves, ratings = [list(pair) for pair in pairs]
return moves, ratings
def pv_string(self):
count = 0
moves = []
while True:
e = self.tt_lookup(self.node)
if e.is_null() or not self.node.is_legal(e.best):
break
# print(self.node.__str__())
# print(self.node.__repr__())
# print(e.best)
# print(self.node.san(e.best))
moves.append(self.node.san(e.best))
self.node.push(e.best)
count += 1
for _ in moves:
self.node.pop()
if count == 0: return ""
return " ".join(moves)
def turnmod(self) -> int:
return -1 if self.node.turn else 1
def show_iteration_data(self, moves: list, values: list, depth: float, start: float) -> tuple:
t = round(time.time()-start, 2)
print(f"{self.node.san(moves[0])} | {-round((self.turnmod()*values[0])/1000, 3)} | {str(t)}s at depth {str(depth + 1)}, {str(self.nodes)} nodes processed, at {str(int(self.nodes / (t+0.00001)))}NPS.\n", f"PV: {self.pv_string()}\n", end="")
return (self.node.san(moves[0]), self.turnmod()*values[0], self.nodes, depth+1, t)
def search(self, ponder: bool = False, readout: bool = True):
val = float("-inf")
start_time = time.time()
self.nodes = 0
moves = [next(self.ordered_moves())]
saved_position = deepcopy(self.node)
alpha, beta = float("-inf"), float("inf")
valWINDOW = PAWN_VALUE / 4
WINDOW_FAILED = False
try:
depth = 1
while depth < 40:
best = self.tt_lookup(self.node).best
time_elapsed = time.time() - start_time
# check if we aren't going to finish the next search in time
if time_elapsed > 0.5 * self.time_limit and not ponder and not WINDOW_FAILED:
return best, val
val = self.negamax(
depth, self.turnmod(), alpha=alpha, beta=beta)
# print(val)
if ((val <= alpha) or (val >= beta)):
# We fell outside the window, so try again with a
# full-width window (and the same depth).
alpha = float("-inf")
beta = float("inf")
WINDOW_FAILED = True
continue
WINDOW_FAILED = False
best = self.tt_lookup(self.node).best
# check if we've run out of time
if time_elapsed > self.time_limit and not ponder:
return best, val
moves = [self.tt_lookup(self.node).best]
values = [self.tt_lookup(self.node).value]
if readout:
self.show_iteration_data(moves, values, depth, start_time)
alpha = val - valWINDOW # Set up the window for the next iteration.
beta = val + valWINDOW
depth += 1
except KeyboardInterrupt:
self.node = saved_position
pass
return moves[0], val
def ponder(self) -> None:
self.origin = self.node.copy()
self.search(ponder=True)
def get_book_move(self):
# book = chess.polyglot.open_reader(
# r"ProDeo292/ProDeo292/books/elo2500.bin")
book = chess.polyglot.open_reader(
r"books/elo2500.bin")
main_entry = book.find(self.node)
choice = book.weighted_choice(self.node)
book.close()
return main_entry.move, choice.move
def engine_move(self) -> Move:
# add flag_func for egtb mode
if self.advancedTC:
self.time_limit = (self.endpoint-time.time())/20
print("Time for move: " + str(round(self.time_limit, 2)) + "s")
if self.inbook:
try:
best, choice = self.get_book_move()
if self.fun:
self.node.push(choice)
return choice
else:
self.node.push(best)
print(chess.pgn.Game.from_board(self.node)[-1])
except IndexError:
self.time_limit = self.time_limit*2
best, _ = self.search()
self.node.push(best)
print(chess.pgn.Game.from_board(self.node)[-1])
self.inbook = False
self.time_limit = self.time_limit/2
else:
best, _ = self.search()
self.node.push(best)
print(chess.pgn.Game.from_board(self.node)[-1])
# self.record_stack()
self.endpoint += self.increment
return best
def user_move(self) -> str:
move = input("enter move: ")
while True:
