id
int64 0
458k
| file_name
stringlengths 4
119
| file_path
stringlengths 14
227
| content
stringlengths 24
9.96M
| size
int64 24
9.96M
| language
stringclasses 1
value | extension
stringclasses 14
values | total_lines
int64 1
219k
| avg_line_length
float64 2.52
4.63M
| max_line_length
int64 5
9.91M
| alphanum_fraction
float64 0
1
| repo_name
stringlengths 7
101
| repo_stars
int64 100
139k
| repo_forks
int64 0
26.4k
| repo_open_issues
int64 0
2.27k
| repo_license
stringclasses 12
values | repo_extraction_date
stringclasses 433
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2,287,500 | constants.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/constants.py |
board_size = 8
SAMPLE_AZ_ARGS = {
"num_net_channels": 512,
"num_net_in_channels": 1,
"net_dropout": 0.3,
"net_action_size": board_size ** 2,
"num_simulations": 1317,
"self_play_games": 300,
"num_iters": 50,
"epochs": 500,
"lr": 0.0032485504583772953,
"max_buffer_size": 100_000,
"num_pit_games": 40,
"random_pit_freq": 3,
"board_size": board_size,
"batch_size": 256,
"tau": 1,
"arena_tau": 0.04139160592420218,
"c": 1,
"checkpoint_dir": None,
"update_threshold": 0.6,
"minimax_depth": 4,
"show_tqdm": True,
"num_workers": 5,
"num_to_win": 5,
"log_epsilon": 1e-9,
"zero_tau_after": 5,
"az_net_linear_input_size": 18432,
"log_dir": "Logs",
"pushbullet_token": None
}
TRAINED_AZ_NET_ARGS = {
"num_net_channels": 512,
"num_net_in_channels": 1,
"net_dropout": 0.3,
"net_action_size": board_size ** 2,
"num_simulations": 1317,
"self_play_games": 3,
"num_iters": 50,
"epochs": 320,
"lr": 0.0032485504583772953,
"max_buffer_size": 100_000,
"num_pit_games": 40,
"random_pit_freq": 3,
"board_size": board_size,
"batch_size": 128,
"tau": 1.0,
"arena_tau": 0, # 0.04139160592420218
"c": 1.15,
"checkpoint_dir": None,
"update_threshold": 0.6,
"minimax_depth": 4,
"show_tqdm": True,
"num_workers": 5,
"num_to_win": 4,
"log_epsilon": 1.4165210108199043e-08,
"log_dir": "Logs",
"pushbullet_token": None
}
SAMPLE_MZ_ARGS = {
"num_net_channels": 512,
"num_net_out_channels": 256,
"num_net_in_channels": 1,
"net_dropout": 0.3,
"net_action_size": 14,
"net_latent_size": 36,
"num_simulations": 240,
"self_play_games": 5,
"K": 5,
"gamma": 0.997,
"frame_buffer_size": 32,
"frame_skip": 4,
"num_steps": 400,
"num_iters": 50,
"epochs": 100,
"lr": 0.001,
"max_buffer_size": 70_000,
"num_pit_games": 40,
"random_pit_freq": 2,
"board_size": board_size,
"batch_size": 255,
"tau": 1,
"arena_tau": 1e-2,
"c": 1,
"c2": 19652,
"alpha": 0.8,
"checkpoint_dir": None,
"update_threshold": 0.6,
"minimax_depth": None, # don't use with muzero
"show_tqdm": False,
"num_workers": 5,
"num_to_win": 5,
"log_epsilon": 1e-9,
"zero_tau_after": 5,
"beta": 1,
"env_id": "ALE/Asteroids-v5",
"pickle_dir": "Pickles/Data",
"target_resolution": (96, 96),
"az_net_linear_input_size": 8192,
"log_dir": "Logs",
"pushbullet_token": None
}
| 2,570 | Python | .py | 102 | 20.421569 | 51 | 0.569631 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,501 | checkpointer.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/checkpointer.py | import atexit
import os
import pickle
import sys
import torch as th
from mu_alpha_zero.AlphaZero.utils import DotDict
from mu_alpha_zero.General.utils import find_project_root
from mu_alpha_zero.config import Config
class CheckPointer:
def __init__(self, checkpoint_dir: str | None, verbose: bool = True) -> None:
self.__checkpoint_dir = checkpoint_dir
self.make_dir()
self.__checkpoint_num = self.initialize_checkpoint_num()
self.__name_prefix = "improved_net_"
self.verbose = verbose
atexit.register(self.cleanup)
def make_dir(self) -> None:
if self.__checkpoint_dir is not None:
os.makedirs(self.__checkpoint_dir, exist_ok=True)
return
root_dir = find_project_root()
checkpoint_dir = f"{root_dir}/Checkpoints/NetVersions"
self.__checkpoint_dir = checkpoint_dir
os.makedirs(checkpoint_dir, exist_ok=True)
def save_checkpoint(self, net: th.nn.Module, opponent: th.nn.Module, optimizer: th.optim,
lr: float,
iteration: int, mu_alpha_zero_config: Config, name: str = None) -> None:
if name is None:
name = self.__name_prefix + str(self.__checkpoint_num)
checkpoint_path = f"{self.__checkpoint_dir}/{name}.pth"
th.save({
"net": net.state_dict(),
"optimizer": optimizer.state_dict() if isinstance(optimizer, th.optim.Optimizer) else optimizer,
"lr": lr,
"iteration": iteration,
"args": mu_alpha_zero_config.to_dict(),
'opponent_state_dict': opponent.state_dict()
}, checkpoint_path)
self.print_verbose(f"Saved checkpoint to {checkpoint_path} at iteration {iteration}.")
self.__checkpoint_num += 1
def save_state_dict_checkpoint(self, net: th.nn.Module, name: str) -> None:
checkpoint_path = f"{self.__checkpoint_dir}/{name}.pth"
th.save(net.state_dict(), checkpoint_path)
self.print_verbose(f"Saved state dict checkpoint.")
def load_state_dict_checkpoint(self, net: th.nn.Module, name: str) -> None:
checkpoint_path = f"{self.__checkpoint_dir}/{name}.pth"
net.load_state_dict(th.load(checkpoint_path))
self.print_verbose(f"Loaded state dict checkpoint.")
def load_checkpoint_from_path(self, checkpoint_path: str) -> tuple:
sys.modules['mem_buffer'] = sys.modules['mu_alpha_zero.mem_buffer']
checkpoint = th.load(checkpoint_path)
self.print_verbose(f"Restoring checkpoint {checkpoint_path} made at iteration {checkpoint['iteration']}.")
if "memory" in checkpoint:
memory = checkpoint["memory"]
else:
memory = None
return checkpoint["net"], checkpoint["optimizer"], memory, checkpoint["lr"], \
DotDict(checkpoint["args"]), checkpoint["opponent_state_dict"]
def load_checkpoint_from_num(self, checkpoint_num: int) -> tuple:
checkpoint_path = f"{self.__checkpoint_dir}/{self.__name_prefix}{checkpoint_num}.pth"
return self.load_checkpoint_from_path(checkpoint_path)
def clear_checkpoints(self) -> None:
# This method doesn't obey the verbose flag as it's a destructive operation.
print("Clearing all checkpoints.")
answer = input("Are you sure?? (y/n): ")
if answer != "y":
print("Aborted.")
return
for file_name in os.listdir(self.__checkpoint_dir):
os.remove(f"{self.__checkpoint_dir}/{file_name}")
print(f"Cleared {len(os.listdir(self.__checkpoint_dir))} saved checkpoints (all).")
def save_temp_net_checkpoint(self, net) -> None:
process_pid = os.getpid()
os.makedirs(f"{self.__checkpoint_dir}/Temp", exist_ok=True)
checkpoint_path = f"{self.__checkpoint_dir}/Temp/temp_net_{process_pid}.pth"
th.save(net.state_dict(), checkpoint_path)
def load_temp_net_checkpoint(self, net) -> None:
process_pid = os.getpid()
checkpoint_path = f"{self.__checkpoint_dir}/Temp/temp_net_{process_pid}.pth"
net.load_state_dict(th.load(checkpoint_path))
def initialize_checkpoint_num(self) -> int:
return len([x for x in os.listdir(self.__checkpoint_dir) if x.endswith(".pth")])
def get_highest_checkpoint_num(self) -> int:
return max([int(file_name.split("_")[2].split(".")[0]) for file_name in os.listdir(self.__checkpoint_dir)])
def get_temp_path(self) -> str:
return f"{self.__checkpoint_dir}/Temp/temp_net.pth"
def get_checkpoint_dir(self) -> str:
return self.__checkpoint_dir
def get_latest_name_match(self, name: str):
name_matches = [os.path.join(self.__checkpoint_dir, x) for x in os.listdir(self.__checkpoint_dir) if name in x]
name_matches.sort(key=lambda x: os.path.getctime(x))
return name_matches[-1]
def get_name_prefix(self):
return self.__name_prefix
def print_verbose(self, msg: str) -> None:
if self.verbose:
print(msg)
def cleanup(self):
import shutil
shutil.rmtree(f"{self.__checkpoint_dir}/Temp/", ignore_errors=True)
def save_losses(self, losses: list[float]):
with open(f"{self.__checkpoint_dir}/training_losses.pkl", "wb") as f:
pickle.dump(losses, f)
| 5,354 | Python | .py | 103 | 43.213592 | 119 | 0.639372 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,502 | utils.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/utils.py | import json
import os
import shutil
import time
from typing import Type, Literal, Callable
import numpy as np
import optuna
# import pygraphviz
import torch as th
from mu_alpha_zero.AlphaZero.constants import SAMPLE_AZ_ARGS as test_args
from mu_alpha_zero.mem_buffer import MemBuffer
from mu_alpha_zero.config import Config, AlphaZeroConfig
from mu_alpha_zero.General.network import GeneralNetwork
class DotDict(dict):
def __getattr__(self, name):
return self[name]
def __setattr__(self, name, value):
self[name] = value
def augment_experience_with_symmetries(game_experience: list, board_size) -> list:
game_experience_ = []
for state, pi, v, _ in game_experience:
pi = np.array([x for x in pi.values()])
game_experience_.append((state, pi, v))
for axis, k in zip([0, 1], [1, 3]):
state_ = np.rot90(state.copy(), k=k)
pi_ = np.rot90(pi.copy().reshape(board_size, board_size), k=k).flatten()
game_experience_.append((state_, pi_, v))
del state_, pi_
state_ = np.flip(state.copy(), axis=axis)
pi_ = np.flip(pi.copy().reshape(board_size, board_size), axis=axis).flatten()
game_experience_.append((state_, pi_, v))
return game_experience_
def rotate_stack(state: np.ndarray, k: int):
for dim in range(state.shape[0]):
state[dim] = np.rot90(state[dim], k=k)
return state
def flip_stack(state: np.ndarray, axis: int):
for dim in range(state.shape[0]):
state[dim] = np.flip(state[dim], axis=axis)
return state
def make_channels(game_experience: list):
experience = []
for state, pi, v, current_player, move in game_experience:
state = make_channels_from_single(state)
experience.append((state, pi, v, current_player, move))
return experience
def make_channels_from_single(state: np.ndarray):
player_one_state = np.where(state == 1, 1, 0) # fill with 1 where player 1 has a piece else 0
player_minus_one_state = np.where(state == -1, 1, 0) # fill with 1 where player -1 has a piece else 0
empty_state = np.where(state == 0, 1, 0) # fill with 1 where empty spaces else 0
return np.stack([state, player_one_state, player_minus_one_state, empty_state], axis=0)
def mask_invalid_actions(probabilities: np.ndarray, mask: np.ndarray) -> np.ndarray:
mask = mask.reshape(probabilities.shape)
valids = probabilities * mask
valids_sum = valids.sum()
return valids / valids_sum
# to_print = "" # for debugging
# # mask = np.where(observations != 0, -5, observations)
# # mask = np.where(mask == 0, 1, mask)
# # mask = np.where(mask == -5, 0, mask)
# valids = probabilities.reshape(-1, board_size ** 2) * mask.reshape(-1, board_size ** 2)
# valids_sum = valids.sum()
# if valids_sum == 0:
# # When no valid moves are available (shouldn't happen) sum of valids is 0, making the returned valids an array
# # of nan's (result of division by zero). In this case, we create a uniform probability distribution.
# to_print += f"Sum of valid probabilities is 0. Creating a uniform probability...\nMask:\n{mask}"
# valids = np.full(valids.shape, 1.0 / np.prod(valids.shape))
# else:
# valids = valids / valids_sum # normalize
#
# if len(to_print) > 0:
# print(to_print, file=open("masking_message.txt", "w"))
# return valids
def mask_invalid_actions_batch(states: th.tensor) -> th.tensor:
masks = []
for state in states:
np_state = state.detach().cpu().numpy()
mask = np.where(np_state != 0, -5, np_state)
mask = np.where(mask == 0, 1, mask)
mask = np.where(mask == -5, 0, mask)
masks.append(mask)
return th.tensor(np.array(masks), dtype=th.float32).squeeze(1)
def check_args(args: dict):
required_keys = ["num_net_channels", "num_net_in_channels", "net_dropout", "net_action_size", "num_simulations",
"self_play_games", "num_iters", "epochs", "lr", "max_buffer_size", "num_pit_games",
"random_pit_freq", "board_size", "batch_size", "tau", "c", "checkpoint_dir", "update_threshold"]
for key in required_keys:
if key not in args:
raise KeyError(f"Missing key {key} in args dict. Please supply all required keys.\n"
f"Required keys: {required_keys}.")
def calculate_board_win_positions(n: int, k: int):
return get_num_horizontal_conv_slides(n, k) * (n * 2 + 2) + 4 * sum(
[get_num_horizontal_conv_slides(x, k) for x in range(k, n)])
def get_num_horizontal_conv_slides(board_size: int, kernel_size: int) -> int:
return (board_size - kernel_size) + 1
def az_optuna_parameter_search(n_trials: int, target_values: list, target_game, config: AlphaZeroConfig,
net_class: Type[GeneralNetwork], results_dir: str, az,refresh_az: Callable):
"""
Performs a hyperparameter search using optuna. This method is meant to be called using the start_jobs.py script.
For this method to work, a mysql database must be running on the storage address and an optuna study with the
given name and the 'maximize' direction must exist.
:param n_trials: num of trials to run the search for.
:param config: The config to use for the search.
:return:
"""
def get_function_from_value(value, trial: optuna.Trial):
if type(value[1]) == int:
return trial.suggest_int(value[0], value[1], value[2])
if type(value[1]) == float:
return trial.suggest_float(value[0], value[1], value[2])
if type(value[1]) == list:
return trial.suggest_categorical(value[0], value[1])
def objective(trial: optuna.Trial):
az = refresh_az()
for value in target_values:
setattr(config, value[0], get_function_from_value(value, trial))
az.trainer.opponent_network.load_state_dict(az.trainer.network.state_dict())
shared_storage_manager = SharedStorageManager()
shared_storage_manager.start()
mem = shared_storage_manager.MemBuffer(az.trainer.memory.max_size, az.trainer.memory.disk,
az.trainer.memory.full_disk,
az.trainer.memory.dir_path, hook_manager=az.trainer.memory.hook_manager)
shared_storage: SharedStorage = shared_storage_manager.SharedStorage(mem)
shared_storage.set_stable_network_params(az.trainer.network.state_dict())
pool = az.trainer.mcts.start_continuous_self_play(
az.trainer.make_n_networks(az.trainer.muzero_alphazero_config.num_workers),
az.trainer.make_n_trees(az.trainer.muzero_alphazero_config.num_workers),
shared_storage, az.trainer.device,
az.trainer.muzero_alphazero_config,
az.trainer.muzero_alphazero_config.num_workers,
az.trainer.muzero_alphazero_config.num_worker_iters)
az.trainer.logger.log(
f"Successfully started a pool of {az.trainer.muzero_alphazero_config.num_workers} workers for "
f"self-play (1/2).")
p2 = Process(target=az.trainer.network.continuous_weight_update,
args=(
shared_storage, az.trainer.muzero_alphazero_config, az.trainer.checkpointer,
az.trainer.logger))
p2.start()
p4 = Process(target=az.trainer.arena.continuous_pit, args=(
az.trainer.net_player.make_fresh_instance(),
az.trainer.net_player.make_fresh_instance(),
RandomPlayer(az.trainer.game_manager.make_fresh_instance(), **{}),
az.trainer.muzero_alphazero_config.num_pit_games,
az.trainer.muzero_alphazero_config.num_simulations,
shared_storage,
az.trainer.checkpointer,
False,
1
))
p4.start()
last_len = 0
max_len = 500
while len(shared_storage.get_combined_losses()) < max_len:
if len(shared_storage.get_combined_losses()) <= last_len:
time.sleep(2)
continue
last_len = len(shared_storage.get_combined_losses())
trial.report(shared_storage.get_combined_losses()[-1], len(shared_storage.get_combined_losses()))
pool.terminate()
p2.terminate()
p4.terminate()
return shared_storage.get_combined_losses()[-1]
from mu_alpha_zero.shared_storage_manager import SharedStorageManager, SharedStorage
from mu_alpha_zero.AlphaZero.Arena.players import RandomPlayer
from mu_alpha_zero.mem_buffer import MemBuffer
from multiprocess.context import Process
config.show_tqdm = False
study = optuna.create_study(study_name="AlphaZeroHyperparameterSearch", direction="minimize")
study.optimize(objective, n_trials=n_trials)
with open(f"{results_dir}/best_params.json", "w") as f:
json.dump(study.best_params, f)
def build_net_from_config(muzero_config: Config, device):
from mu_alpha_zero.AlphaZero.Network.nnet import AlphaZeroNet
network = AlphaZeroNet(muzero_config.num_net_in_channels, muzero_config.num_net_channels,
muzero_config.net_dropout, muzero_config.net_action_size,
muzero_config.az_net_linear_input_size)
return network.to(device)
def build_all_from_config(muzero_alphazero_config: Config, device, lr=None, buffer_size=None):
if lr is None:
lr = muzero_alphazero_config.lr
if buffer_size is None:
buffer_size = muzero_alphazero_config.max_buffer_size
network = build_net_from_config(muzero_alphazero_config, device)
optimizer = th.optim.Adam(network.parameters(), lr=lr)
memory = MemBuffer(max_size=buffer_size)
return network, optimizer, memory
def make_net_from_checkpoint(checkpoint_path: str, args: DotDict | None):
if args is None:
args = DotDict(test_args)
device = th.device("cuda" if th.cuda.is_available() else "cpu")
net = build_net_from_config(args, device)
data = th.load(checkpoint_path)
net.load_state_dict(data["net"])
return net
def visualize_tree(root_node, output_file_name: str, depth_limit: int | None = None):
graph = pygraphviz.AGraph()
graph.graph_attr["label"] = "MCTS visualization"
graph.node_attr["shape"] = "circle"
graph.edge_attr["color"] = "blue"
graph.node_attr["color"] = "gold"
if depth_limit is None:
depth_limit = float("inf")
def make_graph(node, parent, g: pygraphviz.AGraph, d_limit: int):
state_ = None
if node.state is None:
state_ = str(np.random.randint(low=0, high=5, size=parent.state.shape))
else:
state_ = str(node.state)
g.add_node(state_)
if parent != node:
g.add_edge(str(parent.state), state_)
if not node.was_visited() or d_limit <= 0:
return
# queue_ = deque(root_node.children.values())
# depth = 1
# num_children = 25
# children_iterated = 0
# parent = root_node
for child in node.children.values():
make_graph(child, node, g, d_limit=d_limit - 1 if depth_limit != float("inf") else depth_limit)
make_graph(root_node, root_node, graph, d_limit=depth_limit)
graph.layout(prog="dot")
graph.draw(f"{output_file_name}.png")
def cpp_data_to_memory(data: list, memory: MemBuffer, board_size: int):
# import pickle
# test_data = pickle.load(open(f"{find_project_root()}/history.pkl","rb"))
for game_data in data:
for state, pi, v in game_data:
state = th.tensor(state, dtype=th.float32).reshape(board_size, board_size)
pi = th.tensor(pi, dtype=th.float32)
memory.add((state, pi, v))
#
# if __name__ == "__main__":
# cpp_data_to_memory(None,MemBuffer(max_size=10_000),test_args)
| 11,998 | Python | .py | 235 | 42.748936 | 120 | 0.644632 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,503 | __init__.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__init__.py | #from mu_alpha_zero.mem_buffer import MemBuffer
# from AlphaZero.Network.trainer import Trainer
# from AlphaZero.Network.nnet import TicTacToeNet
# from import cbind | 166 | Python | .py | 4 | 40.75 | 49 | 0.834356 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,504 | logger.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/logger.py | import atexit
import datetime
import logging
import os
from pushbullet import API
from mu_alpha_zero.General.utils import find_project_root
class Logger:
def __init__(self, logdir: str or None, token: str or None = None) -> None:
self.logdir = self.init_logdir(logdir)
os.makedirs(self.logdir, exist_ok=True)
self.logger = logging.getLogger("AlphaZeroLogger")
self.logger.setLevel(logging.DEBUG)
self.file_handler = logging.FileHandler(
f"{self.logdir}/{datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S')}.log")
self.file_handler.setLevel(logging.DEBUG)
self.logger.addHandler(self.file_handler)
self.is_token_set = False
self.api = API()
self.init_api_token(token)
formatter = logging.Formatter("[%(asctime)s - %(levelname)s] %(message)s")
self.file_handler.setFormatter(formatter)
atexit.register(self.cleanup)
def log(self, msg: str, level: str = "debug") -> None:
getattr(self.logger, level)(msg)
def init_logdir(self,logdir: str or None):
if logdir is None:
return f"{find_project_root()}/Logs/ProgramLogs"
else:
return logdir
def init_api_token(self, token: str or None) -> None:
if token is None:
return
self.api.set_token(token)
self.is_token_set = True
def pushbullet_log(self, msg: str, algorithm: str = "MuZero") -> None:
if not self.is_token_set:
return
try:
self.api.send_note(f"{algorithm} training notification.", msg)
except Exception as e:
print(e)
def clear_logdir(self):
for file_name in os.listdir(self.logdir):
os.remove(f"{self.logdir}/{file_name}")
def cleanup(self) -> None:
self.file_handler.close()
self.logger.removeHandler(self.file_handler)
class LoggingMessageTemplates:
@staticmethod
def PITTING_START(name1: str, name2: str, num_games: int):
return f"Starting pitting between {name1} and {name2} for {num_games} games."
@staticmethod
def PITTING_END(name1: str, name2: str, wins1: int, wins2: int, total: int, draws: int):
return (f"Pitting ended between {name1} and {name2}. "
f"Player 1 win frequency: {wins1 / total}. "
f"Player 2 win frequency: {wins2 / total}. Draws: {draws}.")
@staticmethod
def SELF_PLAY_START(num_games: int):
return f"Starting self play for {num_games} games."
@staticmethod
def SELF_PLAY_END(wins1: int, wins2: int, draws: int, not_zero_fn: callable):
if wins1 is None or wins2 is None or draws is None:
return "Self play ended. Results not available (This is expected if you are running MuZero)."
return (f"Self play ended. Player 1 win frequency: {wins1 / (not_zero_fn(wins1 + wins2 + draws))}. "
f"Player 2 win frequency: {wins2 / (not_zero_fn(wins1 + wins2 + draws))}. Draws: {draws}.")
@staticmethod
def NETWORK_TRAINING_START(num_epchs: int):
return f"Starting network training for {num_epchs} epochs."
@staticmethod
def NETWORK_TRAINING_END(mean_loss: float):
return f"Network training ended. Mean loss: {mean_loss}"
@staticmethod
def MODEL_REJECT(num_wins: float, update_threshold: float):
return (f"!!! Model rejected, restoring previous version. Win rate: {num_wins}. "
f"Update threshold: {update_threshold} !!!")
@staticmethod
def MODEL_ACCEPT(num_wins: float, update_threshold: float):
return (
f"!!! Model accepted, keeping current version. Win rate: {num_wins}. Update threshold: {update_threshold}"
f" !!!")
@staticmethod
def TRAINING_START(num_iters: int):
return f"Starting training for {num_iters} iterations."
@staticmethod
def TRAINING_END(args_used: dict):
args_used_str = ""
for key, value in args_used.items():
args_used_str += f"{key}: {value}, "
return f"Training ended. Args used: {args_used_str[:-2]}"
@staticmethod
def SAVED(type_: str, path: str):
return f"Saved {type_} to {path}"
@staticmethod
def LOADED(type_: str, path: str):
return f"Restored {type_} from {path}"
@staticmethod
def ITER_FINISHED_PSB(iter: int):
return f"Iteration {iter} of the algorithm training finished!"
