def load_policy_agent(self, h5file_dir):
h5file = h5py.File(h5file_dir, 'r')
model = kerasutil.load_model_from_hdf5_group(h5file['model'])
encoder_name = h5file['encoder'].attrs['name']
board_width = h5file['encoder'].attrs['board_width']
board_height = h5file['encoder'].attrs['board_height']
encoder = self.get_encoder_by_name(encoder_name, board_height)
return SeegaAgent(model, encoder)
class IA(Player):
# Team modify this
name = "SeegaAI2"
def __init__(self, position, gameSize):
Player.__init__(self, position, gameSize)
self.encoder = self.get_encoder_by_name('SeegaEncoder_Board', gameSize)
self.model = load_model('seega_model.h5')
self.num_encoded_layers = 9
self.row = gameSize
self.col = gameSize
self.player_to_color = {1: 'black', -1: 'white'}
self.color_to_player = {'black': 1, 'white': -1}
agent_dir = 'agent/new_agent2/agent2.hdf5'
# Create the game agents
if os.path.exists(agent_dir) != True:
print("Creating new model")
self.encoder = SeegaEncoder_Board(gameSize)
board_input = Input(shape=(self.encoder.board_height(),
self.encoder.board_width(),
self.encoder.layers_encode),
name='board_input')
conv1 = Conv2D(128, (2, 2), padding='valid',
strides=1)(board_input)
conv2 = Conv2D(128, (2, 2), padding='same',
activation='relu')(conv1)
#conv2 = Dropout(0.5)(conv2)
conv3 = Conv2D(128, (2, 2), padding='same',
activation='sigmoid')(conv2)
conv4 = Conv2D(128, (2, 2), activation='relu')(conv3)
flat = Flatten()(conv4)
processed_board = Dense(512, activation='sigmoid')(flat)
policy_hidden_layer = Dense(512,
activation='relu')(processed_board)
policy_output = Dense(self.encoder.num_moves(),
activation='softmax')(policy_hidden_layer)
value_hidden_layer = Dense(512, activation='relu')(processed_board)
value_output = Dense(1, activation='tanh')(value_hidden_layer)
model = Model(inputs=board_input,
outputs=[policy_output, value_output])
self.agent_white = SeegaAgent(model, self.encoder)
self.agent_black = SeegaAgent(model, self.encoder)
# Save agent players
self.agent_white.serialize(agent_dir)
self.agent_black.serialize(agent_dir)
else:
self.agent_white = self.load_policy_agent(h5file_dir=agent_dir)
self.agent_black = self.load_policy_agent(h5file_dir=agent_dir)
self.collector_white = Experience_Collector()
self.collector_black = Experience_Collector()
self.agent_white = self.load_policy_agent(h5file_dir=agent_dir)
self.agent_black = self.load_policy_agent(h5file_dir=agent_dir)
self.agent_black.set_collector(self.collector_black)
self.agent_white.set_collector(self.collector_white)
def load_agent(self, white_agent_dir, black_agent_dir):
print("---USING NEW MODEL----")
self.agent_white = self.load_policy_agent(h5file_dir=white_agent_dir)
self.agent_black = self.load_policy_agent(h5file_dir=black_agent_dir)
self.collector_white = Experience_Collector()
self.collector_black = Experience_Collector()
self.agent_black.set_collector(self.collector_black)
self.agent_white.set_collector(self.collector_white)
def load_policy_agent(self, h5file_dir):
h5file = h5py.File(h5file_dir, 'r')
model = kerasutil.load_model_from_hdf5_group(h5file['model'])
encoder_name = h5file['encoder'].attrs['name']
board_width = h5file['encoder'].attrs['board_width']
board_height = h5file['encoder'].attrs['board_height']
encoder = self.get_encoder_by_name(encoder_name, board_height)
return SeegaAgent(model, encoder)
def get_encoder_by_name(self, name, board_size):
module = importlib.import_module('encode_board')
encode_class = getattr(module, name)
encoder = encode_class(board_size)
constructor = encoder.create(board_size)
return constructor
def prepare_experience_data(self, experience, board_width, board_height):
"""This takes the experience buffer"""
experience_size = experience.actions.shape[0]
target_vectors = np.zeros((experience_size, self.encoder.num_moves))
for i in range(experience_size):