try:
self.node.push_san(move)
break
except Exception:
move = input("enter move: ")
return move
def display_ending(self) -> None:
if self.node.is_stalemate():
print('END BY STALEMATE')
elif self.node.is_insufficient_material():
print('END BY INSUFFICIENT MATERIAL')
elif self.node.is_fivefold_repetition():
print('END BY FIVEFOLD REPETITION')
elif self.node.is_checkmate:
print("BLACK" if self.node.turn == BLACK else "WHITE", 'WINS ON TURN',
self.node.fullmove_number)
else:
print('END BY UNKNOWN REASON')
def run_game(self, indefinite=True) -> str:
while not self.node.is_game_over():
print(self.__repr__())
if self.human and self.node.turn:
try:
self.ponder()
except KeyboardInterrupt:
self.node = self.origin
self.user_move()
else: # SWAP THESE ASAP
self.engine_move()
if not indefinite:
break
self.display_ending()
try:
return str(chess.pgn.Game.from_board(self.node)[-1])
except Exception:
return "PGN ERROR"
def play_viri(self, fen=None):
player_colour = input(
"Enter the human player's colour in the form b/w\n--> ")
while player_colour not in ['b', 'w']:
player_colour = input(
"Enter the human player's colour in the form b/w\n--> ")
player_colour = WHITE if player_colour == 'w' else BLACK
timeControl = int(
input("how many seconds should viri get per move?\n--> "))
self.__init__(human=True, time_limit=timeControl, fen=fen, book=True, fun=False)
self.fun = False
while not self.node.is_game_over():
print(self.__repr__())
if player_colour == self.node.turn:
# try:
# self.ponder()
# except KeyboardInterrupt:
# self.node = self.origin
# self.user_move()
self.user_move()
else:
self.engine_move()
self.display_ending()
def perftx(self, n):
if n == 0:
self.nodes += 1
else:
for move in self.ordered_moves():
self.node.push(move)
self.perftx(n - 1)
self.node.pop()
def perft(self, n):
self.nodes = 0
self.perftx(n)
print(self.nodes)
def uci(self):
start = input()
while start != "uci":
start = input()
print("id", end="")
while True:
command = input()
if command == "ucinewgame":
board = Board()
elif command.split()[0] == "position":
fen = command[command.index(
"fen") + 3: command.index("moves") - 1]
moves = command[command.index("moves"):].split()
if fen == "startpos":
fen = 'rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1'
board = Board(fen)
for move in moves:
board.push(Move.from_uci(move))
def say_last_move(game: chess.Board):
"""Take a chess.BitBoard instance and speaks the last move from it."""
move_parts = {
'K': 'king.ogg',
'B': 'bishop.ogg',
'N': 'knight.ogg',
'R': 'rook.ogg',
'Q': 'queen.ogg',
'P': 'pawn.ogg',
'+': '',
'#': '',
'x': 'takes.ogg',
'=': 'promote.ogg',
'a': 'a.ogg',
'b': 'b.ogg',
'c': 'c.ogg',
'd': 'd.ogg',
'e': 'e.ogg',
'f': 'f.ogg',
'g': 'g.ogg',
'h': 'h.ogg',
'1': '1.ogg',
'2': '2.ogg',
'3': '3.ogg',
'4': '4.ogg',
'5': '5.ogg',
'6': '6.ogg',
'7': '7.ogg',
'8': '8.ogg'
}
bit_board = game.copy()
move = bit_board.pop()
san_move = bit_board.san(move)
voice_parts = []
if san_move.startswith('O-O-O'):
voice_parts += ['castlequeenside.ogg']
elif san_move.startswith('O-O'):
voice_parts += ['castlekingside.ogg']
else:
for part in san_move:
try:
sound_file = move_parts[part]
except KeyError:
logging.warning('unknown char found in san: [%s : %s]', san_move, part)
sound_file = ''
if sound_file:
voice_parts += [sound_file]
if game.is_game_over():
if game.is_checkmate():
wins = 'whitewins.ogg' if game.turn == chess.BLACK else 'blackwins.ogg'
voice_parts += ['checkmate.ogg', wins]
elif game.is_stalemate():
voice_parts += ['stalemate.ogg']
