def load_models(self, dir: str):
conf = Configure()
conf.get_conf()
board_conf_str = "{0}_{1}".format(conf.conf_dict["board_size"], conf.conf_dict["n_in_a_row"])
model_path = pathlib.Path(dir) / board_conf_str
model_path.mkdir(parents=True, exist_ok=True)
self.all_model_path = sorted(item for item in model_path.glob('*/') if item.is_dir())
self.all_model_name = [path.name for path in self.all_model_path]
def start():
conf = Configure()
conf.get_conf()
def player_init(player_selected, name):
if player_selected == 1:
return Human(name=name)
elif player_selected == 2:
search_times, greedy_value = set_AI_conf(search_times=2000, greedy_value=5.0)
return AI_MCTS(name=name,
search_times=search_times,
greedy_value=greedy_value,
is_output_analysis=conf.conf_dict["AI_is_output_analysis"])
elif player_selected == 3:
network = select_network()
search_times, greedy_value = set_AI_conf(search_times=400, greedy_value=5.0)
return AI_MCTS_Net(name=name,
policy_value_function=network.predict,
board_to_xlabel=network.board_to_xlabel,
is_training=False,
search_times=search_times,
greedy_value=greedy_value,
is_output_analysis=conf.conf_dict["AI_is_output_analysis"])
player1_selected, name1 = select_player("Please input first player. Press <Ctrl-C> to end\n"
"1: Human\n"
"2: AI with pure Monte Carlo tree search\n"
"3: AI with Monte Carlo tree search & neural network\n"
": ", allowed_input=[1, 2, 3])
player1 = player_init(player1_selected, name1)
player2_selected, name2 = select_player("Please input second player. Press <Ctrl-C> to end\n"
"1: Human\n"
"2: AI with pure Monte Carlo tree search\n"
"3: AI with Monte Carlo tree search & neural network\n"
": ", allowed_input=[1, 2, 3])
player2 = player_init(player2_selected, name2)
console_renderer = ConsoleRenderer()
print("############### Game Start ###############")
winner = Game.start_until_game_over(player1, player2, board_renderer=console_renderer)
if winner == BOARD.o:
print("Congrats! \"O\" wins.")
elif winner == BOARD.x:
print("Congrats! \"X\" wins.")
else:
print("Draw!")
print("############### Game Over ###############")
def select_model(dir: str, is_training=False, specified_model_name=""):
"""
选择想要使用或训练的网络模型。
Select the network model you want to use or train.
:param dir: 网络模型目录。 Network model directory.
:param is_training: 是否训练。 Whether to train.
:param specified_model_name: 指定想要选择的模型。 Specify the model you want to select.
:return: (<bool>, <str>, <str>) 是否是新的网络模型，和网络模型路径，网络模型记录路径。
Whether it is a new network model, and the network model path, and the network model record path.
"""
conf = Configure()
conf.get_conf()
board_conf_str = "{0}_{1}".format(conf.conf_dict["board_size"],
conf.conf_dict["n_in_a_row"])
model_path = pathlib.Path(dir)
model_path = model_path / board_conf_str
model_path.mkdir(parents=True, exist_ok=True)
all_model_path = sorted(item for item in model_path.glob('*/')
if item.is_dir())
all_model_name = [path.name for path in all_model_path]
if len(specified_model_name) != 0:
model_path = model_path / specified_model_name
model_path.mkdir(parents=True, exist_ok=True)
model_record_path = model_path / "latest.h5"
is_new_model = True
if model_record_path.exists():
is_new_model = False
return is_new_model, str(model_path) + "/", str(model_record_path)
if is_training:
print(
"请选择想要训练的网络模型。按 <Ctrl-C> 退出。\n"
"Please select the network model you want to train. Press <Ctrl-C> to exit."
)
print("0: 创建新的网络模型。 Create a new network model.")
else:
print(
"请选择想要使用的网络模型。按 <Ctrl-C> 退出。\n"
"Please select the network model you want to use. Press <Ctrl-C> to exit."