@staticmethod
def TRAINING_END_PSB():
return "Algorithm Training finished, you can collect the results :)"
| 4,590 | Python | .py | 101 | 37.415842 | 118 | 0.640376 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,505 | alpha_zero.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/alpha_zero.py | import sys
from typing import Type
import numpy as np
import torch as th
from mu_alpha_zero.AlphaZero.Arena.arena import Arena
from mu_alpha_zero.AlphaZero.Arena.players import NetPlayer
from mu_alpha_zero.AlphaZero.MCTS.az_search_tree import McSearchTree
from mu_alpha_zero.trainer import Trainer
from mu_alpha_zero.General.az_game import AlphaZeroGame
from mu_alpha_zero.General.memory import GeneralMemoryBuffer
from mu_alpha_zero.General.network import GeneralNetwork
from mu_alpha_zero.General.utils import net_not_none, find_project_root
from mu_alpha_zero.Hooks.hook_manager import HookManager
from mu_alpha_zero.config import AlphaZeroConfig
class AlphaZero:
def __init__(self, game_instance: AlphaZeroGame):
self.trainer = None
self.net = None
self.game = game_instance
self.device = th.device("cuda" if th.cuda.is_available() else "cpu")
self.alpha_zero_config: AlphaZeroConfig = None
self.tree: McSearchTree = None
def create_new(self, alpha_zero_config: AlphaZeroConfig, network_class: Type[GeneralNetwork],
memory: GeneralMemoryBuffer, headless: bool = True, hook_manager: HookManager or None = None,
checkpointer_verbose: bool = False):
network = network_class.make_from_config(alpha_zero_config, hook_manager=hook_manager).to(
self.device)
tree = McSearchTree(self.game.make_fresh_instance(), alpha_zero_config)
self.tree = tree
net_player = NetPlayer(self.game.make_fresh_instance(), **{"network": network, "monte_carlo_tree_search": tree})
self.alpha_zero_config = alpha_zero_config
self.trainer = Trainer.create(alpha_zero_config, self.game, network, tree, net_player, headless=headless,
checkpointer_verbose=checkpointer_verbose, memory_override=memory,
hook_manager=hook_manager)
self.net = self.trainer.get_network()
def load_checkpoint(self, network_class: Type[GeneralNetwork], path: str, checkpoint_dir: str,
headless: bool = True, hook_manager: HookManager or None = None,
checkpointer_verbose: bool = False, memory: GeneralMemoryBuffer = None):
self.trainer = Trainer.from_checkpoint(network_class, McSearchTree, NetPlayer, path, checkpoint_dir, self.game,
headless=headless, hook_manager=hook_manager,
checkpointer_verbose=checkpointer_verbose, mem=memory)
self.net = self.trainer.get_network()
self.tree = self.trainer.get_tree()
self.args = self.trainer.get_args()
def train(self):
net_not_none(self.net)
self.trainer.train()
def train_parallel(self, use_pitting: bool):
net_not_none(self.net)
self.trainer.train_parallel(False, use_pitting)
def predict(self, x: np.ndarray, tau: float = 0) -> int:
net_not_none(self.net)
assert x.shape == (self.args["board_size"], self.args["board_size"], self.args[
"num_net_in_channels"]), "Input shape is not correct. Expected (board_size, board_size, num_net_in_channels)." \
"Got: " + str(x.shape)
pi, _ = self.tree.search(self.net, x, 1, self.device, tau=tau)
return self.game.select_move(pi, tau=self.alpha_zero_config.tau)
def play(self, p1_name: str, p2_name: str, num_games: int, alpha_zero_config: AlphaZeroConfig, starts: int = 1,
switch_players: bool = True):
net_not_none(self.net)
self.net.to(self.device)
self.net.eval()
manager = self.game.make_fresh_instance()
tree = McSearchTree(manager, alpha_zero_config)
kwargs = {"network": self.net, "monte_carlo_tree_search": tree, "evaluate_fn": manager.eval_board,
"depth": alpha_zero_config.minimax_depth, "player": -1}
path_prefix = find_project_root().replace("\\", "/").split("/")[-1]
p1 = sys.modules[f"{path_prefix}.AlphaZero.Arena.players"].__dict__[p1_name](manager, **kwargs)
p2 = sys.modules[f"{path_prefix}.AlphaZero.Arena.players"].__dict__[p2_name](manager, **kwargs)
arena_manager = self.game.make_fresh_instance()
arena_manager.set_headless(False)
arena = Arena(arena_manager, alpha_zero_config, self.device)
p1_w, p2_w, ds = arena.pit(p1, p2, num_games, alpha_zero_config.num_simulations, one_player=not switch_players,
start_player=starts, add_to_kwargs=kwargs)
print(f"Results: Player 1 wins: {p1_w}, Player 2 wins: {p2_w}, Draws: {ds}")
| 4,714 | Python | .py | 75 | 52.08 | 124 | 0.652193 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,506 | checkpointer.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/checkpointer.cpython-311.pyc | §
-*¢fê ã ól — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dlm Z d dl
m Z G d„ d¦ « Z dS )é N)ÚDotDict)Úfind_project_root)ÚConfigc óˆ — e Zd Zd$dedz deddfd„Zd%d„Z d&dej j d ej j d
ej
d e d e d
e
deddfd„Zdej j deddfd„Zdej j deddfd„Zdedefd„Zde defd„Zd%d„Zd%d„Zd%d„Zde fd„Zde fd„Zdefd„Zdefd „Zdefd „Zd „ Z d eddfd „Z d!„ Z d"e e fd#„Z dS )'Ú CheckPointerTÚcheckpoint_dirNÚverboseÚreturnc ó¼ — || _ | ¦ « | ¦ « | _ d| _ || _ t
j | j ¦ « d S )NÚ
improved_net_) Ú _CheckPointer__checkpoint_dirÚmake_dirÚinitialize_checkpoint_numÚ _CheckPointer__checkpoint_numÚ_CheckPointer__name_prefixr ÚatexitÚregisterÚcleanup)Úselfr r s úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/checkpointer.pyÚ__init__zCheckPointer.__init__ sR € Ø .ˆÔ Ø �
Š
‰ŒˆØ $× >Ò >Ñ @Ô @ˆÔ Ø ,ˆÔØ ˆŒ İŒ˜œ Ñ%Ô%Ğ%Ğ%Ğ%ó c ó® — | j � t j | j d¬¦ « d S t ¦ « }|› d�}|| _ t j |d¬¦ « d S )NT©Úexist_okz/Checkpoints/NetVersions)r
ÚosÚmakedirsr )r Úroot_dirr s r r zCheckPointer.make_dir sd € Ø Ô Ğ ,İ ŒK˜Ô-¸Ğ =Ñ =Ô =Ğ =Ø ˆFİ$Ñ&Ô&ˆØ $Ğ>Ğ>Ğ>ˆØ .ˆÔ İ
Œ �N¨TĞ2Ñ2Ô2Ğ2Ğ2Ğ2r ÚnetÚopponentÚ optimizerÚlrÚ iterationÚmu_alpha_zero_configÚnamec ó¾ — |€ | j t | j ¦ « z }| j › d|› d�}t j | ¦ « t |t j j ¦ « r| ¦ « n||||
¦ « | ¦ « dœ|¦ « | d|› d|› d�¦ « | xj dz
c_ d S )Nú/ú.pth)r r! r" r# ÚargsÚopponent_state_dictzSaved checkpoint to z at iteration ú.é ) r Ústrr r
ÚthÚsaveÚ
state_dictÚ
isinstanceÚoptimÚ OptimizerÚto_dictÚ
print_verbose) r r r r! r" r# r$ r% Úcheckpoint_paths r Úsave_checkpointz CheckPointer.save_checkpoint sü € ğ ˆ<ØÔ%¨DÔ,AÑ(BÔ(BÑBˆDà !Ô 2Ğ?Ğ?°TĞ?Ğ?Ğ?ˆİ
ŒØ—>’>Ñ#Ô#İ3=¸iÍÌÔI[Ñ3\Ô3\Ğk˜×-Ò-Ñ/Ô/Ğ/ĞbkØØ"Ø(×0Ò0Ñ2Ô2Ø#+×#6Ò#6Ñ#8Ô#8ğ
ğ
ğ ñ ô ğ ğ
×ÒĞ]°/Ğ]Ğ]ĞQZĞ]Ğ]Ğ]Ñ^Ô^Ğ^Ø Ğ Ô Ñ"Ğ Ô Ğ Ğ r c ó˜ — | j › d|› d�}t j | ¦ « |¦ « | d¦ « d S )Nr' r( z Saved state dict checkpoint.)r
r. r/ r0 r5 ©r r r% r6 s r Úsave_state_dict_checkpointz'CheckPointer.save_state_dict_checkpoint1 sR € Ø !Ô 2Ğ?Ğ?°TĞ?Ğ?Ğ?ˆİ
Œ�—’Ñ Ô /Ñ2Ô2Ğ2Ø ×ÒĞ:Ñ;Ô;Ğ;Ğ;Ğ;r c ó˜ — | j › d|› d�}| t j |¦ « ¦ « | d¦ « d S )Nr' r( z Loaded state dict checkpoint.)r
Úload_state_dictr. Úloadr5 r9 s r Úload_state_dict_checkpointz'CheckPointer.load_state_dict_checkpoint6 sT € Ø !Ô 2Ğ?Ğ?°TĞ?Ğ?Ğ?ˆØ ×Ò�BœG OÑ 4Ô 4Ñ5Ô5Ğ5Ø ×ÒĞ;Ñ<Ô<Ğ<Ğ<Ğ<r r6 c ó0 — t j d t j d< t j |¦ « }| d|› d|d › d�¦ « d|v r |d }nd }|d |d ||d
t |d ¦ « |d fS )
Nzmu_alpha_zero.mem_bufferÚ
mem_bufferzRestoring checkpoint z made at iteration r# r+ Úmemoryr r! r" r) r* )ÚsysÚmodulesr. r= r5 r )r r6 Ú
checkpointrA s r Úload_checkpoint_from_pathz&CheckPointer.load_checkpoint_from_path; s¬ € İ$'¤KĞ0JÔ$K�Œ �LÑ!İ”W˜_Ñ-Ô-ˆ
Ø ×ÒĞq°?ĞqĞqĞWaĞbmÔWnĞqĞqĞqÑrÔrĞrØ �zĞ !Ğ !Ø Ô)ˆFˆFàˆFؘ%Ô *¨[Ô"9¸6À:ÈdÔCSİ �J˜vÔ&Ñ 'Ô '¨Ğ4IÔ)JğKğ Kr Úcheckpoint_numc óT — | j › d| j › |› d�}| |¦ « S )Nr' r( )r
r rE )r rF r6 s r Úload_checkpoint_from_numz%CheckPointer.load_checkpoint_from_numF s7 € Ø !Ô 2Ğ]Ğ]°TÔ5GĞ]È Ğ]Ğ]Ğ]ˆØ×-Ò-¨oÑ>Ô>Ğ>r c óR — t d¦ « t d¦ « }|dk rt d¦ « d S t j | j ¦ « D ] }t j | j › d|› �¦ « Œ!t dt
t j | j ¦ « ¦ « › d�¦ « d S )NzClearing all checkpoints.zAre you sure?? (y/n): ÚyzAborted.r' zCleared z saved checkpoints (all).)ÚprintÚinputr Úlistdirr
ÚremoveÚlen)r ÚanswerÚ file_names r Úclear_checkpointsz CheckPointer.clear_checkpointsJ s² € õ Ğ)Ñ*Ô*Ğ*İĞ/Ñ0Ô0ˆØ �SŠ=ˆ=İ �*Ñ Ô Ğ Ø ˆFİœ DÔ$9Ñ:Ô:ğ >ğ >ˆIİ ŒI˜Ô.Ğ<Ğ<°Ğ<Ğ<Ñ =Ô =Ğ =Ğ =İ
ĞZ��RœZ¨Ô(=Ñ >Ô >Ñ?Ô?ĞZĞZĞZÑ[Ô[Ğ[Ğ[Ğ[r c óĞ — t j ¦ « }t j | j › d�d¬¦ « | j › d|› d�}t j | ¦ « |¦ « d S )Nz/TempTr ú/Temp/temp_net_r( )r Úgetpidr r
r. r/ r0 ©r r Ú process_pidr6 s r Úsave_temp_net_checkpointz%CheckPointer.save_temp_net_checkpointV si € İ”i‘k”kˆ İ
Œ �tÔ,Ğ3Ğ3Ğ3¸dĞCÑCÔCĞCØ !Ô 2ĞTĞTÀ;ĞTĞTĞTˆİ
Œ�—’Ñ Ô /Ñ2Ô2Ğ2Ğ2Ğ2r c ó” — t j ¦ « }| j › d|› d�}| t j |¦ « ¦ « d S )NrT r( )r rU r
r< r. r= rV s r Úload_temp_net_checkpointz%CheckPointer.load_temp_net_checkpoint\ sJ € İ”i‘k”kˆ Ø !Ô 2ĞTĞTÀ;ĞTĞTĞTˆØ ×Ò�BœG OÑ 4Ô 4Ñ5Ô5Ğ5Ğ5Ğ5r c ób — t d„ t j | j ¦ « D ¦ « ¦ « S )Nc ó< — g | ]}| d ¦ « ¯|‘ŒS )r( )Úendswith)Ú.0Úxs r ú
<listcomp>z:CheckPointer.initialize_checkpoint_num.<locals>.<listcomp>b s) € ĞWĞWĞW˜!ÀAÇJÂJÈvÑDVÔDVĞW�AĞWĞWĞWr )rO r rM r
©r s r r z&CheckPointer.initialize_checkpoint_numa s, € İĞWĞW�rœz¨$Ô*?Ñ @Ô @ĞWÑWÔWÑXÔXĞXr c ób — t d„ t j | j ¦ « D ¦ « ¦ « S )Nc ó� — g | ]C}t | d ¦ « d d¦ « d ¦ « ‘ŒDS )Ú_é r+ r )ÚintÚsplit)r^ rQ s r r` z;CheckPointer.get_highest_checkpoint_num.<locals>.<listcomp>e sD € ĞrĞrĞrÀ9•C˜ Ÿš¨Ñ,Ô,¨QÔ/×5Ò5°cÑ:Ô:¸1Ô=Ñ>Ô>ĞrĞrĞrr )Úmaxr rM r
ra s r Úget_highest_checkpoint_numz'CheckPointer.get_highest_checkpoint_numd s/ € İĞrĞrÕPRÔPZĞ[_Ô[pÑPqÔPqĞrÑrÔrÑsÔsĞsr c ó — | j › d�S )Nz/Temp/temp_net.pth©r
ra s r Ú
get_temp_pathzCheckPointer.get_temp_pathg s € ØÔ'Ğ;Ğ;Ğ;Ğ;r c ó — | j S ©Nrk ra s r Úget_checkpoint_dirz CheckPointer.get_checkpoint_dirj s € ØÔ$Ğ$r c ó� ‡ ‡— ˆˆ fd„t j ‰ j ¦ « D ¦ « }| d„ ¬¦ « |d S )Nc ób •— g | ]+}‰|v ¯t j ‰j |¦ « ‘Œ,S © )r ÚpathÚjoinr
)r^ r_ r% r s €€r r` z6CheckPointer.get_latest_name_match.<locals>.<listcomp>n s8 ø€ ĞwĞwĞwÀ1ĞmqĞuvĞmvĞmv�œŸ š TÔ%:¸AÑ>Ô>ĞmvĞmvĞmvr c ó@ — t j | ¦ « S rn )r rs Úgetctime)r_ s r ú<lambda>z4CheckPointer.get_latest_name_match.<locals>.<lambda>o s € ¬×(8Ò(8¸Ñ(;Ô(;€ r )Úkeyéÿÿÿÿ)r rM r
Úsort)r r% Ú name_matchess `` r Úget_latest_name_matchz"CheckPointer.get_latest_name_matchm sT øø€ ØwĞwĞwĞwĞwÍÌ
ĞSWÔShÑHiÔHiĞwÑwÔwˆ Ø×ÒĞ ;Ğ ;ĞÑ<Ô<Ğ<ؘBÔ Ğ r c ó — | j S rn )r ra s r Úget_name_prefixz CheckPointer.get_name_prefixr s € ØÔ!Ğ!r Úmsgc ó6 — | j rt |¦ « d S d S rn )r rK )r r s r r5 zCheckPointer.print_verboseu s% € Ø Œ<ğ İ �#‰JŒJˆJˆJˆJğ ğ r c óL — dd l }| | j › d�d¬¦ « d S )Nr z/Temp/T)Ú
ignore_errors)ÚshutilÚrmtreer
)r rƒ s r r zCheckPointer.cleanupy s3 € ؈
ˆ
ˆ
�
Š
˜Ô.Ğ6Ğ6Ğ6Àdˆ
ÑKÔKĞKĞKĞKr Úlossesc ó’ — t | j › d�d¦ « 5 }t j ||¦ « d d d ¦ « d S # 1 swxY w Y d S )Nz/training_losses.pklÚwb)Úopenr
ÚpickleÚdump)r r… Úfs r Ú save_losseszCheckPointer.save_losses} s‘ € İ
�TÔ*Ğ@Ğ@Ğ@À$Ñ
GÔ
Gğ #È1İ ŒK˜ Ñ "Ô "Ğ "ğ #ğ #ğ #ñ #ô #ğ #ğ #ğ #ğ #ğ #ğ #ğ #øøøğ #ğ #ğ #ğ #ğ #ğ #s ™<¼A ÁA )T)r
Nrn )!Ú__name__Ú
__module__Ú __qualname__r- Úboolr r r. ÚnnÚModuler2 Úfloatrf r r7 r: r> ÚtuplerE rH rR rX rZ r ri rl ro r| r~ r5 r ÚlistrŒ rr r r r r
sš € € € € € ğ&ğ & s¨T¡zğ &¸Dğ &ÈDğ &ğ &ğ &ğ &ğ3ğ 3ğ 3ğ 3ğ SWğ#ğ # 2¤5¤<ğ #¸2¼5¼<ğ #ĞTVÔT\ğ #Ø !ğ#à#&ğ#à>Dğ#àLOğ#à[_ğ#ğ #ğ #ğ #ğ$<¨b¬e¬lğ <À#ğ <È$ğ <ğ <ğ <ğ <ğ
=¨b¬e¬lğ =À#ğ =È$ğ =ğ =ğ =ğ =ğ
K¸ğ KÀğ Kğ Kğ Kğ Kğ?°sğ ?¸uğ ?ğ ?ğ ?ğ ?ğ
\ğ
\ğ
\ğ
\ğ3ğ 3ğ 3ğ 3ğ 6ğ 6ğ 6ğ 6ğ
Y¨3ğ Yğ Yğ Yğ Yğt¨Cğ tğ tğ tğ tğ<˜sğ <ğ <ğ <ğ <ğ% Cğ %ğ %ğ %ğ %ğ ¨#ğ ğ ğ ğ ğ
"ğ "ğ "ğ ğ ¨ğ ğ ğ ğ ğLğ Lğ Lğ# $ u¤+ğ #ğ #ğ #ğ #ğ #ğ #r r )
r r r‰ rB Útorchr. Ú mu_alpha_zero.AlphaZero.utilsr Úmu_alpha_zero.General.utilsr Úmu_alpha_zero.configr r rr r r ú<module>rš s§ ğØ
€
€
€
Ø € € € Ø
€
€
€
Ø
€
€
€
à Ğ Ğ Ğ à 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø 9Ğ 9Ğ 9Ğ 9Ğ 9Ğ 9Ø 'Ğ 'Ğ 'Ğ 'Ğ 'Ğ 'ğr#ğ r#ğ r#ğ r#ğ r#ñ r#ô r#ğ r#ğ r#ğ r#r | 11,468 | Python | .py | 35 | 326.571429 | 2,143 | 0.360679 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,507 | utils.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/utils.cpython-311.pyc | §
weŠfº* ã
ó4 — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dl m
Z
d dl m Z
d dlmZ d dlmZmZ G d„ de¦ « Zd ed
efd „Zd ej d
efd„Zd ej defd„Zd efd„Zd ej fd„Z dej dej d
ej fd„Z dej d
ej fd„Z defd„Z ded
efd„Z!d ed ed
efd „Z"d ed e#d!e#d"e#d#ef
d$„Z$d%ed
e
fd&„Z%d6d'ed
e&e
ej' j( ef fd(„Z)d)e#dedz fd*„Z*d+„ Z+d7d-ed.e,fd/„Z-d8d0e#d1edz fd2„Z.d3ed4ed efd5„Z/dS )9é N)Ú get_ipython)Ú AlphaZeroNet)ÚSAMPLE_AZ_ARGS)Ú MemBuffer)ÚConfigÚAlphaZeroConfigc ó — e Zd Zd„ Zd„ ZdS )ÚDotDictc ó — | | S ©N© )ÚselfÚnames úL/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/utils.pyÚ __getattr__zDotDict.__getattr__ s € Ø�DŒzĞó c ó — || |< d S r r
)r r Úvalues r Ú __setattr__zDotDict.__setattr__ s € ؈ˆT‰
ˆ
ˆ
r N)Ú__name__Ú
__module__Ú __qualname__r r r
r r r
r
s2 € € € € € ğğ ğ ğğ ğ ğ ğ r r
Úgame_experienceÚreturnc ó& — g }| D �]Š\ }}}}t j d„ | ¦ « D ¦ « ¦ « }| |||f¦ « t ddgddg¦ « D �]$\ }}t j | ¦ « |¬¦ « } t j | ¦ « ||¦ « |¬¦ « ¦ « }
| | |
|f¦ « ~ ~
t j | ¦ « |¬¦ « } t j | ¦ « ||¦ « |¬¦ « ¦ « }
| | |
|f¦ « �Œ&�ŒŒ|S )Nc ó — g | ]}|‘ŒS r
r
)Ú.0Úxs r ú
<listcomp>z6augment_experience_with_symmetries.<locals>.<listcomp> s € Ğ.Ğ.Ğ.˜Q�qĞ.Ğ.Ğ.r r é é ©Úk©Úaxis)
ÚnpÚarrayÚvaluesÚappendÚzipÚrot90ÚcopyÚreshapeÚflattenÚflip) r Ú
board_sizeÚgame_experience_ÚstateÚpiÚvÚ_r% r# Ústate_Úpi_s r Ú"augment_experience_with_symmetriesr8 sw € ØĞØ*ğ
6ñ
6‰ˆˆr�1�aİ
ŒXĞ.Ğ. "§)¢)¡+¤+Ğ.Ñ.Ô.Ñ
/Ô
/ˆØ× Ò ¨¨A Ñ/Ô/Ğ/ݘA˜q˜6 A q 6Ñ*Ô*ğ 6ñ 6‰GˆD�!İ”X˜eŸjšj™lœl¨aĞ0Ñ0Ô0ˆFİ”(˜2Ÿ7š7™9œ9×,Ò,¨Z¸ÑDÔDÈĞJÑJÔJ×RÒRÑTÔTˆCØ × #Ò # V¨S°!Ğ$4Ñ 5Ô 5Ğ 5ؘݔW˜UŸZšZ™\œ\°Ğ5Ñ5Ô5ˆFİ”'˜"Ÿ'š'™)œ)×+Ò+¨J¸
ÑCÔCÈ$ĞOÑOÔO×WÒWÑYÔYˆCØ × #Ò # V¨S°!Ğ$4Ñ 5Ô 5Ğ 5Ñ 5ñ 6ğ Ğr r2 r# c ó~ — t | j d ¦ « D ]!}t j | | |¬¦ « | |< Œ"| S )Nr r" )ÚrangeÚshaper& r+ )r2 r# Údims r Ú rotate_stackr= * sB € İ�U”[ ”^Ñ$Ô$ğ /ğ /ˆİ”X˜e Cœj¨AĞ.Ñ.Ô.ˆˆc‰
ˆ
Ø €Lr r% c ó~ — t | j d ¦ « D ]!}t j | | |¬¦ « | |< Œ"| S )Nr r$ )r: r; r& r/ )r2 r% r< s r Ú
flip_stackr? 0 sB € İ�U”[ ”^Ñ$Ô$ğ 4ğ 4ˆİ”W˜U 3œZ¨dĞ3Ñ3Ô3ˆˆc‰
ˆ
Ø €Lr c ór — g }| D ]1\ }}}}}t |¦ « }| |||||f¦ « Œ2|S r )Úmake_channels_from_singler) )r Ú
experiencer2 r3 r4 Úcurrent_playerÚmoves r Ú
make_channelsrE 6 sX € Ø€JØ.=ğ @ğ @Ñ*ˆˆr�1�n dİ)¨%Ñ0Ô0ˆØ×Ò˜5 " a¨ ¸Ğ>Ñ?Ô?Ğ?Ğ?à Ğr c óÒ — t j | dk dd¦ « }t j | dk dd¦ « }t j | dk dd¦ « }t j | |||gd¬¦ « S )Nr r éÿÿÿÿr$ )r& ÚwhereÚstack)r2 Úplayer_one_stateÚplayer_minus_one_stateÚ empty_states r rA rA ? sj € İ”x ¨¢
¨A¨qÑ1Ô1Ğİ œX e¨r¢k°1°aÑ 8Ô 8Ğİ”(˜5 Aš: q¨!Ñ,Ô,€Kİ
Œ8�UĞ ,Ğ.DÀkĞRĞYZĞ [Ñ [Ô [Ğ[r Ú
probabilitiesÚmaskc óŒ — d}| d|dz ¦ « | d|dz ¦ « z }| ¦ « }|dk r=|d|› �z
}t j |j dt j |j ¦ « z ¦ « }n||z }t
|¦ « dk r t |t dd¦ « ¬ ¦ « |S )
NÚ rG é r zISum of valid probabilities is 0. Creating a uniform probability...