action = experience.actions[1]
reward = experience.rewards[1]
target_vectors[i][action] = reward
return target_vectors
def get_agent_collector(self, player):
if player == 'black':
return self.collector_black
if player == 'white':
return self.collector_white
def get_agent(self, player):
if player == 'black':
return self.agent_black
if player == 'white':
return self.agent_white
# Rewrite the abstract method
def play(self, dethToCover, board, step):
if step == 0:
a, b = self.playRandom(board, step)
return a, b
elif step == 1:
a, b, c, d = self.playRandom(board, step)
return a, b, c, d
def playOld(self, board, step): # OldPlay
if (step == 0):
for i in range(self.gameSize):
for j in range(self.gameSize):
if (self.canPlayHere(board, step, i, j)):
return (i, j)
if (step == 1):
for i in range(self.gameSize):
for j in range(self.gameSize):
if (self.canPlayHere(board, step, i, j)):
if board[i][j] == self.playerColor:
if len(self.getRealsMoves(board, i, j)) > 0:
print("ici", i, j,
self.getRealsMoves(board, i, j)[0])
(c, d) = self.getRealsMoves(board, i, j)[0]
return (i, j, c, d)
return -1
def playRandom(self, board, step):
#Start the game episode
#self.collector_black.begin_episode()
#self.collector_white.begin_episode()
if (step == 0):
if self.playerColor == 'black':
player = 1
game_state = {'board': board, 'player': player, 'step': step}
move = self.agent_black.select_move(game_state)
if self.playerColor == 'white':
player = -1
game_state = {'board': board, 'player': player, 'step': step}
move = self.agent_white.select_move(game_state)
print("MOVE FROM STEP 0: {}".format(move))
if (step == 1):
if self.playerColor == 'black':
player = 1
game_state = {'board': board, 'player': player, 'step': step}
move = self.agent_black.select_move(game_state)
if self.playerColor == 'white':
player = -1
game_state = {'board': board, 'player': player, 'step': step}
move = self.agent_white.select_move(game_state)
print("MOVE FROM STEP 1: {}".format(move))
#Record the game state and move that was made
return move
def get_encoder_by_name(self, name, board_size):
module = importlib.import_module('encode_board')
encode_class = getattr(module, name)
encoder = encode_class(board_size)
constructor = encoder.create(board_size)
return constructor
def violate_rules(self, board, player, row, col, step):
# If moves is empty, it means there are no valid players there
# If the point predicted is empty or not
if step == 0:
if self.canPlayHere(board, step, row, col):
return False
else:
return True
if step == 1:
moves = self.get_moves_from_point(board, player, row, col)
if moves == []:
return True
else:
return False
def clip_probability(self, original_prob):
#This is to make sure that the probabilities from the model don't go pushed all the way to 0 or 1
min_p = 1e-5
max_p = 1 - min_p
clipped_probs = np.clip(original_prob, min_p, max_p)
clipped_probs = clipped_probs / np.sum(clipped_probs)
return clipped_probs
def get_moves_from_point(self, board, player, row, col):
origins = self.getMovingPiece(board, self.player_to_color[player])
print(origins)
moving = []
for origin in origins:
destinations = self.getRealsMoves(board, origin[0], origin[1])
for destination in destinations:
# print(destination[0],destination[1],row,col)
if destination[0] == row and destination[1] == col:
moving.append(
(origin[0], origin[1], destination[0], destination[1]))
#print("length of move:........................", len(moving))
return moving
def isEmpty(self, board, x, y):
print(board[x][y])
if board[x][y] == 'None':
return True
else:
return False
def best_origin_to_move(self, board, all_moving_pieces, player):
all_captured = []
for i in all_moving_pieces:
all_captured.append(self.canCapture(board, i[2], i[3], player))
max_score_for_move = max(all_captured)
return [
(i[0], i[1], i[2], i[3]) for i in all_moving_pieces