)
for i, one_model_name in enumerate(all_model_name):
print("{0}: {1}".format(i + 1, one_model_name))
model_selected = select(": ",
allowed_input=range(0 if is_training else 1,
len(all_model_path) + 1))
if model_selected == 0:
while True:
new_name = input(
"请输入新的模型名称。按 <Ctrl-C> 退出。\n"
"Please enter a new model name. Press <Ctrl-C> to exit.\n"
": ")
if len(new_name) == 0:
print("模型名称为空，请重新输入。\n"
"Model name is empty, please try again.\n")
continue
if new_name in all_model_name:
print("该模型名称已存在，请重新输入。\n"
"The model name already exists, please try again.\n")
continue
model_path = model_path / new_name
model_path.mkdir(parents=True, exist_ok=True)
return True, str(model_path) + "/", None
else:
model_path = all_model_path[model_selected - 1]
model_record_path = sorted(item for item in model_path.glob('*.h5'))
model_record_name = [path.name[:-3] for path in model_record_path]
if is_training:
print(
"请选择想要训练的模型记录。按 <Ctrl-C> 退出。\n"
"Please select the model record you want to train. Press <Ctrl-C> to exit."
)
else:
print(
"请选择想要使用的模型记录。按 <Ctrl-C> 退出。\n"
"Please select the model record you want to use. Press <Ctrl-C> to exit."
)
for i, one_model_record_name in enumerate(model_record_name):
print("{0}: {1}".format(i + 1, one_model_record_name))
model_record_selected = select(": ",
allowed_input=range(
1,
len(model_record_path) + 1))
return False, str(model_path) + "/", str(
model_record_path[model_record_selected - 1])
import numpy as np
from Game.BoardRenderer import BoardRenderer
from Function import coordinates_set
from configure import Configure
conf = Configure()
conf.get_conf()
# Fixed Configuration.
o = conf.conf_dict["o"]
x = conf.conf_dict["x"]
empty = conf.conf_dict["empty"]
# Changeable Configuration.
n_in_a_row = conf.conf_dict["n_in_a_row"] # How many pieces in a row.
o_win = n_in_a_row
x_win = -n_in_a_row
start_player = conf.conf_dict["start_player"] # start player
board_size = conf.conf_dict["board_size"] # The size of the board.
class Board:
def __init__(self):
self.board = np.zeros((board_size, board_size))
self.available_actions = coordinates_set(board_size, board_size)
self.last_action = None # Last move.
self.current_player = start_player # current player
def __copy__(self):
new_board = Board()
def select_model(dir: str, is_training=False, specified_model_name=""):
conf = Configure()
conf.get_conf()
board_conf_str = "{0}_{1}".format(conf.conf_dict["board_size"],
conf.conf_dict["n_in_a_row"])
model_path = pathlib.Path(dir)
model_path = model_path / board_conf_str
model_path.mkdir(parents=True, exist_ok=True)
all_model_path = sorted(item for item in model_path.glob('*/')
if item.is_dir())
all_model_name = [path.name for path in all_model_path]
if len(specified_model_name) != 0:
model_path = model_path / specified_model_name
model_path.mkdir(parents=True, exist_ok=True)
model_record_path = model_path / "latest.h5"
is_new_model = True
if model_record_path.exists():
is_new_model = False
return is_new_model, str(model_path) + "/", str(model_record_path)
if is_training:
print(
"Please select the network model you want to train. Press <Ctrl-C> to exit."
)
print("0: Create a new network model.")
else:
print(
"Please select the network model you want to use. Press <Ctrl-C> to exit."
)
for i, one_model_name in enumerate(all_model_name):
print("{0}: {1}".format(i + 1, one_model_name))
model_selected = select(": ",
allowed_input=range(0 if is_training else 1,
len(all_model_path) + 1))
if model_selected == 0:
while True:
new_name = input(
"Please enter a new model name. Press <Ctrl-C> to exit.\n"
": ")
if len(new_name) == 0:
print("Model name is empty, please try again.\n")
continue
if new_name in all_model_name:
print("The model name already exists, please try again.\n")
continue
model_path = model_path / new_name
model_path.mkdir(parents=True, exist_ok=True)
return True, str(model_path) + "/", None
else:
model_path = all_model_path[model_selected - 1]
model_record_path = sorted(item for item in model_path.glob('*.h5'))
model_record_name = [path.name[:-3] for path in model_record_path]
if is_training:
print(
"Please select the model record you want to train. Press <Ctrl-C> to exit."
)
else:
print(
"Please select the model record you want to use. Press <Ctrl-C> to exit."
)
for i, one_model_record_name in enumerate(model_record_name):
print("{0}: {1}".format(i + 1, one_model_record_name))
model_record_selected = select(": ",
allowed_input=range(
1,
len(model_record_path) + 1))
return False, str(model_path) + "/", str(
model_record_path[model_record_selected - 1])