Mask:
g ğ?zmasking_message.txtÚw)Úfile) r- Úsumr& Úfullr; ÚprodÚlenÚprintÚopen)rM rN r0 Úto_printÚvalidsÚ
valids_sums r Úmask_invalid_actionsr] F sÊ € Ø€Hğ ×
"Ò
" 2 z°Q¡Ñ
7Ô
7¸$¿,º,ÀrÈ:ĞYZÉ?Ñ:[Ô:[Ñ
[€FØ—’‘ ” €JØ�Q‚€ğ ĞhĞbfĞhĞhÑhˆİ”˜œ sR¬W°V´\Ñ-BÔ-BÑ'BÑCÔCˆˆà˜*Ñ$ˆå
ˆ8�}„}�qÒĞİ
ˆh�TĞ"7¸Ñ =Ô =Ğ>Ñ>Ô>Ğ>Ø €Mr Ústatesc óĞ — g }| D ]�}| ¦ « ¦ « ¦ « }t j |dk d|¦ « }t j |dk d|¦ « }t j |dk d|¦ « }| |¦ « Œ�t
j t j |¦ « t j ¬¦ «
d¦ « S )Nr éûÿÿÿr ©Údtype) ÚdetachÚcpuÚnumpyr& rH r) ÚthÚtensorr' Úfloat32Úsqueeze)r^ Úmasksr2 Únp_staterN s r Úmask_invalid_actions_batchrl Z s € Ø €EØğ ğ ˆØ—<’<‘>”>×%Ò%Ñ'Ô'×-Ò-Ñ/Ô/ˆİŒx˜ Aš
r¨8Ñ4Ô4ˆİŒx˜ š 1 dÑ+Ô+ˆİŒx˜ š
A tÑ,Ô,ˆØ
� Š �TÑÔĞĞå
Œ9•R”X˜e‘_”_B¬JĞ 7Ñ 7Ô 7× ?Ò ?ÀÑ BÔ BĞBr Úargsc óL — g d¢}|D ] }|| vrt d|› d|› d�¦ « ‚Œ d S )N)Únum_net_channelsÚnum_net_in_channelsÚ net_dropoutÚnet_action_sizeÚnum_simulationsÚself_play_gamesÚ num_itersÚepochsÚlrÚmax_buffer_sizeÚ
num_pit_gamesÚrandom_pit_freqr0 Ú
batch_sizeÚtauÚcÚcheckpoint_dirÚupdate_thresholdz Missing key z? in args dict. Please supply all required keys.
Required keys: ú.)ÚKeyError)rm Ú
required_keysÚkeys r Ú
check_argsr„ f st € ğvğ vğ v€Mğ ğ ?ğ ?ˆØ �dˆ?ˆ?İğ >¨#ğ >ğ >Ø-:ğ >ğ >ğ >ñ ?ô ?ğ
?ğ ğ?ğ ?r Únc ó� ‡— t | ‰¦ « | dz dz z dt ˆfd„t ‰| ¦ « D ¦ « ¦ « z z S )NrQ é c ó0 •— g | ]}t |‰¦ « ‘ŒS r
)Ú get_num_horizontal_conv_slides)r r r# s €r r z1calculate_board_win_positions.<locals>.<listcomp>s s$ ø€ ĞCĞCĞC°!Õ '¨¨1Ñ -Ô -ĞCĞCĞCr )r‰ rT r: )r… r# s `r Ú calculate_board_win_positionsrŠ q sa ø€ İ )¨!¨QÑ /Ô /°1°q±5¸1±9Ñ =ÀÅCØCĞCĞCĞCµu¸QÀ±{´{ĞCÑCÔCñEEô EEñ AEñ Eğ Er r0 Ú kernel_sizec ó — | |z
dz S )Nr r
)r0 r‹ s r r‰ r‰ v s € Ø ˜Ñ $¨Ñ )Ğ)r Ún_trialsÚ
init_net_pathÚstorageÚ
study_nameÚconfigc ó¦ ‡‡‡‡ ‡
— ˆˆ ˆˆˆ
fd„}ddl mŠ ddlmŠ |Š
d‰
_ t j ||¬¦ « }| || ¬¦ « dS ) aœ
Performs a hyperparameter search using optuna. This method is meant to be called using the start_jobs.py script.
For this method to work, a mysql database must be running on the storage address and an optuna study with the
given name and the 'maximize' direction must exist.
:param n_trials: num of trials to run the search for.
:param init_net_path: The path to the initial network to use for all trials.
:param storage: The mysql storage string. Specifies what database to use.
:param study_name: Name of the study to use.
:param game: The game instance to use.
:param config: The config to use for the search.
:return:
c óX •— | ddd¦ « }| ddd¦ « }| ddd ¦ « }| d
d d d
¬¦ « }| ddd¦ « }| dd d¦ « }| ddd¦ « }| dddd
¬¦ « }|‰_ |‰_ |‰_ |‰_ |‰_ |‰_ |‰_ d‰_ |‰_
‰ ‰ ¦ « ‰¦ « } ‰
‰‰‰| ¦ « }
t d| j › d�¦ « |
¦ « |
¦ « } | | | j ¦ « t d| j › d| › d�¦ « ~
| S ) NÚnum_mc_simulationsé< i@ Únum_self_play_gamesé2 éÈ Ú
num_epochséd i� rw g-C ëâ6?g{®Gáz„?T)ÚlogÚtempg à?g ø?Ú
arena_tempÚcpucté Ú log_epsilong»½×Ùß|Û=gH¯¼šò×z>zTrial z started.z finished with win freq r€ )Ú suggest_intÚ
suggest_floatrs rt rv rw r| r} Ú arena_tauru r Úmake_fresh_instanceÚfrom_state_dictrX ÚnumberÚtrainÚ get_arena_win_frequencies_meanÚreport)Útrialr” r– r™ rw rœ r� r� r Ú search_treeÚtrainerÚwin_freqÚ McSearchTreeÚTrainerÚgamer� Ú trial_configs €€€€€r Ú objectivez-az_optuna_parameter_search.<locals>.objective‰ sÁ ø€ Ø "× .Ò .Ğ/CÀRÈÑ NÔ NĞØ #× /Ò /Ğ0EÀrÈ3Ñ OÔ OĞØ×&Ò& |°S¸#Ñ>Ô>ˆ
Ø
×
Ò
t¨T°tĞ
Ñ
<Ô
<ˆØ×"Ò" 6¨3°Ñ4Ô4ˆØ×(Ò(¨ °t¸SÑAÔAˆ
Ø×#Ò# G¨S°!Ñ4Ô4ˆØ×)Ò)¨-¸ÀÈ$Ğ)ÑOÔOˆ à'9ˆ Ô$Ø':ˆ Ô$Ø (ˆ ÔØ ˆ ŒØ ˆ ÔØ ˆ ŒØ!+ˆ Ô Ø!"ˆ Ô Ø#.ˆ Ô Ø"�l 4×#;Ò#;Ñ#=Ô#=¸|ÑLÔLˆ Ø×)Ò)¨-¸ ÀtÈ[ÑYÔYˆİ
Ğ.�u”|Ğ.Ğ.Ğ.Ñ/Ô/Ğ/Ø�
Š
‰ŒˆØ×9Ò9Ñ;Ô;ˆØ
� Š �X˜uœ|Ñ,Ô,Ğ,İ
ĞH�u”|ĞHĞH¸XĞHĞHĞHÑIÔIĞIØ Øˆr r )r¯ )r® F)r� r� )r� N)Úmu_alpha_zero.trainerr¯ Ú+mu_alpha_zero.AlphaZero.MCTS.az_search_treer® Ú show_tqdmÚoptunaÚ
load_studyÚoptimize) r� r� r� r� r° r‘ r² Ústudyr® r¯ r± s ` ` @@@r Úaz_optuna_parameter_searchrº z s˜ øøøøø€ ğ ğ ğ ğ ğ ğ ğ ğ ğ ğ: .Ğ-Ğ-Ğ-Ğ-Ğ-ØHĞHĞHĞHĞHĞHØ€LØ "€LÔİ Ô ¨¸WĞ EÑ EÔ E€EØ ‡N‚N�9 x€NÑ0Ô0Ğ0Ğ0Ğ0r Ú
muzero_configc ó„ — t | j | j | j | j | j ¦ « }| |¦ « S r )r rp ro rq rr Úaz_net_linear_input_sizeÚto)r» ÚdeviceÚnetworks r Úbuild_net_from_configrÁ ® sB € ݘ=Ô<¸mÔ>\Ø(Ô4°mÔ6SØ(ÔAñCô C€Gğ �:Š:�fÑ Ô Ğ r Úmuzero_alphazero_configc óÖ — |€| j }|€| j }t | |¦ « }t j | ¦ « |¬¦ « }t |¬¦ « }|||fS )N)rw )Úmax_size)rw rx rÁ rf ÚoptimÚAdamÚ
parametersr )r r¿ rw Ú buffer_sizerÀ Ú optimizerÚmemorys r Úbuild_all_from_configrË µ sm € à €zØ
$Ô
'ˆØĞØ-Ô=ˆ İ#Ğ$;¸VÑDÔD€Gİ”—
’
˜g× 0Ò 0Ñ 2Ô 2°r�
Ñ:Ô:€Iİ
Ğ
,Ñ
,Ô
,€FØ �I˜vĞ %Ğ%r Úcheckpoint_pathc ó — |€t t ¦ « }t j t j ¦ « rdnd¦ « }t
||¦ « }t j | ¦ « }| |d ¦ « |S )NÚcudard Únet) r
Ú test_argsrf r¿ rÎ Ú is_availablerÁ ÚloadÚload_state_dict)rÌ rm r¿ rÏ Údatas r Úmake_net_from_checkpointrÕ Á su € Ø €|İ•yÑ!Ô!ˆİ
ŒY¥¤×!5Ò!5Ñ!7Ô!7ĞB�v�v¸UÑ
CÔ
C€Fİ
fÑ
-Ô
-€Cİ
Œ7�?Ñ #Ô #€DØ×Ò˜˜Uœ Ñ$Ô$Ğ$Ø €Jr c óŠ — t ¦ « j j } | dk rdS | dk rdS | dk rdS dS # t $ r Y dS w xY w)NÚZMQInteractiveShellTÚShellÚTerminalInteractiveShellF)r Ú __class__r Ú NameError)Úshells r Ú is_notebookrİ Ë so € ğ İ‘
”
Ô'Ô0ˆØ Ğ)Ò )Ğ )Ø�4Ø
�gÒ
Ğ
Ø�4Ø
Ğ0Ò
0Ğ
0Ø�5à�5øİ ğ ğ ğ ؈uˆuğøøøs ‚ 4 ¢4 ª4 ´
AÁAFÚfilesÚnot_notebook_okc ó( — t ¦ « }|s|rd S |st d¦ « ‚| D ]i}t j d¦ « s3t j d¦ « rt j d¦ « t
j |d¦ « Œjd S )Nz2This method should only be called from a notebook.z"/content/drive/MyDrive/Checkpointsz/content/drive/MyDrive)rİ Ú RuntimeErrorÚosÚpathÚexistsÚmkdirÚshutilr, )rŞ rß Úis_nbtrS s r Úupload_checkpoint_to_gdriverè Ú s¨ € İ
‰]Œ]€FØ ğ �oğ ØˆØ ğ QİĞOÑPÔPĞPàğ @ğ @ˆİŒw�~Š~Ğ BÑCÔCğ ;ÍÌÏÊĞWoÑHpÔHpğ ;İ ŒHĞ9Ñ :Ô :Ğ :İŒ �DĞ>Ñ?Ô?Ğ?Ğ?ğ@ğ @r Úoutput_file_nameÚ depth_limitc ó\ ‡‡— t ¦ « }d|j d< d|j d< d|j d< d|j d< ‰€t d¦ « Šd t j d
t fˆˆfd „ Š ‰| | |‰¬ ¦ « | d
¬¦ « | |› d�¦ « d S )NzMCTS visualizationÚlabelÚcircler; ÚblueÚcolorÚgoldÚinfÚgÚd_limitc
óú •— d }| j €:t t j dd|j j ¬¦ « ¦ « }nt | j ¦ « }| |¦ « || k r(| t |j ¦ « |¦ « | ¦ « r|dk rd S | j
¦ « D ])} ‰|| |‰t d¦ « k r|dz
n‰¬¦ « Œ*d S )Nr rŸ )ÚlowÚhighÚsizerñ r ©ró ) r2 Ústrr& ÚrandomÚrandintr; Úadd_nodeÚadd_edgeÚ was_visitedÚchildrenr( Úfloat)ÚnodeÚparentrò ró r6 Úchildrê Ú
make_graphs €€r r z"visualize_tree.<locals>.make_graphğ s ø€ àˆØ Œ:Ğ İ�œ×*Ò*¨q°q¸v¼|Ô?QĞ*ÑRÔRÑSÔSˆFˆF嘜‘_”_ˆFØ �
Š
�6ÑÔĞØ �TŠ>ˆ>Ø
�JŠJ•s˜6œ<Ñ(Ô(¨&Ñ 1Ô 1Ğ 1Ø× Ò Ñ!Ô!ğ W°¢\ \Ø ˆFğ ”]×)Ò)Ñ+Ô+ğ lğ lˆEØ ˆJ�u˜d A¸kÍUĞSXÉ\Ì\Ò>YĞ>Y¨w¸©{¨{Ğ_jĞ kÑ kÔ kĞ kĞ kğ lğ lr rø Údot)Úprogz.png) Ú
pygraphvizÚAGraphÚ
graph_attrÚ node_attrÚ edge_attrr ÚintÚlayoutÚdraw)Ú root_noderé rê Úgraphr s ` @r Úvisualize_treer ç så øø€ İ × Ò Ñ Ô €EØ 4€EÔ�WÑ Ø '€E„O�GÑ Ø %€E„O�GÑ Ø %€E„O�GÑ ØĞݘE‘l”lˆ ğl¥JÔ$5ğ lÅğ lğ lğ lğ lğ lğ lğ lğ* €Jˆy˜) U°KĞ@Ñ@Ô@Ğ@Ø ‡L‚L�e€LÑ Ô Ğ Ø ‡J‚JĞ"Ğ(Ğ(Ğ(Ñ)Ô)Ğ)Ğ)Ğ)r rÔ rÊ c óú — | D ]w}|D ]r\ }}}t j |t j ¬¦ « ||¦ « }t j |t j ¬¦ « }| |||f¦ « ŒsŒxd S )Nra )rf rg rh r- Úadd)rÔ rÊ r0 Ú game_datar2 r3 r4 s r Úcpp_data_to_memoryr
s� € ğ ğ 'ğ 'ˆ Ø %ğ 'ğ '‰LˆE�2�qİ”I˜e2¬:Ğ6Ñ6Ô6×>Ò>¸zÈ:ÑVÔVˆEİ”˜2¥R¤ZĞ0Ñ0Ô0ˆBØ �JŠJ˜˜r 1�~Ñ &Ô &Ğ &Ğ &ğ 'ğ'ğ 'r )NN)Fr )0râ ræ re r& r¶ Útorchrf ÚIPythonr Ú$mu_alpha_zero.AlphaZero.Network.nnetr Ú!mu_alpha_zero.AlphaZero.constantsr rĞ Úmu_alpha_zero.mem_bufferr Úmu_alpha_zero.configr r Údictr
Úlistr8 Úndarrayr r= r? rE rA r] rg rl r„ rŠ r‰ rù rº rÁ ÚtuplerÅ Ú OptimizerrË rÕ rİ Úboolrè r r r
r r ú<module>r" s� ğØ € € € Ø
€
€
€
Ø Ğ Ğ Ğ Ø
€
€
€
à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ ğ >Ğ =Ğ =Ğ =Ğ =Ğ =Ø IĞ IĞ IĞ IĞ IĞ IØ .Ğ .Ğ .Ğ .Ğ .Ğ .Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8ğğ ğ ğ ğ ˆdñ ô ğ ğ ¸ğ ÈTğ ğ ğ ğ ğ"˜œ
ğ sğ ğ ğ ğ ğ �b”jğ ¨ğ ğ ğ ğ ğ 4ğ ğ ğ ğ ğ\ R¤Zğ \ğ \ğ \ğ \ğ¨¬
ğ ¸"¼*ğ ĞUWÔU_ğ ğ ğ ğ ğ( C r¤yğ C°R´Yğ Cğ Cğ Cğ Cğ?�Tğ ?ğ ?ğ ?ğ ?ğE Sğ E¨Sğ Eğ Eğ Eğ Eğ
*¨sğ *Àğ *Èğ *ğ *ğ *ğ *ğ11¨ğ 11¸Sğ 11È3ğ 11Ğ\_ğ 11Ğo~ğ 11ğ 11ğ 11ğ 11ğh ¨ğ ¸Lğ ğ ğ ğ ğ &ğ &°6ğ &ĞafØ�"”(Ô$ iĞ/ôb1ğ &ğ &ğ &ğ &ğ¨cğ ¸À4¹ ğ ğ ğ ğ ğ ğ ğ ğ
@ğ
@ tğ
@¸dğ
@ğ
@ğ
@ğ
@ğ *ğ *°ğ *À#ÈÁ*ğ *ğ *ğ *ğ *ğF'˜Tğ '¨9ğ 'À#ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'r | 17,643 | Python | .py | 104 | 168.125 | 1,037 | 0.351559 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,508 | alpha_zero.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/alpha_zero.cpython-311.pyc | §
weŠf¤ ã óÀ — d dl Z d dlmZ d dlZd dlZd dlmZ d dl m
Z
d dl m Z d dl
mZ d dlmZ d dlmZ d d lmZ d d
lmZmZ d d lmZ d d lmZ G d
„ d¦ « Z dS )é N)ÚType)ÚArena)Ú NetPlayer)Ú McSearchTree)ÚTrainer)Ú
AlphaZeroGame)ÚGeneralMemoryBuffer)ÚGeneralNetwork)Ú net_not_noneÚfind_project_root)Ú HookManager)ÚAlphaZeroConfigc
óÒ — e Zd Zdefd„Z d dedee ded e d
e
pdd e f d „Z d dee d
e de d e d
e
pdd e f d„Z
d„ Zd dej dedefd„Z d de de dededede f d „ZdS ) Ú AlphaZeroÚ
game_instancec ó¾ — d | _ d | _ || _ t j t j ¦ « rdnd¦ « | _ d | _ d | _ d S )NÚcudaÚcpu) ÚtrainerÚnetÚgameÚthÚdevicer Ú is_availableÚalpha_zero_configÚtree)Úselfr s úQ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/alpha_zero.pyÚ__init__zAlphaZero.__init__ sR € ؈Œ ؈ŒØ!ˆŒ Ý”i"¬'×*>Ò*>Ñ*@Ô*@Ð K ÀeÑLÔLˆŒ Ø26ˆÔ Ø"&ˆŒ ˆ ˆ ó TNFr Ú
network_classÚmemoryÚheadlessÚ hook_managerÚcheckpointer_verbosec ó¦ — | ||¬¦ « | j ¦ « }t | j ¦ « |¦ « }|| _ t | j ¦ « fi ||dœ¤Ž} || _ t j
|| j ||| ||||¬¦ « | _ | j ¦ « | _
d S )N)r$ )ÚnetworkÚmonte_carlo_tree_search)r# r% Úmemory_overrider$ )Úmake_from_configÚtor r r Úmake_fresh_instancer r r r Úcreater Ú get_networkr )
r r r! r" r# r$ r% r' r Ú
net_players
r Ú
create_newzAlphaZero.create_new sÙ € ð ×0Ò0Ð1BÐQ]Ð0Ñ^Ô^×aÒaÐbfÔbmÑnÔnˆÝ˜DœI× 9Ò 9Ñ ;Ô ;Ð=NÑOÔOˆØˆŒ Ý ˜tœy× <Ò <Ñ >Ô >ÐxÐxÈgÐrvÐBwÐBwÐxÐxˆ
Ø!2ˆÔ Ý ”~Ð&7¸¼ÀGÈTÐS]ÐhpØ;OÐagØ3?ðAñ Aô AˆŒ ð ”<×+Ò+Ñ-Ô-ˆŒˆˆr ÚpathÚcheckpoint_dirc ó — t j |t t ||| j |||¬¦ « | _ | j ¦ « | _ | j ¦ « | _ | j
¦ « | _ d S )N)r# r$ r% ) r Úfrom_checkpointr r r r r. r Úget_treer Úget_argsÚargs)r r! r1 r2 r# r$ r% s r Úload_checkpointzAlphaZero.load_checkpoint) s € õ Ô.¨}½lÍIÐW[Ð]kÐmqÔmvØ8@È|ØDXðZñ Zô ZˆŒ ð ”<×+Ò+Ñ-Ô-ˆŒØ”L×)Ò)Ñ+Ô+ˆŒ Ø”L×)Ò)Ñ+Ô+ˆŒ ˆ ˆ r c ó` — t | j ¦ « | j ¦ « d S )N)r r r Útrain)r s r r: zAlphaZero.train3 s, € Ý�T”XÑ Ô Ð Ø Œ ×ÒÑ Ô Ð Ð Ð r r ÚxÚtauÚreturnc ój — t | j ¦ « |j | j d | j d | j d fk s J dt |j ¦ « z ¦ « ‚| j | j |d| j |¬¦ « \ }}| j || j
j ¬¦ « S )NÚ
board_sizeÚnum_net_in_channelszXInput shape is not correct. Expected (board_size, board_size, num_net_in_channels).Got: é )r< ) r r Úshaper7 Ústrr Úsearchr r Ú select_mover r< )r r; r< ÚpiÚ_s r ÚpredictzAlphaZero.predict7 s¹ € Ý�T”XÑ Ô Ð ØŒw˜4œ9 \Ô2°D´I¸lÔ4KÈTÌYØ !ôN#ð $ò $ð $ð $ð&-Ý/2°1´7©|¬|ñ&<ñ$ô $ð $ð ” × Ò ¤¨1¨a°´À#Ð ÑFÔF‰ˆˆAØŒy×$Ò$ R¨TÔ-CÔ-GÐ$ÑHÔHÐHr rA Úp1_nameÚp2_nameÚ num_gamesÚstartsÚswitch_playersc ó0 — t | j ¦ « | j | j ¦ « | j ¦ « | j ¦ « }t ||¦ « }| j ||j |j ddœ} t ¦ « dd¦ « d¦ « d }
t j |
› d� j | |fi | ¤Ž} t j |
› d� j | |fi | ¤Ž} | j ¦ « }
|
d¦ « t# |
|| j ¦ « }| | | ||j | || ¬¦ « \ }}}t) d|› d |› d
|› �¦ « d S ) Néÿÿÿÿ)r' r( Ú evaluate_fnÚdepthÚplayerú\ú/z.AlphaZero.Arena.playersF)Ú
one_playerÚ start_playerÚ
add_to_kwargszResults: Player 1 wins: z, Player 2 wins: z , Draws: )r r r+ r Úevalr r, r Ú
eval_boardÚ
minimax_depthr ÚreplaceÚsplitÚsysÚmodulesÚ__dict__Ú set_headlessr ÚpitÚnum_simulationsÚprint)r rI rJ rK r rL rM Úmanagerr ÚkwargsÚ path_prefixÚp1Úp2Ú
arena_managerÚarenaÚp1_wÚp2_wÚdss r ÚplayzAlphaZero.play? s¨ € å�T”XÑ Ô Ð Ø Œ� Š �D”KÑ Ô Ð Ø Œ�
Š
‰ŒˆØ”)×/Ò/Ñ1Ô1ˆÝ˜GÐ%6Ñ7Ô7ˆØ !œXÀ$ÐW^ÔWiØ,Ô:ÀbðJð Jˆå'Ñ)Ô)×1Ò1°$¸Ñ<Ô<×BÒBÀ3ÑGÔGÈÔKˆ Ý
Œ[˜KÐAÐAÐAÔ
BÔ
KÈGÔ
TÐU\Ð
gÐ
gÐ`fÐ
gÐ
gˆÝ
Œ[˜KÐAÐAÐAÔ
BÔ
KÈGÔ
TÐU\Ð
gÐ
gÐ`fÐ
gÐ
gˆØ œ ×5Ò5Ñ7Ô7ˆ
Ø×"Ò" 5Ñ)Ô)Ð)Ý�mÐ%6¸¼ ÑDÔDˆØ Ÿš 2 r¨9Ð6GÔ6WÐhvÐdvØ06Àfð #ñ Nô N‰ˆˆd�Bå
ÐS¨ÐSÐSÀÐSÐSÈrÐSÐSÑTÔTÐTÐTÐTr )TNF)r )rA T)Ú__name__Ú
__module__Ú __qualname__r r r r r
r Úboolr
r0 rC r8 r: ÚnpÚndarrayÚfloatÚintrH rn © r r r r s‹ € € € € € ð' mð 'ð 'ð 'ð 'ð lpØ05ð .ð .¨Oð .ÈDÐQ_ÔL`ð .Ø.ð .Ø:>ð .ØU`ÐUhÐdhð .à)-ð .ð .ð .ð .ð TXØ5:ð,ð ,¨T°.Ô-Að ,Èð ,Ð^að ,Ø"&ð,Ø=HÐ=PÈDð,à.2ð,ð ,ð ,ð ,ð ð ð ðIð I˜œð I¨%ð I¸ð Ið Ið Ið Ið rsØ$(ðUð U˜Cð U¨#ð U¸#ð UÐRað UÐknð UØ !ðUð Uð Uð Uð Uð Ur r ) r] Útypingr Únumpyrs Útorchr Ú#mu_alpha_zero.AlphaZero.Arena.arenar Ú%mu_alpha_zero.AlphaZero.Arena.playersr Ú+mu_alpha_zero.AlphaZero.MCTS.az_search_treer Úmu_alpha_zero.trainerr Ú mu_alpha_zero.General.az_gamer Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr
Úmu_alpha_zero.General.utilsr r Ú mu_alpha_zero.Hooks.hook_managerr
Úmu_alpha_zero.configr r rw r r ú<module>r… s. ðØ
€
€
€
Ø Ð Ð Ð Ð Ð à Ð Ð Ð Ø Ð Ð Ð à 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø ;Ð ;Ð ;Ð ;Ð ;Ð ;Ø DÐ DÐ DÐ DÐ DÐ DØ )Ð )Ð )Ð )Ð )Ð )Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7Ø <Ð <Ð <Ð <Ð <Ð <Ø 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø GÐ GÐ GÐ GÐ GÐ GÐ GÐ GØ 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø 0Ð 0Ð 0Ð 0Ð 0Ð 0ð=Uð =Uð =Uð =Uð =Uñ =Uô =Uð =Uð =Uð =Ur | 7,308 | Python | .py | 31 | 234.677419 | 1,057 | 0.377851 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,509 | constants.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/constants.cpython-311.pyc | §
føf!