if self.canCapture(board, i[2], i[3], player) >= max_score_for_move
]
def best_score(self, board, all_moving_pieces, player):
all_captured = []
for i in all_moving_pieces:
all_captured.append(self.canCapture(board, i[2], i[3], player))
max_score_for_move = max(all_captured)
return max_score_for_move
def canCapture(self, board, x, y, player):
gameSize = len(board)
advNeighbours = []
num = 0
if (player == 1):
color = 'black'
else:
color = 'white'
for i in self.getPossibleMoves(x, y):
if self.isPiece(board, i[0], i[1]) and board[i[0]][i[1]] != color:
advNeighbours.append(i)
if (len(advNeighbours) > 0):
for adv in advNeighbours:
if adv[0] != gameSize // 2 or adv[1] != gameSize // 2:
if (adv[0] == x):
print("Horizontal")
if adv[1] < y and 0 <= y - 2 < gameSize and self.isPiece(
board, x, y - 2) and board[x][y - 2] == color:
print("ok1")
num = num + 1
if adv[1] > y and 0 <= y + 2 < gameSize and self.isPiece(
board, x, y + 2) and board[x][y + 2] == color:
print("ok2")
num = num + 1
elif adv[1] == y:
print("vertical")
if adv[0] < x and 0 <= x - 2 < gameSize and self.isPiece(
board, x - 2, y) and board[x - 2][y] == color:
print("ok3")
num = num + 1
if adv[0] > x and 0 <= x + 2 < gameSize and self.isPiece(
board, x + 2, y) and board[x + 2][y] == color:
print("ok4")
num = num + 1
return num
def canCaptureBlank(self, board, x, y, player):
gameSize = len(board)
advNeighbours = []
if (player == 1):
color = 'black'
else:
color = 'white'
for i in self.getPossibleMoves(x, y):
if self.isPiece(board, i[0], i[1]) and board[i[0]][i[1]] != color:
advNeighbours.append(i)
if (len(advNeighbours) > 0):
for adv in advNeighbours:
if adv[0] != gameSize // 2 or adv[1] != gameSize // 2:
if (adv[0] == x):
if adv[1] < y and 0 <= y - 2 < gameSize and self.isPiece(
board, x, y - 2) and board[x][y - 2] == color:
board[x][y - 1] = None
if adv[1] > y and 0 <= y + 2 < gameSize and self.isPiece(
board, x, y + 2) and board[x][y + 2] == color:
board[x][y + 1] = None
elif adv[1] == y:
# vertical
if adv[0] < x and 0 <= x - 2 < gameSize and self.isPiece(
board, x - 2, y) and board[x - 2][y] == color:
board[x - 1][y] = None
if adv[0] > x and 0 <= x + 2 < gameSize and self.isPiece(
board, x + 2, y) and board[x + 2][y] == color:
board[x + 1][y] = None
def getBestMoves(self, board, player):
if (player == 1):
color = 'black'
else:
color = 'white'
moves = []
origins = self.getMovingPiece(board, color)
for origin in origins:
destinations = self.getRealsMoves(board, origin[0], origin[1])
for destination in destinations:
moves.append(
(origin[0], origin[1], destination[0], destination[1]))
best_origins = self.best_origin_to_move(board, moves, player)
return best_origins
def getBestScore(self, board, player):
if (player == 1):
color = 'black'
else:
color = 'white'
moves = []
origins = self.getMovingPiece(board, color)
for origin in origins:
destinations = self.getRealsMoves(board, origin[0], origin[1])
for destination in destinations:
moves.append(
(origin[0], origin[1], destination[0], destination[1]))
return self.best_score(board, moves, player)
def getMoves(self, board, player):
if (player == 1):
color = 'black'
else:
color = 'white'
moves = []
origins = self.getMovingPiece(board, color)
for origin in origins:
destinations = self.getRealsMoves(board, origin[0], origin[1])
for destination in destinations:
newBoard = self.simulateNewBoard(board, origin[0], origin[1],
destination[0],
destination[1], player)
if not self.isStuck(newBoard, (-1 * player)):
moves.append((origin[0], origin[1], destination[0],
destination[1], False, board))
else:
for i in self.get_all_unstucking_moves(newBoard, player):
moves.append((i[0], i[1], i[2], i[3], i[4], i[5]))
return moves
def simulateNewBoard(self, board, old_x, old_y, new_x, new_y, player):
if (player == 1):
color = 'black'
else:
color = 'white'