else:
if game.is_seventyfive_moves():
voice_parts += ['75moves.ogg', 'draw.ogg']
elif game.is_insufficient_material():
voice_parts += ['material.ogg', 'draw.ogg']
elif game.is_fivefold_repetition():
voice_parts += ['repetition.ogg', 'draw.ogg']
else:
voice_parts += ['draw.ogg']
elif game.is_check():
voice_parts += ['check.ogg']
if bit_board.is_en_passant(move):
voice_parts += ['enpassant.ogg']
return voice_parts
def main():
"""
https://stackoverflow.com/a/19655992
"""
# pylint: disable=too-many-branches
input_queue = queue.Queue()
input_thread = Thread(target=add_input, args=(input_queue,))
input_thread.daemon = True
input_thread.start()
board = chess.Board(INITIAL)
time_manager = TimeManager()
LOGGER.debug("------ NEW SESSION ------")
while True:
if not input_queue.empty():
smove = input_queue.get()
LOGGER.warning("Command captured: %s", smove)
else:
if time_manager.done:
move_to_go = time_manager.decision
assert move_to_go is not None
print('bestmove', move_to_go)
continue
if smove == 'quit':
LOGGER.debug("UCI Quit called, Engine dismissed")
break
elif smove == "isready":
print("readyok")
elif smove == 'uci':
print("id name Studienarbeitsengine")
print("id author Daniel und Stefan")
elif smove == "ucinewgame":
board = Board()
elif smove == "position startpos":
board = Board()
elif smove.startswith("position startpos moves"):
moves = [Move.from_uci(uci=uci) for uci in smove.split(" ")[3:]]
LOGGER.debug("Setting moves: %s", str(moves))
board = Board()
for move in moves:
board.push(move)
elif smove.startswith("go"):
parts = smove.split(" ")[1:]
LOGGER.debug("Parts of go: %s", parts)
information = process_uci_time_information(*parts)\
if len(parts) % 2 == 0 else {"infinite": True}
time_manager.info_from_uci(**information)
time_manager.perform_search(board.copy())
elif smove.startswith("stop"):
pass
else:
print("Error (unkown command):", smove)
def score(bord: chess.Board, move):
scr = 0
board = bord.copy()
current_turn = board.turn
#Material Value
if board.is_capture(move):
captured_piece = board.piece_type_at(move.to_square)
for c in range(1, 7):
if captured_piece == c:
if current_turn == False:
#print("black caps white")
scr += COSTS[c - 1]
break
else:
scr -= COSTS[c - 1]
#print("white caps black")
break
#Pre-move misc
# if board.is_check():
# if current_turn == 0:
# scr += 5
# else:
# scr -= 5
#Freedom of board
# board.push(move) #New board with testing move
# possible_moves = board.legal_moves
# if current_turn == 0:
# scr += len(list(possible_moves)) * 0.1
# else:
# scr -= len(list(possible_moves)) * 0.1
#Centre control
# for sq in TRUE_CENTER:
# if board.piece_at(sq):
# pc = board.piece_at(sq)
# if pc.color == chess.WHITE:
# scr += 0.5
# else:
# scr -= 0.5
#
# for sq in EXTENDED_CENTER:
# if board.piece_at(sq):
# pc = board.piece_at(sq)
# if pc.color == chess.WHITE:
# scr += 0.3
# else:
# scr -= 0.3
#Attacks and Attackers:
# for square in chess.SQUARES:
#
# if board.piece_at(square):
#
# #TODO Pins
#
# piece = board.piece_at(square)
# attacks = board.attacks(square)
# attackers = board.attackers(not piece.color, square)
#
# for atk_sqr in attacks:
#
# attacked_piece = board.piece_at(atk_sqr)
#
# for d in range(1, 7):
# try:
# if attacked_piece.piece_type == d:
#
# if current_turn == 0:
# scr += COSTS[d - 1] / 2
# break
# else:
# scr -= COSTS[d - 1] / 2
# break
#
# except AttributeError:
# pass
return scr
def say_last_move(game: chess.Board):
"""Take a chess.BitBoard instance and speaks the last move from it."""