ã
ó — d Z i dd“dd“dd“de dz “d d
“d d “d
d“dd“dd“dd“dd“dd“de “dd“d d“d d “d d“d d!d"d#d$d$d%d$d&d'd d(œ ¥Zi dd“dd“dd“de dz “d d
“d d“d
d“dd)“dd“dd“dd“dd“de “dd*“d d+“d d,“d d-“d d!d"d#d$d"d.d'd d/œ ¥Zi dd“d0d“dd“dd“dd1“d2d3“d d4“d d$“d5d$“d6d7“d8d9“d:d"“d;d<“d
d“dd=“dd>“dd?“i dd“dd“de “dd@“d d“d dA“d d“dBdC“dDdE“dFd “dGd!“dHd “dIdJ“dKd$“dLd$“dMd%“dNd$“¥ddOdPdJdQdRd'd dSœ¥Zd S )Té Únum_net_channelsi Únum_net_in_channelsé Ú net_dropoutg333333Ó?Únet_action_sizeé Únum_simulationsi% Úself_play_gamesi, Ú num_itersé2 Úepochsiô Úlrg{äL?´œj?Úmax_buffer_sizei † Ú
num_pit_gamesé( Úrandom_pit_freqé Ú
board_sizeÚ
batch_sizeé ÚtauÚ arena_taugëö‹ÓG1¥?ÚcNg333333ã?é Té g•Ö&è .>i H ÚLogs) Úcheckpoint_dirÚupdate_thresholdÚ
minimax_depthÚ show_tqdmÚ num_workersÚ
num_to_winÚ log_epsilonÚzero_tau_afterÚaz_net_linear_input_sizeÚlog_dirÚpushbullet_tokeni@ é€ g ğ?é gffffffò?g»"¸hkN>) r r r r r! r" r# r& r' Únum_net_out_channelsé Únet_latent_sizeé$ éğ ÚKÚgammag�•C‹lçï?Úframe_buffer_sizeé Ú
frame_skipÚ num_stepsi� éd gü©ñÒMbP?ip éÿ g{®Gáz„?Úc2iÄL Úalphagš™™™™™é?r r r r Fr! r" r# r$ zALE/Asteroids-v5z Pickles/Data)é` r9 i )ÚbetaÚenv_idÚ
pickle_dirÚuse_javaÚtarget_resolutionr% r& r' )r ÚSAMPLE_AZ_ARGSÚTRAINED_AZ_NET_ARGSÚSAMPLE_MZ_ARGS© ó úP/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/constants.pyú<module>rE sî ğğ €
ğ ؘğ à˜1ğ ğ �3ğ ğ �z Q‘ğ ğ
�tğ ğ �sğ
ğ �ğ ğ
ˆcğ ğ Ğ
ğ ğ �wğ ğ �Rğ ğ �qğ ğ �*ğ ğ �#ğ ğ
ˆ1ğ ğ Ğ$ğ! ğ" ˆğ# ğ$ ØØØØØØØØ %ØØ ğ9 ğ ğ €ğ>ؘğà˜1ğğ �3ğğ �z Q‘ğ ğ
�tğ ğ �qğ
ğ �ğğ
ˆcğğ Ğ
ğğ �wğğ �Rğğ �qğğ �*ğğ �#ğ ğ
ˆ3ğ ğ �ğ!ğ" ˆğ#ğ$ ØØØØØØ)ØØ ğ5ğ ğ Ğ ğ:+ؘğ+à˜Cğ+ğ ˜1ğ+ğ �3ğ +ğ
�rğ +ğ �rğ
+ğ �sğ+ğ �qğ+ğ ˆğ+ğ ˆUğ+ğ ˜ğ+ğ �!ğ+ğ �ğ+ğ �ğ +ğ
ˆcğ +ğ ˆ%ğ!+ğ" �vğ#+ğ +ğ$ �Rğ%+ğ& �qğ'+ğ( �*ğ)+ğ* �#ğ++ğ,
ˆ1ğ-+ğ. �ğ/+ğ0 ˆğ1+ğ2 ˆ%ğ3+ğ4 ˆSğ5+ğ6 �dğ7+ğ8 ˜ğ9+ğ: �Tğ;+ğ< �ğ=+ğ> �1ğ?+ğ@ �!ğA+ğB �4ğC+ğD �ağE+ğ +ğF
Ø Ø ØØ!Ø $ØØ ğU+ğ +ğ +€€€rC | 2,697 | Python | .py | 19 | 140.789474 | 589 | 0.454274 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,510 | logger.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/logger.cpython-311.pyc | §
føfî ã ór — d dl Z d dlZd dlZd dlZd dlmZ d dlmZ G d„ d¦ « Z G d„ d¦ « Z dS )é N)ÚAPI©Úfind_project_rootc óŠ — e Zd Zddepddepdddfd„Zddededdfd „Zdepdfd
„Zdepdddfd „Zdded
eddfd„Zd„ Z dd„Z
dS )ÚLoggerNÚlogdirÚtokenÚreturnc ó — | |¦ « | _ t j | j d¬¦ « t j d¦ « | _ | j t j ¦ « t j | j › dt j
¦ « d¦ « › d�¦ « | _
| j
t j ¦ « | j | j
¦ « d| _ t! ¦ « | _ | |¦ « t j d¦ « }| j
|¦ « t+ j | j ¦ « d S ) NT)Úexist_okÚAlphaZeroLoggerú/z%Y-%m-%d_%H-%M-%Sz.logFz)[%(asctime)s - %(levelname)s] %(message)s)Ú init_logdirr ÚosÚmakedirsÚloggingÚ getLoggerÚloggerÚsetLevelÚDEBUGÚ FileHandlerÚdatetimeÚnowÚstrftimeÚ file_handlerÚ
addHandlerÚ is_token_setr ÚapiÚinit_api_tokenÚ FormatterÚ setFormatterÚatexitÚregisterÚcleanup)Úselfr r Ú formatters úM/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/logger.pyÚ__init__zLogger.__init__ s= € Ø×&Ò& vÑ.Ô.ˆŒ İ
Œ �D”K¨$Ğ/Ñ/Ô/Ğ/İ Ô'Ğ(9Ñ:Ô:ˆŒ Ø Œ × Ò �Wœ]Ñ+Ô+Ğ+İ #Ô /ØŒ{Ğ XĞ X�XÔ .× 2Ò 2Ñ 4Ô 4× =Ò =Ğ>QÑ RÔ RĞ XĞ XĞ Xñ Zô ZˆÔà Ô×"Ò"¥7¤=Ñ1Ô1Ğ1Ø Œ × Ò ˜tÔ 0Ñ1Ô1Ğ1Ø !ˆÔİ‘5”5ˆŒØ ×Ò˜EÑ"Ô"Ğ"İÔ%Ğ&QÑRÔRˆ Ø Ô×&Ò& yÑ1Ô1Ğ1İŒ˜œ Ñ%Ô%Ğ%Ğ%Ğ%ó ÚdebugÚmsgÚlevelc óB — t | j |¦ « |¦ « d S ©N)Úgetattrr )r% r+ r, s r' Úlogz
Logger.log s$ € Ø#��” ˜UÑ#Ô# CÑ(Ô(Ğ(Ğ(Ğ(r) c ó, — |€t ¦ « › d�S |S )Nz/Logs/ProgramLogsr )r% r s r' r zLogger.init_logdir s! € Ø ˆ>İ'Ñ)Ô)Ğ<Ğ<Ğ<Ğ <àˆMr) c óP — |€d S | j |¦ « d| _ d S )NT)r Ú set_tokenr )r% r s r' r zLogger.init_api_token% s0 € Ø ˆ=Ø ˆFØ Œ×Ò˜5Ñ!Ô!Ğ!Ø ˆÔĞĞr) ÚMuZeroÚ algorithmc ó¤ — | j sd S | j |› d�|¦ « d S # t $ r}t |¦ « Y d }~d S d }~ww xY w)Nz training notification.)r r Ú send_noteÚ ExceptionÚprint)r% r+ r5 Úes r' Úpushbullet_logzLogger.pushbullet_log+ st € ØÔ ğ Ø ˆFğ Ø ŒH× Ò )Ğ DĞ DĞ DÀcÑ JÔ JĞ JĞ JĞ Jøİğ ğ ğ İ �!‰HŒHˆHˆHˆHˆHˆHˆHˆHøøøøğ øøøs ‹ + «
AµA
Á
Ac óz — t j | j ¦ « D ] }t j | j › d|› �¦ « Œ!d S )Nr )r Úlistdirr Úremove)r% Ú file_names r' Ú clear_logdirzLogger.clear_logdir3 sJ € İœ D¤KÑ0Ô0ğ 4ğ 4ˆIİ ŒI˜œĞ2Ğ2 yĞ2Ğ2Ñ 3Ô 3Ğ 3Ğ 3ğ 4ğ 4r) c óv — | j ¦ « | j | j ¦ « d S r. )r Úcloser Ú
removeHandler)r% s r' r$ zLogger.cleanup7 s6 € Ø Ô× Ò Ñ!Ô!Ğ!Ø Œ ×!Ò! $Ô"3Ñ4Ô4Ğ4Ğ4Ğ4r) r. )r* )r4 )r
N) Ú__name__Ú
__module__Ú __qualname__Ústrr( r0 r r r; r@ r$ © r) r' r r s € € € € € ğ&ğ &˜s˜{ dğ &°3°;¸$ğ &È$ğ &ğ &ğ &ğ &ğ )ğ )�sğ ) 3ğ )°Tğ )ğ )ğ )ğ )ğ ¨ğ ğ ğ ğ ğ ! C K¨4ğ !°Dğ !ğ !ğ !ğ !ğ ğ #ğ °#ğ ÀTğ ğ ğ ğ ğ4ğ 4ğ 4ğ5ğ 5ğ 5ğ 5ğ 5ğ 5r) r c
óÈ — e Zd Zedededefd„¦ « Zededededededef d „¦ « Zedefd
„¦ « Zedededed e fd „¦ « Z
ed
efd„¦ « Z ede fd„¦ « Z
ede de fd„¦ « Zede de fd„¦ « Zedefd„¦ « Zedefd„¦ « Zededefd„¦ « Zededefd „¦ « Zed efd „¦ « Zed „ ¦ « Zd S )!ÚLoggingMessageTemplatesÚname1Úname2Ú num_gamesc ó — d| › d|› d|› d�S )NzStarting pitting between ú and z for ú games.rH )rK rL rM s r' Ú
PITTING_STARTz%LoggingMessageTemplates.PITTING_START> s# € àU¨5ĞUĞU°uĞUĞUÀ9ĞUĞUĞUĞUr) Úwins1Úwins2ÚtotalÚdrawsc ó2 — d| › d|› d||z › d||z › d|› d� S )NzPitting ended between rO z. Player 1 win frequency: ú. Player 2 win frequency: ú . Draws: ú.rH )rK rL rR rS rT rU s r' Ú PITTING_ENDz#LoggingMessageTemplates.PITTING_ENDB se € ğL¨ğ Lğ L°Uğ Lğ LØ+0°5©=ğLğ Là+0°5©=ğLğ LàCHğLğ Lğ Lğ Mr) c ó — d| › d�S )NzStarting self play for rP rH )rM s r' ÚSELF_PLAY_STARTz'LoggingMessageTemplates.SELF_PLAY_STARTH s € à;¨Ğ;Ğ;Ğ;Ğ;r) Ú not_zero_fnc ór — | �|�|€dS d| || |z |z ¦ « z › d| || |z |z ¦ « z › d|› d�S )NzTSelf play ended. Results not available (This is expected if you are running MuZero).z)Self play ended. Player 1 win frequency: rW rX rY rH )rR rS rU r] s r' Ú
SELF_PLAY_ENDz%LoggingMessageTemplates.SELF_PLAY_ENDL sˆ € à ˆ=˜E˜M¨U¨]ØiĞiğk¸EÀ[À[ĞQVĞY^ÑQ^ĞafÑQfÑEgÔEgÑ<hğ kğ kØ+0°K°KÀÈÁ
ĞPUÑ@UÑ4VÔ4VÑ+Wğkğ kØbgğkğ kğ kğ lr) Ú num_epchsc ó — d| › d�S )Nz Starting network training for z epochs.rH )r` s r' ÚNETWORK_TRAINING_STARTz.LoggingMessageTemplates.NETWORK_TRAINING_STARTS s € àC° ĞCĞCĞCĞCr) Ú mean_lossc ó — d| › �S )Nz#Network training ended. Mean loss: rH )rc s r' ÚNETWORK_TRAINING_ENDz,LoggingMessageTemplates.NETWORK_TRAINING_ENDW s € à@°YĞ@Ğ@Ğ@r) Únum_winsÚupdate_thresholdc ó — d| › d|› d�S )Nz:!!! Model rejected, restoring previous version. Win rate: ú. Update threshold: ú !!!rH ©rf rg s r' Ú MODEL_REJECTz$LoggingMessageTemplates.MODEL_REJECT[ s* € ğ<ÈXğ <ğ <Ø%5ğ<ğ <ğ <ğ =r) c ó — d| › d|› d�S )Nz7!!! Model accepted, keeping current version. Win rate: ri rj rH rk s r' Ú MODEL_ACCEPTz$LoggingMessageTemplates.MODEL_ACCEPT` s* € ğ
Àhğ
ğ
Ğdtğ
ğ
ğ
ğ r) Ú num_itersc ó — d| › d�S )NzStarting training for z iterations.rH )ro s r' ÚTRAINING_STARTz&LoggingMessageTemplates.TRAINING_STARTf s € à?¨ Ğ?Ğ?Ğ?Ğ?r) Ú args_usedc ój — d}| ¦ « D ]\ }}||› d|› d�z
}Œd|d d… › �S )NÚ z: z, zTraining ended. Args used: éşÿÿÿ)Úitems)rr Ú
args_used_strÚkeyÚvalues r' Ú TRAINING_ENDz$LoggingMessageTemplates.TRAINING_ENDj sX € àˆ
Ø#Ÿ/š/Ñ+Ô+ğ 1ğ 1‰JˆC�Ø Ğ 0Ğ 0 uĞ 0Ğ 0Ğ 0Ñ 0ˆMˆMØA¨]¸3¸B¸3Ô-?ĞAĞAĞAr) Útype_Úpathc ó — d| › d|› �S )NzSaved z to rH ©r{ r| s r' ÚSAVEDz LoggingMessageTemplates.SAVEDq s € à)˜Ğ)Ğ) 4Ğ)Ğ)Ğ)r) c ó — d| › d|› �S )Nz Restored z from rH r~ s r' ÚLOADEDz LoggingMessageTemplates.LOADEDu s € à.˜5Ğ.Ğ.¨Ğ.Ğ.Ğ.r) Úiterc ó — d| › d�S )Nz
Iteration z$ of the algorithm training finished!rH )r‚ s r' ÚITER_FINISHED_PSBz)LoggingMessageTemplates.ITER_FINISHED_PSBy s € àF˜DĞFĞFĞFĞFr) c ó — dS )Nz;Algorithm Training finished, you can collect the results :)rH rH r) r' ÚTRAINING_END_PSBz(LoggingMessageTemplates.TRAINING_END_PSB} s € àLĞLr) N)rD rE rF Ú staticmethodrG ÚintrQ rZ r\ Úcallabler_ rb Úfloatre rl rn rq Údictrz r r� r„ r† rH r) r' rJ rJ < s € € € € € àğV˜Sğ V¨ğ V¸ğ Vğ Vğ Vñ „\ğVğ ğM˜3ğ M sğ M°3ğ M¸sğ MÈ3ğ MĞWZğ Mğ Mğ Mñ „\ğMğ
ğ< 3ğ <ğ <ğ <ñ „\ğ<ğ ğl˜Sğ l¨ğ l°Sğ lÀxğ lğ lğ lñ „\ğlğ ğD¨#ğ Dğ Dğ Dñ „\ğDğ ğA¨ğ Ağ Ağ Añ „\ğAğ ğ=˜uğ =¸ğ =ğ =ğ =ñ „\ğ=ğ ğ˜uğ ¸ğ ğ ğ ñ „\ğğ
ğ@ #ğ @ğ @ğ @ñ „\ğ@ğ ğB ğ Bğ Bğ Bñ „\ğBğ ğ*�Sğ * ğ *ğ *ğ *ñ „\ğ*ğ ğ/�cğ / ğ /ğ /ğ /ñ „\ğ/ğ ğG ğ Gğ Gğ Gñ „\ğGğ ğMğ Mñ „\ğMğ Mğ Mr) rJ )
r" r r r Ú
pushbulletr Úmu_alpha_zero.General.utilsr r rJ rH r) r' ú<module>r� s¼ ğØ
€
€
€
Ø €€€Ø €€€Ø € € € à Ğ Ğ Ğ Ğ Ğ à 9Ğ 9Ğ 9Ğ 9Ğ 9Ğ 9ğ.5ğ .5ğ .5ğ .5ğ .5ñ .5ô .5ğ .5ğbCMğ CMğ CMğ CMğ CMñ CMô CMğ CMğ CMğ CMr) | 9,693 | Python | .py | 20 | 483.6 | 3,727 | 0.398284 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,511 | arena.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Arena/arena.py | import time
import wandb
from mu_alpha_zero.AlphaZero.Arena.players import Player
# from mu_alpha_zero.General.az_game import AlphaZeroGame
from mu_alpha_zero.General.arena import GeneralArena
from mu_alpha_zero.Hooks.hook_manager import HookManager
from mu_alpha_zero.Hooks.hook_point import HookAt
from mu_alpha_zero.MuZero.utils import scale_action, resize_obs
from mu_alpha_zero.config import AlphaZeroConfig
class Arena(GeneralArena):
def __init__(self, game_manager, alpha_zero_config: AlphaZeroConfig, device,
hook_manager: HookManager or None = None, state_managed: bool = False):
self.game_manager = game_manager
self.state_managed = state_managed
self.device = device
self.hook_manager = hook_manager if hook_manager is not None else HookManager()
self.alpha_zero_config = alpha_zero_config
def pit(self, player1: Player, player2: Player, num_games_to_play: int, num_mc_simulations: int,
one_player: bool = False, start_player: int = 1, add_to_kwargs: dict or None = None, debug: bool = False,
override_player_with_returned_state: bool = False) -> \
tuple[int, int, int]:
"""
Pit two players against each other for a given number of games and gather the results.
:param start_player: Which player should start the game.
:param one_player: If True always only the first player will start the game.
:param player1:
:param player2:
:param num_games_to_play:
:param num_mc_simulations:
:return: number of wins for player1, number of wins for player2, number of draws
"""
results = {"wins_p1": 0, "wins_p2": 0, "draws": 0}
tau = self.alpha_zero_config.arena_tau
if one_player:
num_games_per_player = num_games_to_play
else:
num_games_per_player = num_games_to_play // 2
if self.state_managed:
if player1.name == "NetPlayer":
player1.monte_carlo_tree_search.game_manager = self.game_manager
if player2.name == "NetPlayer":
player2.monte_carlo_tree_search.game_manager = self.game_manager
player1.set_game_manager(self.game_manager)
player2.set_game_manager(self.game_manager)
move = None
for game in range(num_games_to_play):
if game < num_games_per_player:
current_player = start_player
else:
current_player = -start_player
kwargs = {"num_simulations": num_mc_simulations, "current_player": current_player, "device": self.device,
"tau": tau, "unravel": self.alpha_zero_config.unravel, "move": move}
if add_to_kwargs is not None:
kwargs.update(add_to_kwargs)
if not self.alpha_zero_config.requires_player_to_reset:
state = self.game_manager.reset()
else:
state = self.game_manager.reset(player=current_player)
if override_player_with_returned_state:
if self.alpha_zero_config.arena_running_muzero:
current_player = int(state[:, :, -1][0][0])
kwargs["current_player"] = current_player
else:
current_player = self.game_manager.current_player
kwargs["current_player"] = current_player
if self.alpha_zero_config.arena_running_muzero and self.alpha_zero_config.enable_frame_buffer:
try:
player1.monte_carlo_tree_search.buffer.init_buffer(
self.game_manager.get_state_for_passive_player(state, 1),
1)
player2.monte_carlo_tree_search.buffer.init_buffer(
self.game_manager.get_state_for_passive_player(state, -1), -1)
except AttributeError:
pass
if player1.name == "NetworkPlayer":
player1.monte_carlo_tree_search.step_root(None)
if player2.name == "NetworkPlayer":
player2.monte_carlo_tree_search.step_root(None)
# time.sleep(0.01)
while True:
self.game_manager.render()
if current_player == 1:
move = player1.choose_move(state, **kwargs)
else:
move = player2.choose_move(state, **kwargs)
self.hook_manager.process_hook_executes(self, self.pit.__name__, __file__, HookAt.MIDDLE,
args=(move, kwargs, current_player))
if not self.state_managed:
state = self.game_manager.get_next_state(state, move, current_player)
status = self.game_manager.game_result(current_player, state)
else:
state = self.game_manager.get_next_state(move, current_player)[0]
status = self.game_manager.game_result(current_player)
self.game_manager.render()
if self.alpha_zero_config.arena_running_muzero:
if move is not None:
move = scale_action(move, self.game_manager.get_num_actions()) if isinstance(move, int) else \
scale_action(move[0] * state.shape[0] + move[1], self.game_manager.get_num_actions())
try:
if current_player == 1:
if player2.name == "NetPlayer":
player2.monte_carlo_tree_search.buffer.add_frame(state, move, -1)
player1.monte_carlo_tree_search.buffer.add_frame(
self.game_manager.get_state_for_passive_player(state, 1), move, 1)
else:
if player1.name == "NetPlayer":
player1.monte_carlo_tree_search.buffer.add_frame(state, move, 1)
player2.monte_carlo_tree_search.buffer.add_frame(
self.game_manager.get_state_for_passive_player(state, -1), move, -1)
except AttributeError:
# Player is probably not a net player and doesn't have monte_carlo_tree_search.
pass
if status is not None:
if status == 1:
if current_player == 1:
results["wins_p1"] += 1
else:
results["wins_p2"] += 1
elif status == -1:
if current_player == 1:
results["wins_p2"] += 1
else:
results["wins_p1"] += 1
else:
results["draws"] += 1
# if debug:
# self.wait_keypress()
print(f"player {current_player} wins")
if (player1.name == "HumanPlayer" or player2.name == "HumanPlayer") and debug:
time.sleep(1)
try:
wandb.log(
{"wins_p1": results["wins_p1"], "wins_p2": results["wins_p2"], "draws": results["draws"]})
except wandb.errors.Error:
pass
break
current_player *= -1
kwargs["current_player"] = current_player
return results["wins_p1"], results["wins_p2"], results["draws"]
def wait_keypress(self):
inpt = input("Press any key to continue...")
return inpt
| 7,791 | Python | .py | 142 | 37.852113 | 118 | 0.541547 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,512 | players.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Arena/players.py | import copy
import time
from abc import ABC, abstractmethod
import numpy as np
from mu_alpha_zero.Game.tictactoe_game import TicTacToeGameManager
class Player(ABC):
"""
To create a custom player, extend this class and implement the choose_move method.
You can see different implementations below.