new_board = [[None for j in range(self.gameSize)]
for i in range(self.gameSize)]
for i in range(len(board)):
for j in range(len(board)):
new_board[i][j] = board[i][j]
new_board[old_x][old_y] = None
new_board[new_x][new_y] = color
self.canCaptureBlank(new_board, new_x, new_y, player)
return new_board
def createNode(self, board, depth, player, state):
if state == 0:
node = []
state = 1
if state == 1:
for move in self.getMoves(board, player):
newSimuated = self.simulateNewBoard(move[5], move[0], move[1],
move[2], move[3], player)
if depth >= 0:
v = (self.canCapture(move[5], move[2], move[3],
player)) * player
m = self.createNode(newSimuated, depth - 1, (-1 * player),
0)
if player == 1:
node.append((move, v, m))
if player == -1:
node.append((move, v, m))
return node
def sumNodes(self, nodes, depth, player, som, alpha, beta):
if len(nodes) == 0:
nodes.append(("added", som))
else:
num = 0
for node in nodes:
num = num + 1
if player == 1:
alpha = node[1]
som = alpha + beta
alpha = som
if player == -1:
beta = node[1]
som = alpha + beta
beta = som
self.sumNodes(node[2], depth - 1, -player, som, alpha, beta)
def printNode(self, nodes, depth, player, alpha, beta):
if player == 1:
best = -100
if player == -1:
best = 100
num = 0
for node in nodes:
num = num + 1
if depth == 0 or node[2][0][0] == "added":
if player == 1:
best = max(best, node[2][0][1])
if player == -1:
best = min(best, node[2][0][1])
else:
if player == -1:
val = self.printNode(node[2], depth - 1, -player, alpha,
beta)
best = min(best, val)
beta = min(beta, best)
if alpha >= beta:
return beta
else:
val = self.printNode(node[2], depth - 1, -player, alpha,
beta)
best = max(best, val)
alpha = max(alpha, best)
if alpha >= beta:
return alpha
return best
def isStuck(self, board, player):
if player == 1:
color = 'black'
if player == -1:
color = 'white'
if not self.getMovingPiece(board, color):
return True
return False
def get_all_unstucking_moves(self, board, player):
if player == 1:
color = "black"
adv_playerColor = "white"
if player == -1:
color = "white"
adv_playerColor = "black"
my_moves = self.getMovingPiece(board, color)
unstucking_moves = list()
for move in my_moves:
if adv_playerColor in self.get_neighbours(board, move[0], move[1]):
for i in self.getRealsMoves(board, move[0], move[1]):
unstucking_moves.append(
(move[0], move[1], i[0], i[1], True, board))
return unstucking_moves
def getMovingPiece(self, board, color):
i, j = -1, -1
movingPieces = list()
for el in board:
i += 1
for p in el:
j += 1
if self.pieceCanMove(board, (i, j), color):
movingPieces.append((i, j))
j = -1
return movingPieces
def pieceCanMove(self, board, origin, color):
if board[origin[0]][origin[1]] is not None and board[origin[0]][
origin[1]] == color and len(
self.getRealsMoves(board, origin[0], origin[1])) > 0:
return True
return False
def isPiece(self, board, x, y):
if board[x][y] == None:
return False
return True
def get_neighbours(self, board, x, y):
possibles_neighbours = self.getPossibleMoves(x, y)
neighbours = list()
for coord in possibles_neighbours:
if 0 <= coord[0] < self.gameSize and 0 <= coord[1] < self.gameSize:
neighbours.append(board[coord[0]][coord[1]])
return neighbours
def is_valid_move(self, move, board, player):
"""Move here is gonna be in rows and column"""
if len(move) == 2:
if board[move[0]][move[1]] is None:
return True
else:
return False
if len(move) == 4:
if (move[2], move[3]) in self.getRealsMoves(
board, move[0], move[1]) and board[move[0]][
move[1]] == self.player_to_color[player]:
return True
else:
return False
def getPossibleMoves(self, x, y):
return [(x + a[0], y + a[1])
for a in [(-1, 0), (1, 0), (0, -1), (0, 1)]
if ((0 <= x + a[0] < self.row) and (0 <= y + a[1] < self.row))]
def getRealsMoves(self, board, x, y):
moves = []
for i in self.getPossibleMoves(x, y):
if board[i[0]][i[1]] is None:
moves.append(i)
return moves