PicoTalkerDisplay.c_taken = False #molli
PicoTalkerDisplay.c_castle = False #molli
PicoTalkerDisplay.c_knight = False #molli
PicoTalkerDisplay.c_rook = False #molli
PicoTalkerDisplay.c_king = False #molli
PicoTalkerDisplay.c_bishop = False #molli
PicoTalkerDisplay.c_pawn = False #molli
PicoTalkerDisplay.c_queen = False #molli
PicoTalkerDisplay.c_check = False #molli
PicoTalkerDisplay.c_mate = False #molli
PicoTalkerDisplay.c_stalemate = False #molli
PicoTalkerDisplay.c_draw = False #molli
move_parts = {
'K': 'king.ogg',
'B': 'bishop.ogg',
'N': 'knight.ogg',
'R': 'rook.ogg',
'Q': 'queen.ogg',
'P': 'pawn.ogg',
'+': '',
'#': '',
'x': 'takes.ogg',
'=': 'promote.ogg',
'a': 'a.ogg',
'b': 'b.ogg',
'c': 'c.ogg',
'd': 'd.ogg',
'e': 'e.ogg',
'f': 'f.ogg',
'g': 'g.ogg',
'h': 'h.ogg',
'1': '1.ogg',
'2': '2.ogg',
'3': '3.ogg',
'4': '4.ogg',
'5': '5.ogg',
'6': '6.ogg',
'7': '7.ogg',
'8': '8.ogg'
}
bit_board = game.copy()
move = bit_board.pop()
san_move = bit_board.san(move)
voice_parts = []
if san_move.startswith('O-O-O'):
voice_parts += ['castlequeenside.ogg']
PicoTalkerDisplay.c_castle = True
elif san_move.startswith('O-O'):
voice_parts += ['castlekingside.ogg']
PicoTalkerDisplay.c_castle = True
else:
for part in san_move:
try:
sound_file = move_parts[part]
except KeyError:
logging.warning('unknown char found in san: [%s : %s]',
san_move, part)
sound_file = ''
if sound_file:
voice_parts += [sound_file]
if sound_file == 'takes.ogg':
PicoTalkerDisplay.c_taken = True #molli
elif sound_file == 'knight.ogg':
PicoTalkerDisplay.c_knight = True #molli
elif sound_file == 'king.ogg':
PicoTalkerDisplay.c_king = True #molli
elif sound_file == 'rook.ogg':
PicoTalkerDisplay.c_rook = True #molli
elif sound_file == 'pawn.ogg':
PicoTalkerDisplay.c_pawn = True #molli
elif sound_file == 'bishop.ogg':
PicoTalkerDisplay.c_bishop = True #molli
elif sound_file == 'queen.ogg':
PicoTalkerDisplay.c_queen = True #molli
if game.is_game_over():
if game.is_checkmate():
wins = 'whitewins.ogg' if game.turn == chess.BLACK else 'blackwins.ogg'
voice_parts += ['checkmate.ogg', wins]
PicoTalkerDisplay.c_mate = True
elif game.is_stalemate():
voice_parts += ['stalemate.ogg']
PicoTalkerDisplay.c_stalemate = True
else:
PicoTalkerDisplay.c_draw = True
if game.is_seventyfive_moves():
voice_parts += ['75moves.ogg', 'draw.ogg']
elif game.is_insufficient_material():
voice_parts += ['material.ogg', 'draw.ogg']
elif game.is_fivefold_repetition():
voice_parts += ['repetition.ogg', 'draw.ogg']
else:
voice_parts += ['draw.ogg']
elif game.is_check():
voice_parts += ['check.ogg']
PicoTalkerDisplay.c_check = True
if bit_board.is_en_passant(move):
voice_parts += ['enpassant.ogg']
return voice_parts