"""
@abstractmethod
def __init__(self, game_manager, **kwargs):
pass
@abstractmethod
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
pass
@abstractmethod
def make_fresh_instance(self):
pass
def init_kwargs(self, kwargs: dict):
for key in kwargs.keys():
setattr(self, key, kwargs[key])
@abstractmethod
def set_game_manager(self, game_manager):
pass
class RandomPlayer(Player):
def __init__(self, game_manager: TicTacToeGameManager, **kwargs):
self.game_manager = game_manager
self.name = self.__class__.__name__
self.kwargs = kwargs
self.init_kwargs(kwargs)
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
move = self.game_manager.get_random_valid_action(board, **kwargs)
if "unravel" in kwargs.keys():
unravel = kwargs["unravel"]
else:
unravel = True
return tuple(move) if unravel else int(move)
def make_fresh_instance(self):
return RandomPlayer(self.game_manager.make_fresh_instance(), **self.kwargs)
def set_game_manager(self, game_manager):
self.game_manager = game_manager
class PerfectConnect4Player(Player):
def __init__(self, game_manager, **kwargs):
self.game_manager = game_manager
self.name = self.__class__.__name__
self.kwargs = kwargs
self.init_kwargs(kwargs)
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
assert kwargs.get("move") is not None, "Please provide the last move made."
from mu_alpha_zero.PerfectConnect4Negamax.board import Board
from mu_alpha_zero.PerfectConnect4Negamax.solver import solve
bd = Board()
bd.play(kwargs["move"])
move = solve(bd)
return move
def make_fresh_instance(self):
pass
def set_game_manager(self, game_manager):
pass
class NetPlayer(Player):
def __init__(self, game_manager: TicTacToeGameManager, **kwargs):
self.game_manager = game_manager
self.name = self.__class__.__name__
self.kwargs = kwargs
self.init_kwargs(kwargs)
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
try:
current_player = kwargs["current_player"]
device = kwargs["device"]
tau = kwargs["tau"]
except KeyError:
raise KeyError("Missing keyword argument. Please supply kwargs: current_player, device, "
"tau")
pi, _ = self.monte_carlo_tree_search.search(self.network, board, current_player, device, tau=tau)
move = self.game_manager.select_move(pi, tau=tau)
self.monte_carlo_tree_search.step_root(None)
if "unravel" in kwargs.keys():
unravel = kwargs["unravel"]
else:
unravel = True
if unravel:
return self.game_manager.network_to_board(move)
return int(move)
def make_fresh_instance(self):
return NetPlayer(self.game_manager.make_fresh_instance(), **{"network": copy.deepcopy(self.network),
"monte_carlo_tree_search": self.monte_carlo_tree_search.make_fresh_instance()})
def set_network(self, network):
self.network = network
def set_game_manager(self, game_manager):
self.game_manager = game_manager
class HumanPlayer(Player):
def __init__(self, game_manager: TicTacToeGameManager, **kwargs):
self.name = self.__class__.__name__
self.game_manager = game_manager
self.kwargs = kwargs
self.init_kwargs(kwargs)
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
time.sleep(0.4)
if self.game_manager.headless:
raise RuntimeError("Cannot play with a human player in headless mode.")
move = self.game_manager.get_human_input(board)
return move
def make_fresh_instance(self):
return HumanPlayer(self.game_manager.make_fresh_instance(), **self.kwargs)
def set_game_manager(self, game_manager):
self.game_manager = game_manager
class Connect4MinimaxPlayer(Player):
def __init__(self, game_manager, **kwargs):
self.game_manager = game_manager
self.name = self.__class__.__name__
self.kwargs = kwargs
self.init_kwargs(kwargs)
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
time.sleep(0.4)
from mu_alpha_zero.MuZero.MinimaxOpponent.player import PlayerMM
from mu_alpha_zero.MuZero.MinimaxOpponent.board import Board
bd = Board.from_game_state(board[:, :, 0], (1, kwargs["move"]))
player = PlayerMM(6, False)
move = player.findMove(bd)
return move
def make_fresh_instance(self):
return Connect4MinimaxPlayer(self.game_manager.make_fresh_instance(), **self.kwargs)
def set_game_manager(self, game_manager):
self.game_manager = game_manager
class NoopPlayer(Player):
def __init__(self, game_manager, **kwargs):
self.game_manager = game_manager
self.name = self.__class__.__name__
self.kwargs = kwargs
self.init_kwargs(kwargs)
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
return None
def make_fresh_instance(self):
pass
def set_game_manager(self, game_manager):
pass
class MinimaxPlayer(Player):
def __init__(self, game_manager: TicTacToeGameManager, **kwargs):
# self.evaluate_fn = evaluate_fn
self.game_manager = game_manager
self.name = self.__class__.__name__
self.kwargs = kwargs
self.init_kwargs(kwargs)
def choose_move(self, board: np.ndarray, **kwargs) -> tuple[int, int]:
try:
depth = kwargs["depth"]
player = kwargs["player"]
except KeyError:
raise KeyError("Missing keyword argument. Please supply kwargs: depth, player")
move = self.minimax(board.copy(), depth, True, player)[1]
return tuple(move)
def minimax(self, board: np.ndarray, depth: int, is_max: bool, player: int, alpha=-float("inf"),
beta=float("inf")) -> tuple:
# self.game_manager.check_pg_events()
eval_ = self.evaluate_fn(board)
if eval_ is not None:
return eval_, None
if depth == 0:
return self.evaluate_fn(board, check_end=False), None
if is_max:
best_score = -float("inf")
best_move = None
for move in self.game_manager.get_valid_moves(board):
board[move[0]][move[1]] = player
score = self.minimax(board.copy(), depth - 1, False, -player, alpha, beta)[0]
# print(score)
board[move[0]][move[1]] = 0
if score > best_score:
best_move = move
best_score = max(score, best_score)
alpha = max(alpha, best_score)
if alpha >= beta:
break
# if best_move is None:
# return self.evaluate_fn(board, player), None
return best_score, best_move
else:
best_score = float("inf")
best_move = None
for move in self.game_manager.get_valid_moves(board):
board[move[0]][move[1]] = player
score = self.minimax(board.copy(), depth - 1, True, -player, alpha, beta)[0]
if score is None:
continue
# print(score)
board[move[0]][move[1]] = 0
best_score = min(score, best_score)
beta = min(beta, best_score)
if beta <= alpha:
break
# if best_move is None:
# return self.evaluate_fn(board, player), None
return best_score, best_move
def make_fresh_instance(self):
return MinimaxPlayer(self.game_manager.make_fresh_instance(), **self.kwargs)
def set_game_manager(self, game_manager):
self.game_manager = game_manager
| 8,576 | Python | .py | 198 | 33.540404 | 148 | 0.606607 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,513 | players.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Arena/__pycache__/players.cpython-311.pyc | §
�Œf$ ã ó — d dl Z d dlZd dlmZmZ d dlZd dlZd dlm Z d dl
m Z d dl m
Z
G d„ de¦ « Z G d„ d e¦ « Z G d
„ d e¦ « Z G d „ d
e¦ « Z G d„ de¦ « Z G d„ de¦ « Z G d„ de¦ « ZdS )é N)ÚABCÚabstractmethod)Ú AlphaZeroNet)ÚTicTacToeGameManager)Ú
AlphaZeroGamec óš — e Zd ZdZed„ ¦ « Zedej dee e f fd„¦ « Z
ed„ ¦ « Z de fd„Z
ed „ ¦ « Zd
S ) ÚPlayerz�
To create a custom player, extend this class and implement the choose_move method.
You can see different implementations below.
c ó — d S ©N© ©ÚselfÚ game_managerÚkwargss úT/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Arena/players.pyÚ__init__zPlayer.__init__ ó € à ˆó ÚboardÚreturnc ó — d S r r )r r r s r Ú choose_movezPlayer.choose_move r r c ó — d S r r ©r s r Úmake_fresh_instancezPlayer.make_fresh_instance r r r c ób — | ¦ « D ]}t | ||| ¦ « Œd S r )ÚkeysÚsetattr)r r Úkeys r Ú init_kwargszPlayer.init_kwargs s< € Ø—;’;‘=”=ğ ,ğ ,ˆCİ �D˜#˜v cœ{Ñ +Ô +Ğ +Ğ +ğ ,ğ ,r c ó — d S r r ©r r s r Úset_game_managerzPlayer.set_game_manager# r r N)Ú__name__Ú
__module__Ú __qualname__Ú__doc__r r ÚnpÚndarrayÚtupleÚintr r Údictr r# r r r r r
s € € € € € ğğ ğ
ğ
ğ
ñ „^ğ
ğ ğ
¤ğ
¸%ÀÀSÀ¼/ğ
ğ
ğ
ñ „^ğ
ğ ğ
ğ
ñ „^ğ
ğ, $ğ ,ğ ,ğ ,ğ ,ğ ğ
ğ
ñ „^ğ
ğ
ğ
r r c óP — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z dS ) Ú RandomPlayerr c ón — || _ | j j | _ || _ | |¦ « d S r ©r Ú __class__r$ Únamer r r
s r r zRandomPlayer.__init__) ó7 € Ø (ˆÔØ”NÔ+ˆŒ Ø ˆŒ Ø ×Ò˜Ñ Ô Ğ Ğ Ğ r r r c óª — | j j |fi |¤�}d| ¦ « v r |d }nd}|rt |¦ « nt |¦ « S )NÚunravelT)r Úget_random_valid_actionr r* r+ )r r r Úmover5 s r r zRandomPlayer.choose_move/ s^ € Ø8ˆtÔ Ô8¸ĞIĞIÀ&ĞIĞIˆØ ˜Ÿ š ™
œ
Ğ %Ğ %Ø ˜YÔ'ˆGˆGàˆGØ %Ğ4�u�T‰{Œ{ˆ{3¨t©9¬9Ğ4r c óT — t | j ¦ « fi | j ¤�S r )r. r r r r s r r z RandomPlayer.make_fresh_instance7 s) € ݘDÔ -× AÒ AÑ CÔ CĞSĞSÀtÄ{ĞSĞSĞSr c ó — || _ d S r ©r r" s r r# z RandomPlayer.set_game_manager: ó € Ø (ˆÔĞĞr N© r$ r% r& r r r( r) r* r+ r r r# r r r r. r. ( s} € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ 5 ¤ğ 5¸%ÀÀSÀ¼/ğ 5ğ 5ğ 5ğ 5ğTğ Tğ Tğ)ğ )ğ )ğ )ğ )r r. c óV — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z d„ Z d S )
Ú NetPlayerr c ón — || _ | j j | _ || _ | |¦ « d S r r0 r
s r r zNetPlayer.__init__? r3 r r r c ó — |d }|d }|d }n # t $ r t d¦ « ‚w xY w| j | j ||||¬¦ « \ }}| j ||¬¦ « }| j d ¦ « d| ¦ « v r |d } nd} | r| j |¦ « S t |¦ « S )NÚcurrent_playerÚdeviceÚtauúKMissing keyword argument. Please supply kwargs: current_player, device, tau©rC r5 T)
ÚKeyErrorÚmonte_carlo_tree_searchÚsearchÚnetworkr Ú select_moveÚ step_rootr Únetwork_to_boardr+ )
r r r rA rB rC ÚpiÚ_r7 r5 s
r r zNetPlayer.choose_moveE s € ğ "Ø #Ğ$4Ô 5ˆNؘHÔ%ˆFؘ”-ˆCˆCøİğ "ğ "ğ "İğ !ñ "ô "ğ
"ğ "øøøğ Ô,×3Ò3°D´LÀ%È ĞY_ĞehĞ3ÑiÔi‰ˆˆAØÔ ×,Ò,¨R°CĞ,Ñ8Ô8ˆØ Ô$×.Ò.¨tÑ4Ô4Ğ4Ø ˜Ÿ š ™
œ
Ğ %Ğ %Ø ˜YÔ'ˆGˆGàˆGØ ğ <ØÔ$×5Ò5°dÑ;Ô;Ğ ;İ�4‰yŒyĞó ‚ ›5c ó¬ — t | j ¦ « fi t j | j ¦ « | j ¦ « dœ¤�S )N)rI rG )r> r r ÚcopyÚdeepcopyrI rG r s r r z NetPlayer.make_fresh_instanceY sv € ݘÔ*×>Ò>Ñ@Ô@ğ Uğ UÕPTÔP]Ğ^bÔ^jÑPkÔPkØ`dÔ`|÷ aQò aQñ aSô aSğETğ ETğ Uğ Uğ Ur c ó — || _ d S r )rI )r rI s r Ú set_networkzNetPlayer.set_network] s
€ Ø ˆŒ ˆ ˆ r c ó — || _ d S r r: r" s r r# zNetPlayer.set_game_manager` r; r N)
r$ r% r& r r r( r) r* r+ r r rT r# r r r r> r> > sŒ € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ(Uğ Uğ Uğ ğ ğ ğ)ğ )ğ )ğ )ğ )r r> c ód — e Zd Zdededefd„Zdefd„Zdej de
e e f fd„Z d „ Z
d
„ Zd S ) ÚTrainingNetPlayerrI r Úargsc ó — t d¦ « ‚)Nz8Don't use this class yet, it produces incorrect results.)
ÚNotImplementedErrorr1 r$ r2 Ú _TrainingNetPlayer__init_argsrX rI r ÚtraceÚ traced_path)r rI r rX s r r zTrainingNetPlayer.__init__e s € İ!Ğ"\Ñ]Ô]Ğ]r r c ó‚ — dD ];} | |¦ « Œ# t $ r t d|› d�¦ « Y Œ8w xY w|S )N)Úcheckpoint_dirÚ max_depthzKey z
not present.)ÚpoprF Úprint)r rX r s r Ú __init_argsz TrainingNetPlayer.__init_argsm sg € Ø2ğ 1ğ 1ˆCğ
1Ø—’˜‘
”
�
�
øİğ
1ğ
1ğ
1İĞ/˜SĞ/Ğ/Ğ/Ñ0Ô0Ğ0Ğ0Ğ0ğ
1øøøàˆ s † œ <»<r c ó, — |d }|d }|d }n # t $ r t d¦ « ‚w xY wt |||| j | j ¦ « }| j ||¬¦ « }| j |¦ « S )NrA rB rC rD rE )rF Ú
CpSelfPlayÚ CmctsSearchrX r] r rJ rL )r r r rA rB rC rM r7 s r r z TrainingNetPlayer.choose_moveu s € ğ "Ø #Ğ$4Ô 5ˆNؘHÔ%ˆFؘ”-ˆCˆCøİğ "ğ "ğ "İğ !ñ "ô "ğ
"ğ "øøøõ ×
#Ò
# E¨>¸3ÀÄ È4ÔK[Ñ
\Ô
\ˆØÔ ×,Ò,¨R°CĞ,Ñ8Ô8ˆØÔ ×1Ò1°$Ñ7Ô7Ğ7rO c ó — t ‚r )rZ r s r r z%TrainingNetPlayer.make_fresh_instance� s € İ!Ğ!r c ó — || _ d S r r: r" s r r# z"TrainingNetPlayer.set_game_manager„ r; r N)r$ r% r& r r r, r r[ r( r) r* r+ r r r# r r r rW rW d s« € € € € € ğG ğ GĞ<Pğ GĞX\ğ Gğ Gğ Gğ Gğ 4ğ ğ ğ ğ ğ
8 ¤ğ
8¸%ÀÀSÀ¼/ğ
8ğ
8ğ
8ğ
8ğ"ğ "ğ "ğ)ğ )ğ )ğ )ğ )r rW c óP — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z dS ) Ú HumanPlayerr c ón — | j j | _ || _ || _ | |¦ « d S r )r1 r$ r2 r r r r
s r r zHumanPlayer.__init__‰ s7 € Ø”NÔ+ˆŒ Ø (ˆÔØ ˆŒ Ø ×Ò˜Ñ Ô Ğ Ğ Ğ r r r c óp — | j j rt d¦ « ‚| j |¦ « }|S )Nz1Cannot play with a human player in headless mode.)r ÚheadlessÚ RuntimeErrorÚget_human_input)r r r r7 s r r zHumanPlayer.choose_move� s; € Ø Ô Ô %ğ Tİ Ğ RÑSÔSĞ SØÔ ×0Ò0°Ñ7Ô7ˆØˆ r c óT — t | j ¦ « fi | j ¤�S r )rj r r r r s r r z HumanPlayer.make_fresh_instance• s) € ݘ4Ô,×@Ò@ÑBÔBĞRĞRÀdÄkĞRĞRĞRr c ó — || _ d S r r: r" s r r# z HumanPlayer.set_game_manager˜ r; r Nr< r r r rj rj ˆ s} € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ Sğ Sğ Sğ)ğ )ğ )ğ )ğ )r rj c óV — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z d„ Z d S )
ÚJavaMinimaxPlayerr c óÄ — t j d¦ « t j ¦ « || _ | j j | _ || _ t j | j
¦ « d S )NzFC:\Users\Skyr\IdeaProjects\Minimax\build\libs\Minimax-1.0-SNAPSHOT.jar) ÚjpypeÚ addClassPathÚstartJVMr r1 r$ r2 r ÚatexitÚregisterÚ on_shutdownr
s r r zJavaMinimaxPlayer.__init__� sV € İ
ÔĞdÑeÔeĞeİ
ŒÑÔĞØ (ˆÔØ”NÔ+ˆŒ Ø ˆŒ İŒ˜Ô(Ñ)Ô)Ğ)Ğ)Ğ)r r r c ór — |d }|d }n # t $ r t d¦ « ‚w xY wt j d¦ « } |¦ « }t | j j ¦ « t j |¦ « }| ||||d| j j ¦ « }t |¦ « S )NÚdepthÚplayerú=Missing keyword argument. Please supply kwargs: depth, playerzdev.skyr.MinimaxT)
rF ru ÚJClassrb r Ú
num_to_winÚJArrayÚofÚrunr* ) r r r r| r} ÚMinimaxÚminimaxÚjar7 s r r z JavaMinimaxPlayer.choose_move¦ sº € ğ \ؘ7”OˆEؘHÔ%ˆFˆFøİğ \ğ \ğ \İĞZÑ[Ô[Ğ [ğ \øøøå”,Ğ 1Ñ2Ô2ˆØ�'‘)”)ˆİ
ˆdÔ Ô*Ñ+Ô+Ğ+İ
Œ\�_Š_˜UÑ
#Ô
#ˆØ�{Š{˜5 &¨&°%¸¸tÔ?PÔ?[Ñ\Ô\ˆİ�T‰{Œ{Ğó ‚ “-c ó — d S r r r s r r z%JavaMinimaxPlayer.make_fresh_instance³ s € Ø ˆr c ó, — t j ¦ « d S r )ru Ú shutdownJVMr s r rz z JavaMinimaxPlayer.on_shutdown¶ s € İ
ÔÑÔĞĞĞr c ó — || _ d S r r: r" s r r# z"JavaMinimaxPlayer.set_game_manager¹ r; r N)
r$ r% r& r r r( r) r* r+ r r rz r# r r r rs rs œ sˆ € € € € € ğ* ]ğ *ğ *ğ *ğ *ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ
ğ
ğ
ğ ğ ğ ğ)ğ )ğ )ğ )ğ )r rs c ó¢ — e Zd Zdefd„Zdej deeef fd„Z e
d¦ « e
d¦ « fdej dede d edef
d
„Z d „ Z
d „ Zd
S )Ú
MinimaxPlayerr c ón — || _ | j j | _ || _ | |¦ « d S r r0 r
s r r zMinimaxPlayer.__init__¾ s7 € à (ˆÔØ”NÔ+ˆŒ Ø ˆŒ Ø ×Ò˜Ñ Ô Ğ Ğ Ğ r r r c óŞ — |d }|d }n # t $ r t d¦ « ‚w xY w| | ¦ « |d|¦ « d }t |¦ « S )Nr| r} r~ Té )rF r… rQ r* )r r r r| r} r7 s r r zMinimaxPlayer.choose_moveÅ s{ € ğ \ؘ7”OˆEؘHÔ%ˆFˆFøİğ \ğ \ğ \İĞZÑ[Ô[Ğ [ğ \øøøà�|Š|˜EŸJšJ™LœL¨%°°vÑ>Ô>¸qÔAˆİ�T‰{Œ{Ğr‡ Úinfr| Úis_maxr} c óŠ — | |¦ « }|�|d fS |dk r| |d¬¦ « d fS |rÇt d¦ « }d } | j |¦ « D ]–}
|||
d |
d < | | ¦ « |dz
d| ||¦ « d } d||
d |
d < | |k r|
} t
| |¦ « }t
||¦ « }||k r nŒ—|| fS t d¦ « }d } | j |¦ « D ]‘}
|||
d |
d < | | ¦ « |dz
d| ||¦ « d } | €ŒRd||
d |
d < t | |¦ « }t ||¦ « }||k r nŒ’|| fS )Nr F)Ú check_endr‘ r� T)Ú evaluate_fnÚfloatr Úget_valid_movesr… rQ ÚmaxÚmin) r r r| r’ r} ÚalphaÚbetaÚeval_Ú
best_scoreÚ best_mover7 Úscores r r… zMinimaxPlayer.minimaxÎ s € ğ × Ò Ñ'Ô'ˆØ Рؘ$�;Ğ Ø �AŠ:ˆ:Ø×#Ò# E°UĞ#Ñ;Ô;¸TĞAĞ Aà ğ # )İ ™,œ,˜ ˆJØ ˆIØ Ô)×9Ò9¸%Ñ@Ô@ğ
ğ
�Ø*0��d˜1”g”˜t AœwÑ'Ø Ÿ š U§Z¢Z¡\¤\°5¸1±9¸eÀfÀWÈeĞUYÑZÔZĞ[\Ô]�à*+��d˜1”g”˜t AœwÑ'ؘ:Ò%Ğ%Ø $�Iİ ¨
Ñ 3Ô 3�
ݘE :Ñ.Ô.�à˜D’=�=Ø�Eğ !ğ ˜yĞ(Ğ (å ˜u™ œ ˆJØ ˆIØ Ô)×9Ò9¸%Ñ@Ô@ğ
ğ
�Ø*0��d˜1”g”˜t AœwÑ'Ø Ÿ š U§Z¢Z¡\¤\°5¸1±9¸dÀVÀGÈUĞTXÑYÔYĞZ[Ô\�Ø�=Ø à*+��d˜1”g”˜t AœwÑ'İ ¨
Ñ 3Ô 3�
ݘ4 Ñ,Ô,�à˜5’=�=Ø�Eğ !ğ ˜yĞ(Ğ (r c óT — t | j ¦ « fi | j ¤�S r )r� r r r r s r r z!MinimaxPlayer.make_fresh_instanceü s) € İ ˜TÔ .× BÒ BÑ DÔ DĞTĞTÈÌ ĞTĞTĞTr c ó — || _ d S r r: r" s r r# z MinimaxPlayer.set_game_managerÿ r; r N)r$ r% r& r r r( r) r* r+ r r– Úboolr… r r# r r r r� r� ½ sŞ € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ X]ĞW\Ğ]bÑWcÔWcĞVcØ�U˜5‘\”\ğ,)ğ ,)˜RœZğ ,)°ğ ,)¸Tğ ,)È3ğ ,)Ø&+ğ,)ğ ,)ğ ,)ğ ,)ğ\Uğ Uğ Uğ)ğ )ğ )ğ )ğ )r r� )rx rQ Úabcr r ru Únumpyr( Ú$mu_alpha_zero.AlphaZero.Network.nnetr Ú!mu_alpha_zero.Game.tictactoe_gamer Ú mu_alpha_zero.General.az_gamer r r. r> rW rj rs r� r r r ú<module>r¨ s· ğØ
€
€
€
Ø € € € Ø #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #à € € € Ø Ğ Ğ Ğ à =Ğ =Ğ =Ğ =Ğ =Ğ =Ø BĞ BĞ BĞ BĞ BĞ BØ 7Ğ 7Ğ 7Ğ 7Ğ 7Ğ 7ğ
ğ
ğ
ğ
ğ
ˆSñ
ô
ğ
ğ6)ğ )ğ )ğ )ğ )�6ñ )ô )ğ )ğ,#)ğ #)ğ #)ğ #)ğ #)�ñ #)ô #)ğ #)ğL!)ğ !)ğ !)ğ !)ğ !)˜ñ !)ô !)ğ !)ğH)ğ )ğ )ğ )ğ )�&ñ )ô )ğ )ğ( )ğ )ğ )ğ )ğ )˜ñ )ô )ğ )ğBC)ğ C)ğ C)ğ C)ğ C)�Fñ C)ô C)ğ C)ğ C)ğ C)r | 15,725 | Python | .py | 55 | 284.672727 | 1,489 | 0.309936 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,514 | arena.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Arena/__pycache__/arena.cpython-311.pyc | §
kŒ¢fH ã ó~ — d dl Z d dlZd dlmZ d dlmZ d dlmZ d dlm Z d dl
m Z m Z d dl
mZ G d„ d e¦ « ZdS )
é N)ÚPlayer)Ú GeneralArena)Ú HookManager)ÚHookAt)Ú scale_actionÚ
resize_obs)ÚAlphaZeroConfigc óv — e Zd Z ddedepddefd„Z dded ed
ed ed ed
ede pddede
eeef fd„Z d„ Z dS )ÚArenaNFÚalpha_zero_configÚ hook_managerÚ
state_managedc ól — || _ || _ || _ |�|n
t ¦ « | _ || _ d S )N)Ú game_managerr Údevicer r
r )Úselfr r r r
r s úR/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Arena/arena.pyÚ__init__zArena.__init__ s< € à (ˆÔØ *ˆÔØ ˆŒ Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ!2ˆÔ Ğ Ğ ó é Úplayer1Úplayer2Únum_games_to_playÚnum_mc_simulationsÚ
one_playerÚ start_playerÚ
add_to_kwargsÚdebugÚreturnc
óÔ
— ddddœ} | j j }
|r|} n|dz } | j rl|j dk r| j |j _ |j dk r| j |j _ | | j ¦ « | | j ¦ « t |¦ « D �]²} | | k r|}
n| }
||
| j |
| j j dœ}|�|
|¦ « | j j s| j ¦ « }n| j |
¬¦ « }| j j
r�| j j r„ |j j | j |d¦ « d¦ « |j j | j |d ¦ « d ¦ « n# t$ $ r Y nw xY w|j d
k r|j d¦ « |j d
k r|j d¦ « | j ¦ « |
dk r |j |fi |¤�}n |j |fi |¤�}| j | | j j t4 t6 j |||
f¬ ¦ « | j s8| j |||
¦ « }| j |
|¦ « }n;| j ||
¦ « d }| j |
¦ « }| j ¦ « | j j
�ret? |t@ ¦ « r'tC || j " ¦ « ¦ « nCtC |d |j# d z |d z | j " ¦ « ¦ « } |
dk rg|j dk r!|j j $ ||d ¦ « |j j $ | j |d¦ « |d¦ « nf|j dk r!|j j $ ||d¦ « |j j $ | j |d ¦ « |d ¦ « n# t$ $ r Y nw xY w|�Ã|dk r(|
dk r| d
xx dz
cc< nO| dxx dz
cc< n>|d k r(|
dk r| dxx dz
cc< n!| d
xx dz
cc< n| dxx dz
cc< |j dk s |j dk r|rtK j& d¦ « tO j( | d
| d | d dœ¦ « n |
d z }
|
|d< �ŒM�Œ´| d
| d | d fS )aÌ
Pit two players against each other for a given number of games and gather the results.
