class AlphaGoPlayer():
def __init__(self, _, seed, player):
self.game = GoGame(13, 7.5) # THe Go Game class
#THERE IS NO init state
self.board = self.game.get_starting_board()
self.seed = seed
self.player = -1 if player == 1 else 1
self.args = parse_args()
self.nnet = NetTrainer(self.game, self.args)
self.nnet.load_checkpoint(self.args.best_model_path)
self.mct = MCT(self.nnet, self.game, self.args, noise=False)
def get_action(self, _, opponent_action):
if opponent_action != -1: # MEANS
self.board = self.game.get_next_state(self.board, -1 * self.player,
opponent_action)
self.board.set_move_num(0)
action_probs = self.mct.actionProb(self.board, self.player, 0)
self.board.set_move_num(-1)
best_action = np.argmax(action_probs)
self.board = self.game.get_next_state(self.board, self.player,
best_action)
return best_action
class AlphaGoPlayer():
def __init__(self, init_state, seed, player):
self.game = GoGame(init_state) # THe Go Game class
#THERE IS NO init state
# self.init_state = init_state
self.seed = seed
self.player = player
def get_action(self, cur_state, opponent_action):
# State is the board
#Do we have to play oppenent's action ? why don't we get the played board then ?
#This above step is wasting time.
#Run the opponent's move
cur_board = self.game.get_next_state(cur_state, -1 * self.player,
opponent_action)
#Get the possible actions
possible_actions = self.game.get_valid_moves(cur_board, self.player)
# ADDITIONAL LOGIC IF POSSIBLE ACTIONS SHAPED INCORRECTLY
# # print(actions)
# selected_action = None
# possible_actions = []
# for action , indicator in enumerate(actions):
# if indicator == 1:
# # selected_action = action
# # break
# possible_actions.append(action)
#can possible actions be null ? or will resign already be in there ?
high_score = 0
greedy_action = None
for action in possible_actions:
new_board = self.game.get_next_state(cur_board, self.player,
action)
score = self.game.get_score(new_board, self.player)
if score >= high_score: # Modify if multiple high scores
greedy_action = action # The greedy action
# assuming best action isn't None and resign and pass will be in the possible actions
#No need for final board
final_board = self.game.get_next_state(cur_board, self.player,
greedy_action)
return greedy_action
class AlphaGoPlayer():
def __init__(self, init_state, seed, player):
self.game = GoGame(init_state) # THe Go Game class
#THERE IS NO init state
# self.init_state = init_state
self.seed = seed
self.player = player
# WHERE ARE WE GETTING THE ARGS FROM ?
self.args = parse_args()
self.nnet = NetTrainer(self.game, self.args)
self.nnet.load_checkpoint(self.args.best_model_path +
str(self.args.type))
self.mct = MCT(self.nnet, self.game, self.args)
def get_action(self, cur_state, opponent_action):
cur_board = self.game.get_next_state(cur_state, -1 * self.player,
opponent_action)
action_probs = self.mct.actionProb(cur_board, self.player, 0)
best_action = np.argmax(action_probs)
return best_action
if __name__ == "__main__":
game = GoGame(13, 5.5)
board = game.get_starting_board()
player = -1
while True:
if game.get_game_ended(board, player):
break
actions = game.get_valid_moves(board, player)
actions[-1] = 0
selected_action = None
possible_actions = []
for action, indicator in enumerate(actions):
if indicator == 1:
possible_actions.append(action)
if len(possible_actions) > 0:
selected_action = random.choice(possible_actions)
else:
selected_action = game.get_action_space_size() - 1
board = game.get_next_state(board, player, selected_action)
print(selected_action)
board.print_board()
player = -player