:param start_player: Which player should start the game.
:param one_player: If True always only the first player will start the game.
:param player1:
:param player2:
:param num_games_to_play:
:param num_mc_simulations:
:return: number of wins for player1, number of wins for player2, number of draws
r )Úwins_p1Úwins_p2Údrawsé Ú NetPlayer)Únum_simulationsÚcurrent_playerr ÚtauÚunravelN)Úplayerr éÿÿÿÿÚ
NetworkPlayerT)Úargsr! r" r# Ú HumanPlayergš™™™™™É?r' ))r Ú arena_taur Únamer Úmonte_carlo_tree_searchÚset_game_managerÚranger r) ÚupdateÚrequires_player_to_resetÚresetÚarena_running_muzeroÚenable_frame_bufferÚbufferÚ init_bufferÚ get_state_for_passive_playerÚAttributeErrorÚ step_rootÚrenderÚ choose_mover
Úprocess_hook_executesÚpitÚ__name__Ú__file__r ÚMIDDLEÚget_next_stateÚ game_resultÚ
isinstanceÚintr Úget_num_actionsÚshapeÚ add_frameÚtimeÚsleepÚwandbÚlog)r r r r r r r r r Úresultsr( Únum_games_per_playerÚgamer' ÚkwargsÚstateÚmoveÚstatuss r rA z Arena.pit sš € ğ ¨A¸Ğ:Ğ:ˆØÔ$Ô.ˆØ ğ :Ø#4Ğ Ğ à#4¸Ñ#9Ğ à Ô ğ 8ØŒ|˜{Ò*Ğ*Ø?CÔ?P�Ô/Ô<ØŒ|˜{Ò*Ğ*Ø?CÔ?P�Ô/Ô<à × $Ò $ TÔ%6Ñ 7Ô 7Ğ 7Ø × $Ò $ TÔ%6Ñ 7Ô 7Ğ 7åĞ+Ñ,Ô,ğ V :ñ V :ˆDØĞ*Ò*Ğ*Ø!-��à". �à);È~ĞimÔitØ ¨TÔ-CÔ-KğMğ MˆFà Ğ(Ø—
’
˜mÑ,Ô,Ğ,ØÔ)ÔBğ
GØ Ô)×/Ò/Ñ1Ô1��à Ô)×/Ò/°~Ğ/ÑFÔF�àÔ%Ô:ğ
¸tÔ?UÔ?iğ
ğØÔ3Ô:×FÒFØ Ô)×FÒFÀuÈaÑPÔPØñô ğ ğ Ô3Ô:×FÒFØ Ô)×FÒFÀuÈbÑQÔQĞSUñWô Wğ Wğ Wøå%ğ ğ ğ Ø�Dğøøøğ Œ|˜Ò.Ğ.ØÔ/×9Ò9¸$Ñ?Ô?Ğ?ØŒ|˜Ò.Ğ.ØÔ/×9Ò9¸$Ñ?Ô?Ğ?ğ8
:ØÔ!×(Ò(Ñ*Ô*Ğ*Ø! QÒ&Ğ&Ø.˜7Ô.¨uĞ?Ğ?¸Ğ?Ğ?�D�Dà.˜7Ô.¨uĞ?Ğ?¸Ğ?Ğ?�DØÔ!×7Ò7¸¸d¼hÔ>OÕQYÕ[aÔ[hØ>BÀFÈNĞ=[ğ 8ñ ]ô ]ğ ]àÔ)ğ KØ Ô -× <Ò <¸UÀDÈ.Ñ YÔ Y�EØ !Ô .× :Ò :¸>È5Ñ QÔ Q�F�Fà Ô -× <Ò <¸TÀ>Ñ RÔ RĞSTÔ U�EØ !Ô .× :Ò :¸>Ñ JÔ J�FØÔ!×(Ò(Ñ*Ô*Ğ*ØÔ)Ô>ñ İV`ĞaeÕgjÑVkÔVkğ n�<¨¨dÔ.?×.OÒ.OÑ.QÔ.QÑRÔRĞRİ$ T¨!¤W¨u¬{¸1¬~Ñ%=ÀÀQÄÑ%GÈÔIZ×IjÒIjÑIlÔIlÑmÔmğ ğ Ø)¨QÒ.Ğ.Ø &œ|¨{Ò :Ğ :Ø 'Ô ?Ô F× PÒ PĞQVĞW[Ğ\^Ñ _Ô _Ğ _Ø #Ô ;Ô B× LÒ LÈTÔM^×M{ÒM{ğ }Bğ DEñ NFô NFğ GKğ LMñ Nô Nğ Nğ Nà &œ|¨{Ò :Ğ :Ø 'Ô ?Ô F× PÒ PĞQVĞW[Ğ\]Ñ ^Ô ^Ğ ^Ø #Ô ;Ô B× LÒ LÈTÔM^×M{ÒM{ğ }Bğ DFñ NGô NGğ HLğ MOñ Pô Pğ Pøøå)ğ ğ ğ à ˜ğ øøøğ Ğ%Ø ’{�{Ø)¨QÒ.Ğ.Ø # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à 2š ˜ Ø)¨QÒ.Ğ.Ø # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à Ğ(Ğ(Ô(¨AÑ-Ğ(Ğ(Ñ(ğ œ ¨
Ò5Ğ5¸¼ È Ò9VĞ9VĞ\aĞ9Vİ œ
3™œ˜å”I¨'°)Ô*<ÈĞQZÔI[ĞfmĞnuÔfvĞ wĞ wÑxÔxĞxØà "Ñ$�Ø+9�Ğ'Ñ(ñq8
:ñj ğ �yÔ! 7¨9Ô#5°w¸wÔ7GĞGĞGs& Ä6A2F)Æ)
F6Æ5F6Î CQ0Ñ0
Q=Ñ<Q=c ó$ — t d¦ « }|S )Nz Press any key to continue...)Úinput)r Úinpts r Ú
wait_keypresszArena.wait_keypress� s € İĞ3Ñ4Ô4ˆØˆ r )NF)Fr NF)
rB Ú
__module__Ú __qualname__r r Úboolr r rH ÚdictÚtuplerA rZ © r r r r sí € € € € € àQVğ3ğ 3¸ğ 3Ø *Ğ 2¨dğ3ØJNğ3ğ 3ğ 3ğ 3ğ puğuHğ uH˜6ğ uH¨Fğ uHÀsğ uHĞ`cğ uHØ ğuHØ47ğuHØLPÈLĞTXğuHØhlğuHà �#�s˜C�-Ô ğuHğ uHğ uHğ uHğnğ ğ ğ ğ r r )rL rN Ú%mu_alpha_zero.AlphaZero.Arena.playersr Úmu_alpha_zero.General.arenar Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr Úmu_alpha_zero.MuZero.utilsr r Úmu_alpha_zero.configr r r` r r ú<module>rg sÇ ğØ € € € à € € € à 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8à 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ø 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0ğBğ Bğ Bğ Bğ BˆLñ Bô Bğ Bğ Bğ Br | 7,879 | Python | .py | 29 | 268.103448 | 1,827 | 0.361992 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,515 | az_search_tree.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/MCTS/az_search_tree.py | import copy
from typing import Tuple, List, Any, Dict
import wandb
from multiprocess import set_start_method
set_start_method("spawn", force=True)
from multiprocess.pool import Pool
import numpy as np
import torch as th
from mu_alpha_zero.AlphaZero.MCTS.az_node import AlphaZeroNode
from mu_alpha_zero.AlphaZero.utils import augment_experience_with_symmetries, mask_invalid_actions
from mu_alpha_zero.Game.tictactoe_game import TicTacToeGameManager
from mu_alpha_zero.General.memory import GeneralMemoryBuffer
from mu_alpha_zero.General.network import GeneralNetwork
from mu_alpha_zero.General.search_tree import SearchTree
from mu_alpha_zero.Hooks.hook_manager import HookManager
from mu_alpha_zero.Hooks.hook_point import HookAt
from mu_alpha_zero.config import AlphaZeroConfig
from mu_alpha_zero.mem_buffer import SingleGameData, DataPoint
from mu_alpha_zero.shared_storage_manager import SharedStorage
class McSearchTree(SearchTree):
def __init__(self, game_manager: TicTacToeGameManager, alpha_zero_config: AlphaZeroConfig,
hook_manager: HookManager or None = None):
self.game_manager = game_manager
self.alpha_zero_config = alpha_zero_config
self.hook_manager = hook_manager if hook_manager is not None else HookManager()
self.root_node = None
def play_one_game(self, network: GeneralNetwork, device: th.device) -> tuple[
list | list[tuple[int | Any, Any, float, int]], int, int, int]:
"""
Plays a single game using the Monte Carlo Tree Search algorithm.
Args:
network: The neural network used for searching and evaluating moves.
device: The device (e.g., CPU or GPU) on which the network is located.
Returns:
A tuple containing the game history, and the number of wins, losses, and draws.
The game history is a list of tuples, where each tuple contains:
- The game state multiplied by the current player
- The policy vector (probability distribution over moves)
- The game result (1 for a win, -1 for a loss, 0 for a draw)
- The current player (-1 or 1)
"""
# tau = self.args["tau"]
state = self.game_manager.reset()
current_player = 1
game_history = []
# game_data = SingleGameData()
results = {"1": 0, "-1": 0, "D": 0}
move_number = 0
tau = self.alpha_zero_config.tau
while True:
if move_number > self.alpha_zero_config.zero_tau_after_first_n_moves != 0:
tau = 0
pi, _ = self.search(network, state, current_player, device)
move = self.game_manager.select_move(pi, tau=tau)
# self.step_root([move])
self.step_root(None)
# pi = [x for x in pi.values()]
game_history.append((state * current_player, pi, None, current_player,
self.game_manager.get_invalid_actions(state, current_player)))
state = self.game_manager.get_next_state(state, self.game_manager.network_to_board(move), current_player)
r = self.game_manager.game_result(current_player, state)
if r is not None:
if r == current_player:
results["1"] += 1
elif -1 < r < 1:
results["D"] += 1
else:
results["-1"] += 1
if -1 < r < 1:
game_history = [(x[0], x[1], r, x[3], x[4]) for x in game_history]
else:
game_history = [(x[0], x[1], r * current_player * x[3], x[3], x[4]) for x in game_history]
break
current_player *= -1
move_number += 1
# game_history = make_channels(game_history)
if self.alpha_zero_config.augment_with_symmetries:
game_history = self.game_manager.augment_game_history_with_symmetries(game_history)
self.hook_manager.process_hook_executes(self, self.play_one_game.__name__, __file__, HookAt.TAIL,
args=(game_history, results))
# for state, pi, r, player, move_mask in game_history:
# game_data.add_data_point(DataPoint(pi, r, None, None, player, state, move_mask))
return game_history, results["1"], results["-1"], results["D"]
def search(self, network, state, current_player, device, tau=None):
"""
Perform a Monte Carlo Tree Search on the current state starting with the current player.
:param tau:
:param network:
:param state:
:param current_player:
:param device:
:return:
"""
num_simulations = self.alpha_zero_config.num_simulations
if tau is None:
tau = self.alpha_zero_config.tau
if self.root_node is None:
self.root_node = AlphaZeroNode(current_player, times_visited_init=0)
state_ = self.game_manager.get_canonical_form(state, current_player)
# state_ = make_channels_from_single(state_)
state_ = th.tensor(state_, dtype=th.float32, device=device).unsqueeze(0)
probabilities, v = network.predict(state_, muzero=False)
if self.alpha_zero_config.add_dirichlet_noise:
probabilities = (
1 - self.alpha_zero_config.dirichlet_alpha) * probabilities + self.alpha_zero_config.dirichlet_alpha * np.random.dirichlet(
[0.03] * len(probabilities))
probabilities = mask_invalid_actions(probabilities,
self.game_manager.get_invalid_actions(state.copy(), current_player))
probabilities = probabilities.flatten().tolist()
self.root_node.expand(state, probabilities)
for simulation in range(num_simulations):
current_node = self.root_node
path = [current_node]
action = None
while current_node.was_visited():
current_node, action = current_node.get_best_child(c=self.alpha_zero_config.c)
if current_node is None: # This was for testing purposes
th.save(self.root_node, "root_node.pt")
th.save(network.state_dict(), f"network_none_checkpoint_{current_player}.pt")
raise ValueError("current_node is None")
path.append(current_node)
# leaf node reached
next_state = self.game_manager.get_next_state(current_node.parent().state.copy(),
self.game_manager.network_to_board(action),
current_node.parent().current_player)
next_state_ = self.game_manager.get_canonical_form(next_state, current_node.current_player)
v = self.game_manager.game_result(current_node.current_player, next_state)
if v is None:
# next_state_ = make_channels_from_single(next_state_)
next_state_ = th.tensor(next_state_, dtype=th.float32, device=device).unsqueeze(0)
probabilities, v = network.predict(next_state_, muzero=False)
probabilities = mask_invalid_actions(probabilities, self.game_manager.get_invalid_actions(next_state,
current_node.current_player))
v = v.flatten().tolist()[0]
probabilities = probabilities.flatten().tolist()
current_node.expand(next_state, probabilities)
self.backprop(v, path)
self.hook_manager.process_hook_executes(self, self.search.__name__, __file__, HookAt.TAIL,
args=(tau, self.root_node))
return self.root_node.get_self_action_probabilities(), None
def backprop(self, v, path):
"""
Backpropagates the value `v` through the search tree, updating the relevant nodes.
Args:
v (float): The value to be backpropagated.
path (list): The path from the leaf node to the root node.
Returns:
None
"""
for node in reversed(path):
v *= -1
node.total_value += v
node.update_q(v)
node.times_visited += 1
def step_root(self, actions: list | None) -> None:
if actions is not None:
if self.root_node is not None:
if not self.root_node.was_visited():
return
for action in actions:
self.root_node = self.root_node.children[action]
self.root_node.parent = None
else:
# reset root node
self.root_node = None
def make_fresh_instance(self):
return McSearchTree(self.game_manager.make_fresh_instance(), self.alpha_zero_config)
def self_play(self, net: GeneralNetwork, device: th.device, num_games: int, memory: GeneralMemoryBuffer) -> tuple[
int, int, int]:
wins_p1, wins_p2, draws = 0, 0, 0
for game in range(num_games):
game_results, wins_p1_, wins_p2_, draws_ = self.play_one_game(net, device)
wins_p1 += wins_p1_
wins_p2 += wins_p2_
draws += draws_
memory.add_list(game_results)
return wins_p1, wins_p2, draws
@staticmethod
def parallel_self_play(nets: list, trees: list, memory: GeneralMemoryBuffer, device: th.device, num_games: int,
num_jobs: int):
with Pool(num_jobs) as p:
if not memory.is_disk:
results = p.starmap(p_self_play,
[(nets[i], trees[i], copy.deepcopy(device), num_games // num_jobs, None) for i in
range(len(nets))])
else:
results = p.starmap(p_self_play, [
(nets[i], trees[i], copy.deepcopy(device), num_games // num_jobs, copy.deepcopy(memory)) for i in
range(len(nets))])
wins_p1, wins_p2, draws = 0, 0, 0
for result in results:
wins_p1 += result[0]
wins_p2 += result[1]
draws += result[2]
if not memory.is_disk:
memory.add_list(result[3])
return wins_p1, wins_p2, draws
@staticmethod
def start_continuous_self_play(nets: list, trees: list, shared_storage: SharedStorage, device: th.device,
config: AlphaZeroConfig, num_jobs: int, num_worker_iters: int) -> Pool:
pool = Pool(num_jobs)
for i in range(num_jobs):
pool.apply_async(c_p_self_play, args=(
nets[i], trees[i], copy.deepcopy(device), config, i, shared_storage, num_worker_iters // num_jobs
))
return pool
def p_self_play(net, tree, device, num_games, memory):
wins_p1, wins_p2, draws = 0, 0, 0
data = []
for game in range(num_games):
game_results, wp1, wp2, ds = tree.play_one_game(net, device)
wins_p1 += wp1
wins_p2 += wp2
draws += ds
if memory is not None:
memory.add_list(game_results)
else:
data.extend(game_results)
if memory is None:
return wins_p1, wins_p2, draws, data
return wins_p1, wins_p2, draws
def c_p_self_play(net, tree, device, config: AlphaZeroConfig, p_num: int, shared_storage: SharedStorage,
num_worker_iters: int):
if p_num == 0:
wandb.init(project=config.wandbd_project_name, name="Self play")
net = net.to(device)
for iter_ in range(num_worker_iters):
if shared_storage.get_experimental_network_params() is None:
params = shared_storage.get_stable_network_params()
else:
params = shared_storage.get_experimental_network_params()
# params = shared_storage.get_stable_network_params()
net.load_state_dict(params)
net.eval()
game_results, wins_p1, wins_p2, draws = tree.play_one_game(net, device)
shared_storage.add_list(game_results)
if p_num == 0:
wandb.log({"Iteration": iter_, "Wins p1": wins_p1, "Wins p2": wins_p2, "Draws": draws})
| 12,367 | Python | .py | 238 | 39.344538 | 159 | 0.589271 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,516 | az_node.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/MCTS/az_node.py | import math
import weakref
import numpy as np
from mu_alpha_zero.Game.tictactoe_game import TicTacToeGameManager as GameManager
class AlphaZeroNode:
"""
This class defines a node in the search tree. It stores all the information needed for DeepMind's AlphaZero algorithm.
"""
def __init__(self, current_player, select_probability=0, parent=None, times_visited_init=0):
self.times_visited = times_visited_init
self.was_init_with_zero_visits = times_visited_init == 0
self.children = {}
self.parent = weakref.ref(parent) if parent is not None else None
self.select_probability = select_probability
self.q = None
self.current_player = current_player
self.state = None
self.total_value = 0
def expand(self, state, action_probabilities) -> None:
"""
Expands the newly visited node with the given action probabilities and state.
:param state: np.ndarray of shape (board_size, board_size) representing the state current game board.
:param action_probabilities: list of action probabilities for each action.
:return: None
"""
self.state = state.copy() # copying here is probably not necessary.
for action, probability in enumerate(action_probabilities):
node = AlphaZeroNode(self.current_player * (-1), select_probability=probability, parent=self)
self.children[action] = node
def was_visited(self):
return len(self.children) > 0
def update_q(self, v):
# Based on DeepMind's AlphaZero paper.
if self.q is None:
self.q = v
else:
self.q = (self.times_visited * self.q + v) / (self.times_visited + 1)
def get_best_child(self, c=1.5):
best_utc = -float("inf")
best_child = None
best_action = None
valids_for_state = np.where(self.state != 0, -5, self.state)
valids_for_state = np.where(valids_for_state == 0, 1, valids_for_state)
utcs = []
for action, child in self.children.items():
action_ = np.unravel_index(action, self.state.shape)
if valids_for_state[action_] != 1:
continue
child_utc = child.calculate_utc(c=c)
utcs.append(child_utc)
if child_utc > best_utc:
best_utc = child_utc
best_child = child
best_action = action
printable_children = [[child.state, child.times_visited, child.q, child.select_probability] for child in
self.children.values()]
if best_child is None:
# This was for testing purposes
print("Best child is None. Possibly important info:\n", self.state, "\n",
valids_for_state, printable_children, self.was_visited(), utcs,
file=open("important_info"
".txt", "w"))
# raise Exception("Best child is None,terminating.")
return best_child, best_action
def calculate_utc(self, c=1.5):
parent = self.parent()
if self.q is None:
# Inspiration taken from https://github.com/suragnair/alpha-zero-general/blob/master/MCTS.py
utc = c * self.select_probability * math.sqrt(parent.times_visited + 1e-8)
else:
utc = self.q + c * (
self.select_probability * ((math.sqrt(parent.times_visited)) / (1 + self.times_visited)))
return utc
def get_self_value(self):
return self.total_value / self.times_visited if self.times_visited > 0 else 0
def get_self_action_probabilities(self):
total_times_visited = self.times_visited
action_probs = {}
for action, child in self.children.items():
action_probs[action] = child.times_visited / total_times_visited
return action_probs
def get_latent(self):
return self.state
| 3,970 | Python | .py | 83 | 37.626506 | 122 | 0.619798 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,517 | az_node.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/MCTS/__pycache__/az_node.cpython-311.pyc | §
’Ô†f‚ ã óD — d dl Z d dlZd dlZd dlmZ G d„ d¦ « ZdS )é N)ÚTicTacToeGameManagerc óP — e Zd ZdZdd„Zdd„Zd„ Zd„ Zdd
„Zdd „Z d „ Z
d
„ Z d„ Z dS )Ú
AlphaZeroNodez€
This class defines a node in the search tree. It stores all the information needed for DeepMind's AlphaZero algorithm.
r Nc ó¸ — || _ |dk | _ i | _ |�t j |¦ « nd | _ || _ d | _ || _ d | _ d| _
d S ©Nr ) Ú
times_visitedÚwas_init_with_zero_visitsÚchildrenÚweakrefÚrefÚparentÚselect_probabilityÚqÚcurrent_playerÚstateÚ total_value)Úselfr r r
Útimes_visited_inits úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/MCTS/az_node.pyÚ__init__zAlphaZeroNode.__init__
se € Ø /ˆÔØ);¸qÒ)@ˆÔ&؈Œ
Ø-3Ğ-?•g”k &Ñ)Ô)Ğ)ÀTˆŒ Ø"4ˆÔ ؈ŒØ ,ˆÔ؈Œ
Ø ˆÔĞĞó Úreturnc óª — | ¦ « | _ t |¦ « D ])\ }}t | j dz || ¬¦ « }|| j |< Œ*dS )a6
Expands the newly visited node with the given action probabilities and state.
:param state: np.ndarray of shape (board_size, board_size) representing the state current game board.
:param action_probabilities: list of action probabilities for each action.
:return: None
éÿÿÿÿ)r r
N)Úcopyr Ú enumerater r r
)r r Úaction_probabilitiesÚactionÚ probabilityÚnodes r ÚexpandzAlphaZeroNode.expand sg € ğ —Z’Z‘\”\ˆŒ
å#,Ğ-AÑ#BÔ#Bğ )ğ )Ñ ˆF�Kİ Ô!4¸Ñ!;ĞP[ĞdhĞiÑiÔiˆDØ$(ˆDŒM˜&Ñ !Ğ !ğ )ğ )r c ó2 — t | j ¦ « dk S r )Úlenr
©r s r Ú was_visitedzAlphaZeroNode.was_visited& s € İ�4”=Ñ!Ô! AÒ%Ğ%r c ój — | j € || _ d S | j | j z |z | j dz z | _ d S )Né )r r )r Úvs r Úupdate_qzAlphaZeroNode.update_q) s; € à Œ6ˆ>؈DŒFˆFˆFàÔ(¨4¬6Ñ1°AÑ5¸$Ô:LÈqÑ:PÑQˆDŒFˆFˆFr ç ø?c
óx — t d¦ « }d }d }t j | j dk d| j ¦ « }t j |dk d|¦ « }g }| j ¦ « D ]h\ }}t j || j j ¦ « } || dk rŒ1| |¬¦ « }
| |
¦ « |
|k r|
}|}|}Œid„ | j
¦ « D ¦ « } |€<t d| j d|| | ¦ « |t d d
¦ « ¬ ¦ « ||fS ) NÚinfr éûÿÿÿr' )Úcc óB — g | ] }|j |j |j |j g‘Œ S © )r r r r )Ú.0Úchilds r ú
<listcomp>z0AlphaZeroNode.get_best_child.<locals>.<listcomp>B s8 € ğ 6ğ 6ğ 6Ğhm˜uœ{¨EÔ,?ÀÄÈ%ÔJbĞ cğ 6ğ 6ğ 6r z-Best child is None. Possibly important info:
Ú
zimportant_info.txtÚw)Úfile)ÚfloatÚnpÚwherer r
ÚitemsÚ
unravel_indexÚshapeÚ
calculate_utcÚappendÚvaluesÚprintr% Úopen) r r. Úbest_utcÚ
best_childÚ best_actionÚvalids_for_stateÚutcsr r2 Úaction_Ú child_utcÚprintable_childrens r Úget_best_childz AlphaZeroNode.get_best_child0 so € ݘ%‘L”L�=ˆØˆ
؈ İ œ8 D¤J°!¢O°R¸¼ÑDÔDĞİ œ8Ğ$4¸Ò$9¸1Ğ>NÑOÔOĞØˆØ !œ]× 0Ò 0Ñ 2Ô 2ğ %ğ %‰MˆF�EİÔ& v¨t¬zÔ/?Ñ@Ô@ˆGØ Ô(¨AÒ-Ğ-ØØ ×+Ò+¨aĞ+Ñ0Ô0ˆIØ �KŠK˜ Ñ "Ô "Ğ "ؘ8Ò#Ğ#Ø$�Ø "�
Ø $� øğ 6ğ 6Ø "œm× 2Ò 2Ñ 4Ô 4ğ 6ñ 6ô 6Ğà Ğ å ĞBÀDÄJĞPTØ"Ğ$6¸×8HÒ8HÑ8JÔ8JÈDİğ #Ø$'ñ)ô )ğ
*ñ
*ô
*ğ
*ğ
˜;Ğ&Ğ&r c óú — | ¦ « }| j €(|| j z t j |j dz ¦ « z }n7| j || j t j |j ¦ « d| j z z z z z }|S )Ng:Œ0â�yE>r' )r
r r ÚmathÚsqrtr )r r. r
Úutcs r r= zAlphaZeroNode.calculate_utcM s€ € Ø—’‘ ” ˆØ Œ6ˆ>à�dÔ-Ñ-µ´ ¸&Ô:NĞQUÑ:UÑ0VÔ0VÑVˆCˆCà”&˜1ØÔ+µ´ ¸&Ô:NÑ0OÔ0OĞTUĞX\ÔXjÑTjÑ/kÑlñ nñ nˆCğ ˆ
r c ó: — | j dk r| j | j z ndS r )r r r$ s r Úget_self_valuez AlphaZeroNode.get_self_valueX s& € Ø8<Ô8JÈQÒ8NĞ8NˆtÔ $Ô"4Ñ4Ğ4ĞTUĞUr c óp — | j }i }| j ¦ « D ]\ }}|j |z ||< Œ|S ©N)r r
r: )r Útotal_times_visitedÚ action_probsr r2 s r Ú get_self_action_probabilitiesz+AlphaZeroNode.get_self_action_probabilities[ sO € à "Ô 0Ğ؈ Ø !œ]× 0Ò 0Ñ 2Ô 2ğ Mğ M‰MˆF�EØ#(Ô#6Ğ9LÑ#LˆL˜Ñ Ğ àĞr c ó — | j S rR )r r$ s r Ú
get_latentzAlphaZeroNode.get_latentd s
€ ØŒzĞr )r Nr )r N)r* )
Ú__name__Ú
__module__Ú __qualname__Ú__doc__r r! r% r) rJ r= rP rU rW r0 r r r r s¿ € € € € € ğğ ğ ğ ğ ğ ğ )ğ )ğ )ğ )ğ &ğ &ğ &ğRğ Rğ Rğ'ğ 'ğ 'ğ 'ğ: ğ ğ ğ ğVğ Vğ Vğ ğ ğ ğğ ğ ğ ğ r r )rL r Únumpyr8 Ú!mu_alpha_zero.Game.tictactoe_gamer Ú GameManagerr r0 r r ú<module>r_ sq ğØ € € € Ø €€€Ø Ğ Ğ Ğ à QĞ QĞ QĞ QĞ QĞ Qğ]ğ ]ğ ]ğ ]ğ ]ñ ]ô ]ğ ]ğ ]ğ ]r | 5,913 | Python | .py | 35 | 166.514286 | 775 | 0.387651 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,518 | az_search_tree.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/MCTS/__pycache__/az_search_tree.cpython-311.pyc | §
’Ô†fe( ã ó¼ — d dl Z d dlmZ d dlZd dlZd dlmZ d dl m
Z
m Z d dl m
Z
d dlmZ d dlmZ d dlmZ d d lmZ d d
lmZ d d lmZ G d „ d
e¦ « Zd„ ZdS )é N)ÚPool)Ú
AlphaZeroNode)Ú"augment_experience_with_symmetriesÚmask_invalid_actions)ÚTicTacToeGameManager)ÚGeneralMemoryBuffer)ÚGeneralNetwork)Ú
SearchTree)Ú HookManager)ÚHookAt)ÚAlphaZeroConfigc
óø — e Zd Z ddededepdfd„Zdedej de
e e e e f fd „Z
dd
„Zd „ Zd e dz ddfd
„Zd„ Zdedej de dede
e e e f f
d„Zede de dedej de de f d„¦ « ZdS )Ú McSearchTreeNÚ game_managerÚalpha_zero_configÚ hook_managerc ó^ — || _ || _ |�|n
t ¦ « | _ d | _ d S ©N)r r r r Ú root_node)Úselfr r r s úZ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/MCTS/az_search_tree.pyÚ__init__zMcSearchTree.__init__ s2 € à (ˆÔØ!2ˆÔ Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ ˆŒˆˆó ÚnetworkÚdeviceÚreturnc ó¨ ‡ ‡
— | j ¦ « }dŠ g }ddddœ} | ||‰ |¦ « \ }}| j || j j ¬¦ « }| d¦ « | |‰ z |d‰ f¦ « | j || j |¦ « ‰ ¦ « }| j
‰ |¦ « Š
‰
�w‰
‰ k r|dxx dz
cc< n1d‰
cxk rdk rn n|d xx dz
cc< n|d
xx dz
cc< d‰
cxk rdk rn nˆ
fd „|D ¦ « }nˆ ˆ
fd „|D ¦ « }n‰ dz Š �ŒAt || j j ¦ « }| j
| | j j t" t$ j ||f¬
¦ « ||d |d
|d fS )a¸
Plays a single game using the Monte Carlo Tree Search algorithm.
Args:
network: The neural network used for searching and evaluating moves.
device: The device (e.g., CPU or GPU) on which the network is located.
Returns:
A tuple containing the game history, and the number of wins, losses, and draws.
The game history is a list of tuples, where each tuple contains:
- The game state multiplied by the current player
- The policy vector (probability distribution over moves)
- The game result (1 for a win, -1 for a loss, 0 for a draw)
- The current player (-1 or 1)
é r )Ú1ú-1ÚDT)ÚtauNr éÿÿÿÿr! r c ó@ •— g | ]}|d |d ‰|d f‘ŒS ©r r é © )Ú.0ÚxÚrs €r ú
<listcomp>z.McSearchTree.play_one_game.<locals>.<listcomp>A s. ø€ Ğ#PĞ#PĞ#P¸a Q q¤T¨1¨Q¬4°°A°a´DĞ$9Ğ#PĞ#PĞ#Pr c óX •— g | ]&}|d |d ‰‰z |d z |d f‘Œ'S r% r' )r( r) Úcurrent_playerr* s €€r r+ z.McSearchTree.play_one_game.<locals>.<listcomp>C s= ø€ Ğ#hĞ#hĞ#hĞVW Q q¤T¨1¨Q¬4°°^Ñ1CÀaÈÄdÑ1JÈAÈaÌDĞ$QĞ#hĞ#hĞ#hr ©Úargs)r ÚresetÚsearchÚ select_mover r" Ú step_rootÚappendÚget_next_stateÚnetwork_to_boardÚ game_resultr Ú
board_sizer Úprocess_hook_executesÚ
play_one_gameÚ__name__Ú__file__r ÚTAIL) r r r ÚstateÚ game_historyÚresultsÚpiÚ_Úmover- r* s @@r r: zMcSearchTree.play_one_game s/ øø€ ğ" Ô!×'Ò'Ñ)Ô)ˆØˆØˆ Ø ¨Ğ+Ğ+ˆğ !Ø—K’K ¨°ÀÑGÔG‰EˆB�ØÔ$×0Ò0°¸Ô8NÔ8RĞ0ÑSÔSˆDà �NŠN˜4Ñ Ô Ğ à × Ò ¨ Ñ!7¸¸TÀ>Ğ RÑ SÔ SĞ SØÔ%×4Ò4°U¸DÔ<M×<^Ò<^Ğ_cÑ<dÔ<dĞftÑuÔuˆEØÔ!×-Ò-¨n¸eÑDÔDˆA؈}ؘÒ&Ğ&ؘC�L�L”L AÑ%�L�L‘L�Lؘ!�Z�Z’Z�Z˜a’Z�Z�Z�Z�ZؘC�L�L”L AÑ%�L�L‘L�Là˜D�M�M”M QÑ&�M�M‘Mà˜�:�:’:�:˜A’:�:�:�:�:Ø#PĞ#PĞ#PĞ#PÀ<Ğ#PÑ#PÔ#P�L�Là#hĞ#hĞ#hĞ#hĞ#hĞ[gĞ#hÑ#hÔ#h�LØØ ˜bÑ ˆNñ- !õ2 :¸,ÈÔHYÔHdÑeÔeˆ Ø Ô×/Ò/°°dÔ6HÔ6QÕS[Õ]cÔ]hØ6BÀGĞ5Lğ 0ñ Nô Nğ Nà˜W Sœ\¨7°4¬=¸'À#¼,ĞFĞFr c ó� — | j j }|€ | j j }| j €t |d¬¦ « | _ | j ||¦ « }t j |t j |¬¦ «
d¦ « }| |d¬¦ « \ }} |t j
| j j g| j j z ¦ « dd¦ « z }t% || j | ¦ « |¦ « | j j ¦ « }| ¦ « ¦ « }| j ||¦ « t3 |¦ « D �]n}
| j } | g} d}
| ¦ « r¢| | j j ¬ ¦ « \ } }
| €Tt j | j d
¦ « t j | ¦ « d |› d �¦ « t? d
¦ « ‚| | ¦ « | ¦ « °¢| j ! | " ¦ « j# | j $ |
¦ « | " ¦ « j% ¦ « }| j || j% ¦ « }| j & | j% |¦ « } | €ït j |t j |¬¦ «
d¦ « }| |d¬¦ « \ }} t% || j || j% ¦ « | j j ¦ « }| ¦ « ¦ « d } | ¦ « ¦ « }| ||¦ « | ' | | ¦ « �Œp| j( ) | | j* j+ tX tZ j. || j f¬¦ « | j / ¦ « dfS )zó
Perform a Monte Carlo Tree Search on the current state starting with the current player.
:param tau:
:param network:
:param state:
:param current_player:
:param device:
:return:
Nr )Útimes_visited_init)Údtyper F)Úmuzeror r# )Úcz root_node.ptÚnetwork_none_checkpoint_z.ptzcurrent_node is Noner. )0r Únum_simulationsr" r r r Úget_canonical_formÚthÚtensorÚfloat32Ú unsqueezeÚpredictÚnpÚrandomÚ dirichletÚdirichlet_alphaÚnet_action_sizeÚreshaper Úget_invalid_actionsÚcopyr8 ÚflattenÚtolistÚexpandÚrangeÚ was_visitedÚget_best_childrH ÚsaveÚ
state_dictÚ
ValueErrorr4 r5 Úparentr> r6 r- r7 Úbackpropr r9 r1 r; r< r r= Ú get_self_action_probabilities)r r r> r- r r" rJ Ústate_Ú
probabilitiesÚvÚ
simulationÚ current_nodeÚpathÚactionÚ
next_stateÚ next_state_s r r1 zMcSearchTree.searchM s
€ ğ Ô0Ô@ˆØ ˆ;ØÔ(Ô,ˆCØ Œ>Ğ !İ *¨>ÈaĞ PÑ PÔ PˆDŒNØÔ"×5Ò5°e¸^ÑLÔLˆå”˜6¬¸FĞCÑCÔC×MÒMÈaÑPÔPˆØ"Ÿ?š?¨6¸%˜?Ñ@Ô@ш
�qØ%¬ ×(;Ò(;Ø
Ô
#Ô
3Ğ 4°tÔ7MÔ7]Ñ ]ñ)_ô )_ß_fÒ_fĞghØgiñ`kô `kñkˆ
õ -¨]Ø-1Ô->×-RÒ-RĞSX×S]ÒS]ÑS_ÔS_ĞaoÑ-pÔ-pØ-1Ô->Ô-IñKô Kˆ
ğ &×-Ò-Ñ/Ô/×6Ò6Ñ8Ô8ˆ
Ø Œ× Ò ˜e ]Ñ3Ô3Ğ3İ Ñ0Ô0ğ #ñ #ˆJØ œ>ˆLØ �>ˆD؈FØ ×*Ò*Ñ,Ô,ğ
*Ø'3×'BÒ'BÀTÔE[ÔE]Ğ'BÑ'^Ô'^Ñ$� ˜fØ Ğ'İ”G˜DœN¨NÑ;Ô;Ğ;İ”G˜G× .Ò .Ñ 0Ô 0Ğ2`È^Ğ2`Ğ2`Ğ2`ÑaÔaĞaİ$Ğ%;Ñ<Ô<Ğ<Ø— ’ ˜LÑ)Ô)Ğ)ğ
×*Ò*Ñ,Ô,ğ
*ğ Ô*×9Ò9¸,×:MÒ:MÑ:OÔ:OÔ:UØ:>Ô:K×:\Ò:\Ğ]cÑ:dÔ:dØ:F×:MÒ:MÑ:OÔ:OÔ:^ñ`ô `ˆJğ Ô+×>Ò>¸zÈ<ÔKfÑgÔgˆKØÔ!×-Ò-¨lÔ.IÈ:ÑVÔVˆA؈yå œi¨ ½2¼:ÈfĞ UÑ UÔ U× _Ò _Ğ`aÑ bÔ b� Ø#*§?¢?°;Àu ?Ñ#MÔ#MÑ �
˜qİ 4°]ÀDÔDU×DiÒDiĞjtØjvô kFñEGô EGà59Ô5FÔ5Qñ!Sô !S�
ğ —I’I‘K”K×&Ò&Ñ(Ô(¨Ô+�Ø -× 5Ò 5Ñ 7Ô 7× >Ò >Ñ @Ô @�
Ø ×#Ò# J°
Ñ>Ô>Ğ>à �MŠM˜!˜TÑ "Ô "Ğ "Ñ "à Ô×/Ò/°°d´kÔ6JÍHÕV\ÔVaØ69¸4¼>Ğ5Jğ 0ñ Lô Lğ LàŒ~×;Ò;Ñ=Ô=¸tĞCĞCr c ó� — t |¦ « D ]<}|dz }|xj |z
c_ | |¦ « |xj dz
c_ Œ=dS )a
Backpropagates the value `v` through the search tree, updating the relevant nodes.
Args:
v (float): The value to be backpropagated.
path (list): The path from the leaf node to the root node.
Returns:
None
r# r N)ÚreversedÚ total_valueÚupdate_qÚ
times_visited)r rg rj Únodes r rc zMcSearchTree.backprop‹ si € õ ˜T‘N”Nğ $ğ $ˆDØ
�‰GˆAØ Ğ Ô Ñ !Ğ Ô Ø �MŠM˜!Ñ Ô Ğ Ø Ğ Ô !Ñ #Ğ Ô Ğ ğ $ğ $r Úactionsc ó´ — |�N| j �E| j ¦ « sd S |D ]}| j j | | _ Œd | j _ d S d S d | _ d S r )r r] Úchildrenrb )r rt rk s r r3 zMcSearchTree.step_rootœ st € Ø Ğ ØŒ~Ğ)Ø”~×1Ò1Ñ3Ô3ğ Ø�FØ %ğ Eğ E�FØ%)¤^Ô%<¸VÔ%D�D”N�NØ(,�”Ô%Ğ%Ğ%ğ *Ğ)ğ "ˆDŒNˆNˆNr c óZ — t | j ¦ « | j ¦ « S r )r r Úmake_fresh_instancer )r s r rx z McSearchTree.make_fresh_instance¨ s$ € ݘDÔ -× AÒ AÑ CÔ CÀTÔE[Ñ\Ô\Ğ\r ÚnetÚ num_gamesÚmemoryc óº — d\ }}}t |¦ « D ]A}| ||¦ « \ } }
} } ||
z
}|| z
}|| z
}| | ¦ « ŒB|||fS ©N©r r r )r\ r: Úadd_list)
r ry r rz r{ Úwins_p1Úwins_p2ÚdrawsÚgameÚ game_resultsÚwins_p1_Úwins_p2_Údraws_s
r Ú self_playzMcSearchTree.self_play« s‚ € à")Ñ ˆ�˜%ݘ)Ñ$Ô$ğ *ğ *ˆDØ7;×7IÒ7IÈ#ÈvÑ7VÔ7VÑ 4ˆL˜( H¨fØ �xÑ ˆGØ �xÑ ˆGØ �V‰OˆEØ �OŠO˜LÑ )Ô )Ğ )Ğ )à˜ Ğ&Ğ&r ÚnetsÚtreesÚnum_jobsc
ó( ‡ ‡‡‡‡‡— t ‰¦ « 5 }‰j sF| t ˆˆ ˆˆˆfd„t t ‰ ¦ « ¦ « D ¦ « ¦ « }nF| t ˆˆˆ ˆˆˆfd„t t ‰ ¦ « ¦ « D ¦ « ¦ « }d d d ¦ « n # 1 swxY w Y d\ }} }
|D ]E} || d z
}| | d z
} |
| d z
}
‰j s‰ | d ¦ « ŒF|| |
fS )Nc ó` •— g | ]*}‰| ‰| t j ‰¦ « ‰‰z d f‘Œ+S r ©rX Údeepcopy)r( Úir r‰ rz r‹ rŠ s €€€€€r r+ z3McSearchTree.parallel_self_play.<locals>.<listcomp>½ sK ø€ ğ %7ğ %7ğ %7Ğqr d¨1¤g¨u°Q¬x½¼ ÀvÑ9NÔ9NĞPYĞ]eÑPeĞgkĞ%lğ %7ğ %7ğ %7r c ó„ •— g | ]<}‰| ‰| t j ‰¦ « ‰‰z t j ‰¦ « f‘Œ=S r' r� )r( r� r r{ r‰ rz r‹ rŠ s €€€€€€r r+ z3McSearchTree.parallel_self_play.<locals>.<listcomp>À sU ø€ ğ 2&ğ 2&ğ 2&Øqr�T˜!”W˜e Aœh¬
°fÑ(=Ô(=¸yÈHÑ?TÕVZÔVcĞdjÑVkÔVkĞlğ2&ğ 2&ğ 2&r r~ r r é r& )r Úis_diskÚstarmapÚ p_self_playr\ Úlenr ) r‰ rŠ r{ r rz r‹ Úpr@ r€ r� r‚ Úresults `````` r Úparallel_self_playz McSearchTree.parallel_self_play· sª øøøøøø€ õ �(‰^Œ^ğ '˜qØ”>ğ
'ØŸ)š)¥Kğ%7ğ %7ğ %7ğ %7ğ %7ğ %7ğ %7ğ %7İ%*3¨t©9¬9Ñ%5Ô%5ğ%7ñ %7ô %7ñ8ô 8��ğ Ÿ)š)¥Kğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&å�#˜d™)œ)Ñ$Ô$ğ2&ñ 2&ô 2&ñ 'ô '�ğ
'ğ 'ğ 'ñ 'ô 'ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'øøøğ 'ğ 'ğ 'ğ 'ğ #*Ñ ˆ�˜%Ø ğ +ğ +ˆFØ �v˜a”yÑ ˆGØ �v˜a”yÑ ˆGØ �V˜A”YÑ ˆEØ”>ğ
+Ø—’ q¤ Ñ*Ô*Ğ*øؘ Ğ&Ğ&s –BB6Â6B:Â=B:r )r; Ú
__module__Ú __qualname__r r
r r r rL r ÚtupleÚlistÚintr: r1 rc r3 rx r rˆ Ú staticmethodr™ r' r r r r sŒ € € € € € à59ğ ğ Ğ%9ğ Èoğ Ø *Ğ 2¨dğ ğ ğ ğ ğ1G ^ğ 1G¸R¼Yğ 1GÈ5ĞQUĞWZĞ\_ĞadĞQdÔKeğ 1Gğ 1Gğ 1Gğ 1Gğf<Dğ <Dğ <Dğ <Dğ|$ğ $ğ $ğ"
" ¨¡ğ
"°ğ
"ğ
"ğ
"ğ
"ğ]ğ ]ğ ]ğ
'˜^ğ
'°R´Yğ
'È3ğ
'ĞXkğ
'ĞpuØ ˆS�#ˆ
ôqğ
'ğ
'ğ
'ğ
'ğ ğ' ğ '¨dğ 'Ğ<Oğ 'ĞY[ÔYbğ 'Ğorğ 'Ø%(ğ'ğ 'ğ 'ñ „\ğ'ğ 'ğ 'r r c óş — d\ }}}g }t |¦ « D ]Y} | | |¦ « \ }
} } }
|| z
}|| z
}||
z
}|�| |
¦ « ŒD| |
¦ « ŒZ|€||||fS |||fS r} )r\ r: r Úextend)ry Útreer rz r{ r€ r� r‚ Údatarƒ r„ Úwp1Úwp2Údss r r• r• Í s´ € Ø %Ñ€GˆW�eØ
€Dİ�iÑ Ô ğ &ğ &ˆØ%)×%7Ò%7¸¸VÑ%DÔ%DÑ"ˆ �c˜3 Ø�3‰ˆØ�3‰ˆØ
�‰ ˆØ Ğ Ø �OŠO˜LÑ )Ô )Ğ )Ğ )à �KŠK˜ Ñ %Ô %Ğ %Ğ %Ø
€~ؘ ¨Ğ,Ğ,Ø �G˜UĞ "Ğ"r ) rX Úmultiprocessingr ÚnumpyrQ ÚtorchrL Ú$mu_alpha_zero.AlphaZero.MCTS.az_noder Ú mu_alpha_zero.AlphaZero.utilsr r Ú!mu_alpha_zero.Game.tictactoe_gamer Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr Ú!mu_alpha_zero.General.search_treer
Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr Úmu_alpha_zero.configr
r r• r' r r ú<module>r³ s, ğØ € € € Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ø Ğ Ğ Ğ à >Ğ >Ğ >Ğ >Ğ >Ğ >Ø bĞ bĞ bĞ bĞ bĞ bĞ bĞ bØ BĞ BĞ BĞ BĞ BĞ BØ <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0ğx'ğ x'ğ x'ğ x'ğ x'�:ñ x'ô x'ğ x'ğv#ğ #ğ #ğ #ğ #r | 14,571 | Python | .py | 85 | 167.270588 | 1,422 | 0.350549 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,519 | nnet.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Network/nnet.py | import atexit
import glob
import os
import time
from itertools import chain
import numpy as np
import torch as th
import torch.nn as nn
import torch.nn.functional as F
import wandb
from torch.nn.functional import mse_loss
from torch.nn.modules.module import T
from mu_alpha_zero.AlphaZero.checkpointer import CheckPointer
from mu_alpha_zero.AlphaZero.logger import Logger
from mu_alpha_zero.General.network import GeneralAlphZeroNetwork
from mu_alpha_zero.Hooks.hook_manager import HookManager
from mu_alpha_zero.Hooks.hook_point import HookAt
from mu_alpha_zero.config import AlphaZeroConfig, Config
from mu_alpha_zero.mem_buffer import MemBuffer
from mu_alpha_zero.shared_storage_manager import SharedStorage
class AlphaZeroNet(nn.Module, GeneralAlphZeroNetwork):
def __init__(self, in_channels: int, num_channels: int, dropout: float, action_size: int, linear_input_size: int,
hook_manager: HookManager or None = None):
super(AlphaZeroNet, self).__init__()
self.in_channels = in_channels
self.num_channels = num_channels
self.dropout_p = dropout
self.action_size = action_size
self.linear_input_size = linear_input_size
self.hook_manager = hook_manager if hook_manager is not None else HookManager()
# Convolutional layers
self.conv1 = nn.Conv2d(in_channels, num_channels, 3, padding=1)
self.bn1 = nn.BatchNorm2d(num_channels)
self.conv2 = nn.Conv2d(num_channels, num_channels, 3, padding=1)
self.bn2 = nn.BatchNorm2d(num_channels)
self.conv3 = nn.Conv2d(num_channels, num_channels, 3, padding=1)
self.bn3 = nn.BatchNorm2d(num_channels)
self.conv4 = nn.Conv2d(num_channels, num_channels, 3)
self.bn4 = nn.BatchNorm2d(num_channels)
# Fully connected layers
# 4608 (5x5) or 512 (3x3) or 32768 (10x10) or 18432 (8x8) # or 8192 for atari (6x6)
self.fc1 = nn.Linear(self.linear_input_size, 1024)
self.fc1_bn = nn.BatchNorm1d(1024)
self.fc2 = nn.Linear(1024, 512)
self.fc2_bn = nn.BatchNorm1d(512)
self.dropout = nn.Dropout(dropout)
# Output layers
self.pi = nn.Linear(512, action_size) # probability head
self.v = nn.Linear(512, 1) # value head
atexit.register(self.clear_traces)
def forward(self, x, muzero: bool = True):
if not muzero:
x = x.unsqueeze(1)
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = F.relu(self.bn3(self.conv3(x)))
x = F.relu(self.bn4(self.conv4(x)))
x = x.reshape(x.size(0), -1) # Flatten
x = F.relu(self.fc1_bn(self.fc1(x)))
x = self.dropout(x)
x = F.relu(self.fc2_bn(self.fc2(x)))
x = self.dropout(x)
pi = F.log_softmax(self.pi(x), dim=1)
v = F.tanh(self.v(x))
return pi, v
@th.no_grad()
def predict(self, x, muzero=True):
pi, v = self.forward(x, muzero=muzero)
pi = th.exp(pi)
return pi.detach().cpu().numpy(), v.detach().cpu().numpy()
def to_traced_script(self, board_size: int = 10):
return th.jit.trace(self, th.rand(1, 256, board_size, board_size).cuda())
def trace(self, board_size: int) -> str:
traced = self.to_traced_script(board_size=board_size)
path = "traced.pt"
traced.save(path)
return path
def clear_traces(self) -> None:
from mu_alpha_zero.General.utils import find_project_root
for trace_file in glob.glob(f"{find_project_root()}/Checkpoints/Traces/*.pt"):
os.remove(trace_file)
def make_fresh_instance(self):
return AlphaZeroNet(self.in_channels, self.num_channels, self.dropout_p, self.action_size,
self.linear_input_size)
@staticmethod
def make_from_config(config: AlphaZeroConfig, hook_manager: HookManager or None = None):
return AlphaZeroNet(config.num_net_in_channels, config.num_net_channels, config.net_dropout,
config.net_action_size, config.az_net_linear_input_size, hook_manager=hook_manager)
def train_net(self, memory_buffer: MemBuffer, alpha_zero_config: AlphaZeroConfig):
from mu_alpha_zero.AlphaZero.utils import mask_invalid_actions_batch
device = th.device("cuda" if th.cuda.is_available() else "cpu")
losses = []
optimizer = th.optim.Adam(self.parameters(), lr=alpha_zero_config.lr)
memory_buffer.shuffle()
for epoch in range(alpha_zero_config.epochs):
for experience_batch in memory_buffer(alpha_zero_config.batch_size):
if len(experience_batch) <= 1:
continue
states, pi, v = zip(*experience_batch)
states = th.tensor(np.array(states), dtype=th.float32, device=device)
pi = th.tensor(np.array(pi), dtype=th.float32, device=device)
v = th.tensor(v, dtype=th.float32, device=device).unsqueeze(1)
pi_pred, v_pred = self.forward(states, muzero=False)
masks = mask_invalid_actions_batch(states)
loss = mse_loss(v_pred, v) + self.pi_loss(pi_pred, pi, masks, device)
losses.append(loss.item())
# self.summary_writer.add_scalar("Loss", loss.item(), i * epochs + epoch)
optimizer.zero_grad()
loss.backward()
optimizer.step()
self.hook_manager.process_hook_executes(self, self.train_net.__name__, __file__, HookAt.MIDDLE,
args=(experience_batch, loss.item(), epoch))
self.hook_manager.process_hook_executes(self, self.train_net.__name__, __file__, HookAt.TAIL, args=(losses,))
return sum(losses) / len(losses), losses
def pi_loss(self, y_hat, y, masks, device: th.device):
masks = masks.reshape(y_hat.shape).to(device)
masked_y_hat = masks * y_hat
return -th.sum(y * masked_y_hat) / y.size()[0]
def to_shared_memory(self):
for param in self.parameters():
param.share_memory_()
def run_at_training_end(self):
self.hook_manager.process_hook_executes(self, self.run_at_training_end.__name__, __file__, HookAt.ALL)
class OriginalAlphaZeroNetwork(nn.Module, GeneralAlphZeroNetwork):
def __init__(self, in_channels: int, num_channels: int, dropout: float, action_size: int,
linear_input_size: list[int], support_size: int,
state_linear_layers: int, pi_linear_layers: int, v_linear_layers: int, linear_head_hidden_size: int,
is_atari: bool,
latent_size: list[int] = [6, 6],
hook_manager: HookManager or None = None, num_blocks: int = 8, muzero: bool = False,
is_dynamics: bool = False, is_representation: bool = False):
super(OriginalAlphaZeroNetwork, self).__init__()
self.in_channels = in_channels
self.num_channels = num_channels
self.dropout_p = dropout
self.action_size = action_size
self.linear_input_size = linear_input_size
self.support_size = support_size
self.num_blocks = num_blocks
self.muzero = muzero
self.latent_size = latent_size
self.is_dynamics = is_dynamics
self.is_representation = is_representation
self.state_linear_layers = state_linear_layers
self.pi_linear_layers = pi_linear_layers
self.v_linear_layers = v_linear_layers
self.linear_head_hidden_size = linear_head_hidden_size
self.is_atari = is_atari
self.optimizer = None
self.scheduler = None
self.hook_manager = hook_manager if hook_manager is not None else HookManager()
self.conv1 = nn.Conv2d(in_channels, num_channels, 3, padding=1)
self.bn1 = nn.BatchNorm2d(num_channels)
self.dropout = nn.Dropout(dropout)
self.blocks = nn.ModuleList([OriginalAlphaZeroBlock(num_channels, num_channels) for _ in range(num_blocks)])
if not is_representation:
self.value_head = ValueHead(muzero, linear_input_size[0], support_size, num_channels, v_linear_layers,
linear_head_hidden_size)
else:
self.value_head = th.nn.Identity()
if is_dynamics or is_representation:
self.policy_state_head = StateHead(linear_input_size[2], num_channels, latent_size, state_linear_layers,
linear_head_hidden_size)
else:
self.policy_state_head = PolicyHead(action_size, linear_input_size[1], num_channels, pi_linear_layers,
linear_head_hidden_size)
def forward(self, x, muzero: bool = False, return_support: bool = False):
if not muzero:
x = x.unsqueeze(1)
x = F.relu(self.bn1(self.conv1(x)))
for block in self.blocks:
x = block(x)
# x = self.dropout(x)
val_h_output = self.value_head(x)
pol_h_output = self.policy_state_head(x)
if self.is_representation:
return pol_h_output
if not return_support and self.muzero:
from mu_alpha_zero.MuZero.utils import invert_scale_reward_value
# multiply arange by softmax probabilities
val_h_output = th.exp(val_h_output)
support_range = th.arange(-self.support_size, self.support_size + 1, 1, dtype=th.float32,
device=x.device).unsqueeze(0)
output = th.sum(val_h_output * support_range, dim=1)
if self.is_atari:
output = invert_scale_reward_value(output)
return pol_h_output, output.unsqueeze(1)
return pol_h_output, val_h_output
@th.no_grad()
def predict(self, x, muzero=True):
pi, v = self.forward(x, muzero=muzero)
pi = th.exp(pi)
return pi.detach().cpu().numpy(), v.detach().cpu().numpy()
def pi_loss(self, y_hat, y, masks, device: th.device):
masks = masks.reshape(y_hat.shape).to(device)
masked_y_hat = masks * y_hat
return -th.sum(y * masked_y_hat) / y.size()[0]
def make_fresh_instance(self):
return OriginalAlphaZeroNetwork(self.in_channels, self.num_channels, self.dropout_p, self.action_size,
self.linear_input_size, self.support_size,
self.state_linear_layers, self.pi_linear_layers, self.v_linear_layers,
self.linear_head_hidden_size,
self.is_atari,
self.latent_size,
hook_manager=self.hook_manager,
num_blocks=self.num_blocks, muzero=self.muzero, is_dynamics=self.is_dynamics)
@classmethod
def make_from_config(cls, config: Config, hook_manager: HookManager or None = None):
return OriginalAlphaZeroNetwork(config.num_net_in_channels, config.num_net_channels, config.net_dropout,
config.net_action_size, config.az_net_linear_input_size,
hook_manager=hook_manager,
state_linear_layers=config.state_linear_layers,
pi_linear_layers=config.pi_linear_layers,
v_linear_layers=config.v_linear_layers,
linear_head_hidden_size=config.linear_head_hidden_size,
num_blocks=config.num_blocks, muzero=config.muzero,
is_atari=config.is_atari,
support_size=config.support_size, latent_size=config.net_latent_size)
def train_net(self, memory_buffer, muzero_alphazero_config: Config) -> tuple[float, list[float]]:
if memory_buffer.train_length() <= 1:
return 0, []
losses = []
if self.optimizer is None:
self.optimizer = th.optim.Adam(self.parameters(), lr=muzero_alphazero_config.lr,
weight_decay=muzero_alphazero_config.l2)
if muzero_alphazero_config.lr_scheduler is not None and self.scheduler is None:
self.scheduler = muzero_alphazero_config.lr_scheduler(self.optimizer,
**muzero_alphazero_config.lr_scheduler_kwargs)
# memory_buffer.shuffle()
for epoch in range(muzero_alphazero_config.epochs):
for experience_batch in memory_buffer.batch(muzero_alphazero_config.batch_size):
loss, v_loss, pi_loss = self.calculate_loss(experience_batch, muzero_alphazero_config)
losses.append(loss.item())
wandb.log({"combined_loss": loss.item(), "loss_v": v_loss.item(), "loss_pi": pi_loss.item()})
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
if self.scheduler is not None:
self.scheduler.step()
wandb.log({"lr": self.scheduler.get_last_lr()[-1]})
self.hook_manager.process_hook_executes(self, self.train_net.__name__, __file__, HookAt.MIDDLE,
args=(experience_batch, loss.item(), epoch))
self.hook_manager.process_hook_executes(self, self.train_net.__name__, __file__, HookAt.TAIL, args=(losses,))
return sum(losses) / len(losses), losses
def eval_net(self, memory_buffer, muzero_alphazero_config: Config) -> None:
if memory_buffer.eval_length() <= 1:
return
if self.optimizer is None:
self.optimizer = th.optim.Adam(self.parameters(), lr=muzero_alphazero_config.lr,
weight_decay=muzero_alphazero_config.l2)
# memory_buffer.shuffle(is_eval=True)
for epoch in range(muzero_alphazero_config.eval_epochs):
for experience_batch in memory_buffer.batch(muzero_alphazero_config.batch_size, is_eval=True):
loss, v_loss, pi_loss = self.calculate_loss(experience_batch, muzero_alphazero_config)
wandb.log(
{"eval_combined_loss": loss.item(), "eval_loss_v": v_loss.item(), "eval_loss_pi": pi_loss.item()})
def calculate_loss(self, experience_batch, muzero_alphazero_config):
from mu_alpha_zero.AlphaZero.utils import mask_invalid_actions_batch
device = th.device("cuda" if th.cuda.is_available() else "cpu")
states, pi, v, _, masks = experience_batch[0], experience_batch[1], experience_batch[2], experience_batch[3], \
experience_batch[4]
pi = [[y for y in x.values()] if isinstance(x, dict) else x for x in pi]
# game = [[y.frame,y.pi,y.v,y.action_mask] for y in experience_batch.datapoints]
# states, pi, v, masks = zip(*game)
states = th.tensor(np.array(states), dtype=th.float32, device=device)
pi = th.tensor(np.array(pi), dtype=th.float32, device=device)
v = th.tensor(v, dtype=th.float32, device=device).unsqueeze(1)
masks = th.tensor(np.array(masks), dtype=th.float32, device=device)
pi_pred, v_pred = self.forward(states, muzero=muzero_alphazero_config.muzero)
v_loss = mse_loss(v_pred, v)
pi_loss = self.pi_loss(pi_pred, pi, masks, device)
loss = v_loss + pi_loss
return loss, v_loss, pi_loss
def continuous_weight_update(self, shared_storage: SharedStorage, alpha_zero_config: AlphaZeroConfig,
checkpointer: CheckPointer or None,
logger: Logger or None):
wandb.init(project=alpha_zero_config.wandbd_project_name, name="Continuous Weight Update")
alpha_zero_config.epochs = 1
alpha_zero_config.eval_epochs = 50
self.train()
while shared_storage.train_length() < 200:
time.sleep(5)
params = shared_storage.get_stable_network_params()
self.load_state_dict(params)
for iter_ in range(alpha_zero_config.num_worker_iters):
# print(iter_)
# if not shared_storage.get_was_pitted():
# print("Waiting for pitting to finish")
# time.sleep(5)
# continue
avg_iter_losses = self.train_net(shared_storage, alpha_zero_config)
shared_storage.set_experimental_network_params(self.state_dict())
shared_storage.set_training_iter(iter_)
shared_storage.add_combined_loss(avg_iter_losses[0])
# shared_storage.set_was_pitted(False)
if iter_ % alpha_zero_config.eval_interval == 0 and iter_ != 0:
self.eval_net(shared_storage, alpha_zero_config)
checkpointer.save_checkpoint(self, self, self.optimizer, alpha_zero_config.lr, iter_, alpha_zero_config)
print(f"Saved checkpoint at iteration {iter_}")
class OriginalAlphaZeroBlock(th.nn.Module):
def __init__(self, in_channels: int, num_channels: int):
super(OriginalAlphaZeroBlock, self).__init__()
self.conv1 = nn.Conv2d(in_channels, num_channels, 3, padding=1)
self.bn1 = nn.BatchNorm2d(num_channels)
self.conv2 = nn.Conv2d(num_channels, num_channels, 3, padding=1)
self.bn2 = nn.BatchNorm2d(num_channels)
def forward(self, x):
x_skip = x
x = F.relu(self.bn1(self.conv1(x)))
x = self.bn2(self.conv2(x))
x += x_skip
return F.relu(x)
class ValueHead(th.nn.Module):
def __init__(self, muzero: bool, linear_input_size: int, support_size: int, num_channels: int, num_layers: int,
linear_hidden_size: int):
super(ValueHead, self).__init__()
self.conv = nn.Conv2d(num_channels, linear_hidden_size, 1)
self.bn = nn.BatchNorm2d(linear_hidden_size)
self.fc1 = nn.Linear(linear_input_size, 256)
# self.fc = HeadLinear(256, 256, num_layers, linear_hidden_size)
if muzero:
self.fc2 = nn.Linear(256, 2 * support_size + 1)
self.act = nn.LogSoftmax(dim=1)
else:
self.fc2 = nn.Linear(256, 1)
self.act = nn.Tanh()
def forward(self, x):
x = F.relu(self.bn(self.conv(x)))
x = x.reshape(x.size(0), -1)
x = F.relu(self.fc1(x))
# x = F.relu(self.fc(x))
x = self.fc2(x)
return self.act(x)
class PolicyHead(th.nn.Module):
def __init__(self, action_size: int, linear_input_size_policy: int, num_channels: int, num_layers: int,
linear_hidden_size: int):
super(PolicyHead, self).__init__()
self.conv = nn.Conv2d(num_channels, linear_hidden_size, 1)
self.bn = nn.BatchNorm2d(linear_hidden_size)
self.fc1 = nn.Linear(linear_input_size_policy, 256)
# self.fc = HeadLinear(256, 256, num_layers, linear_hidden_size)
self.fc2 = nn.Linear(256, action_size)
def forward(self, x):
x = F.relu(self.bn(self.conv(x)))
x = x.reshape(x.size(0), -1)
x = F.relu(self.fc1(x))
# x = F.relu(self.fc(x))
x = self.fc2(x)
return F.log_softmax(x, dim=1)
class StateHead(th.nn.Module):
def __init__(self, linear_input_size: int, out_channels: int, latent_size: list[int], num_layers: int,
linear_hidden_size: int):
super(StateHead, self).__init__()
self.conv = nn.Conv2d(out_channels, out_channels, 1)
self.bn = nn.BatchNorm2d(out_channels)
# self.fc = HeadLinear(linear_input_size, out_channels, num_layers, linear_hidden_size)
self.fc3 = nn.Linear(linear_input_size, latent_size[0] * latent_size[1] * out_channels)
def forward(self, x):
x = F.relu(self.bn(self.conv(x)))
x = x.reshape(x.size(0), -1)
# x = F.relu(self.fc(x))
x = F.relu(self.fc3(x))
return x
class HeadLinear(nn.Module):
def __init__(self, in_channels: int, out_channels: int, num_layers: int, hidden_size: int):
super(HeadLinear, self).__init__()
self.fc1 = nn.Linear(in_channels, hidden_size)
self.fc = nn.Sequential(*[nn.Linear(hidden_size, hidden_size) for _ in range(num_layers)])
self.fc2 = nn.Linear(hidden_size, out_channels)
def forward(self, x):
x = F.relu(self.fc1(x))
x = F.relu(self.fc(x))
return self.fc2(x)
| 20,720 | Python | .py | 373 | 43.356568 | 120 | 0.604543 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,520 | nnet.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Network/__pycache__/nnet.cpython-311.pyc | §
*¤fZD ã óî — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dl mc m
Z d dl Z d dl m
Z
d dlmZ d dlmZ d dlmZ d dlmZ d dlmZmZ d dlmZ d d lm Z G d
„ d ej e¦ « Z G d „ d
ej e¦ « Z G d„ dej j ¦ « Z G d„ dej j ¦ « Z! G d„ dej j ¦ « Z" G d„ dej j ¦ « Z# G d„ dej ¦ « Z$dS )é N)Úmse_loss)ÚT)ÚGeneralAlphZeroNetwork)Ú HookManager)ÚHookAt)ÚAlphaZeroConfigÚConfig)Ú MemBuffer)Úinvert_scale_reward_valuec
ó ‡ — e Zd Z d dedededededepdf ˆ fd„
Zd d
efd „Z e j
¦ « d d „¦ « Z d defd„Z dede
fd„Zd d„Zd„ Zed dedepdfd„¦ « Zdedefd„Zde j fd„Zd„ Zd „ Zˆ xZS )!Ú AlphaZeroNetNÚ in_channelsÚ num_channelsÚdropoutÚ action_sizeÚlinear_input_sizeÚ hook_managerc ó •— t t | ¦ « ¦ « || _ || _ || _ || _ || _ |�|n
t ¦ « | _ t j ||dd¬¦ « | _ t j
|¦ « | _ t j ||dd¬¦ « | _ t j
|¦ « | _ t j ||dd¬¦ « | _ t j
|¦ « | _ t j ||d¦ « | _ t j
|¦ « | _ t j | j d¦ « | _ t j d¦ « | _ t j dd¦ « | _ t j d¦ « | _ t j |¦ « | _ t j d|¦ « | _ t j dd¦ « | _ t? j | j! ¦ « d S )Né é ©Úpaddingi i )"Úsuperr
Ú__init__r r Ú dropout_pr r r r ÚnnÚConv2dÚconv1Ú BatchNorm2dÚbn1Úconv2Úbn2Úconv3Úbn3Úconv4Úbn4ÚLinearÚfc1Ú BatchNorm1dÚfc1_bnÚfc2Úfc2_bnÚDropoutr ÚpiÚvÚatexitÚregisterÚ clear_traces)Úselfr r r r r r Ú __class__s €úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Network/nnet.pyr zAlphaZeroNet.__init__ s¡ ø€ å
�l˜DÑ!Ô!×*Ò*Ñ,Ô,Ğ,Ø&ˆÔØ (ˆÔØ ˆŒØ&ˆÔØ!2ˆÔ Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔå”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1Ñ=Ô=ˆŒ
İ”> ,Ñ/Ô/ˆŒõ ”9˜TÔ 3°TÑ:Ô:ˆŒİ”n TÑ*Ô*ˆŒ İ”9˜T 3Ñ'Ô'ˆŒİ”n SÑ)Ô)ˆŒ å”z 'Ñ*Ô*ˆŒ õ ”)˜C Ñ-Ô-ˆŒİ”˜3 Ñ"Ô"ˆŒİŒ˜Ô)Ñ*Ô*Ğ*Ğ*Ğ*ó TÚmuzeroc ó6 — |s| d¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | |
|¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j |
| |¦ « ¦ « ¦ « }| |¦ « }t j | | |¦ « ¦ « ¦ « }| |¦ « }t j | |¦ « d¬¦ « }t j | |¦ « ¦ « }||fS )Nr r éÿÿÿÿ©Údim)Ú unsqueezeÚFÚrelur r r" r! r$ r# r& r% ÚreshapeÚsizer* r( r r, r+ Ú log_softmaxr. Útanhr/ ©r3 Úxr7 r. r/ s r5 ÚforwardzAlphaZeroNet.forward7 sr € Øğ Ø— ’ ˜A‘”ˆAİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆà
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆå
Œ]˜4Ÿ7š7 1™:œ:¨1Ğ
-Ñ
-Ô
-ˆİ
ŒF�4—6’6˜!‘9”9Ñ Ô ˆà�1ˆuˆ r6 c ó> — | ||¬¦ « \ }}t j |¦ « }| ¦ « ¦ « ¦ « | ¦ « ¦ « ¦ « fS ©N©r7 ©rE ÚthÚexpÚdetachÚcpuÚnumpyrC s r5 ÚpredictzAlphaZeroNet.predictL óo € à— ’ ˜Q v� Ñ.Ô.‰ˆˆAİ
ŒV�B‰ZŒZˆØ�yŠy‰{Œ{�ŠÑ Ô ×&Ò&Ñ(Ô(¨!¯(ª(©*¬*¯.ª.Ñ*:Ô*:×*@Ò*@Ñ*BÔ*BĞBĞBr6 é
Ú
board_sizec ó� — t j | t j dd||¦ « ¦ « ¦ « S )Nr é )rJ ÚjitÚtraceÚrandÚcuda)r3 rR s r5 Úto_traced_scriptz AlphaZeroNet.to_traced_scriptR s4 € İŒv�|Š|˜D¥"¤'¨!¨S°*¸jÑ"IÔ"I×"NÒ"NÑ"PÔ"PÑQÔQĞQr6 Úreturnc ó` — | |¬¦ « }d}| |¦ « |S )N)rR z traced.pt)rY Úsave)r3 rR ÚtracedÚpaths r5 rV zAlphaZeroNet.traceU s4 € Ø×&Ò&°*Ğ&Ñ=Ô=ˆØˆØ� Š �DÑÔĞ؈ r6 c ó~ — ddl m} t j |¦ « › d�¦ « D ]}t j |¦ « Œd S )Nr )Úfind_project_rootz/Checkpoints/Traces/*.pt)Úmu_alpha_zero.General.utilsr` ÚglobÚosÚremove)r3 r` Ú
trace_files r5 r2 zAlphaZeroNet.clear_traces[ s] € ØAĞAĞAĞAĞAĞAİ œ)Ğ'8Ğ'8Ñ':Ô':Ğ$TĞ$TĞ$TÑUÔUğ "ğ "ˆJİ ŒI�jÑ !Ô !Ğ !Ğ !ğ "ğ "r6 c óZ — t | j | j | j | j | j ¦ « S ©N)r
r r r r r ©r3 s r5 Úmake_fresh_instancez AlphaZeroNet.make_fresh_instance` s. € ݘDÔ ,¨dÔ.?ÀÄ ĞQUÔQaØ Ô 2ñ4ô 4ğ 4r6 Úconfigc ó^ — t | j | j | j | j | j |¬¦ « S )N)r )r
Únum_net_in_channelsÚnum_net_channelsÚ net_dropoutÚnet_action_sizeÚaz_net_linear_input_size)rj r s r5 Úmake_from_configz AlphaZeroNet.make_from_configd s; € å˜FÔ 6¸Ô8OĞQWÔQcØ "Ô 2°FÔ4SĞbnğpñ pô pğ pr6 Ú
memory_bufferÚalpha_zero_configc ó” — ddl m} t j t j ¦ « rdnd¦ « }g }t j | ¦ « |j ¬¦ « }|
¦ « t |j ¦ « D �]Ñ} ||j
¦ « D �]¼}t |¦ « dk rŒt |� \ } }
} t j t# j | ¦ « t j |¬¦ « } t j t# j |
¦ « t j |¬¦ « }
t j | t j |¬¦ « d¦ « } | | d¬ ¦ « \ } }
|| ¦ « }t- |
| ¦ « | | |
||¦ « z }| | ¦ « ¦ « | ¦ « | ¦ « | ¦ « | j | | j j tB tD j# || ¦ « |f¬
¦ « �Œ¾�ŒÓ| j | | j j tB tD j$ |f¬
¦ « tK |¦ « t |¦ « z |fS ) Nr ©Úmask_invalid_actions_batchrX rM )Úlrr ©ÚdtypeÚdeviceFrH ©Úargs)&Ú mu_alpha_zero.AlphaZero.utilsrv rJ rz rX Ú is_availableÚoptimÚAdamÚ
parametersrw ÚshuffleÚrangeÚepochsÚ
batch_sizeÚlenÚzipÚtensorÚnpÚarrayÚfloat32r< rE r Úpi_lossÚappendÚitemÚ zero_gradÚbackwardÚstepr Úprocess_hook_executesÚ train_netÚ__name__Ú__file__r ÚMIDDLEÚTAILÚsum)r3 rr rs rv rz ÚlossesÚ optimizerÚepochÚexperience_batchÚstatesr. r/ Úpi_predÚv_predÚmasksÚlosss r5 r“ zAlphaZeroNet.train_neti sƒ € ØLĞLĞLĞLĞLĞLİ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆØˆİ”H—M’M $§/¢/Ñ"3Ô"3Ğ8IÔ8L�MÑMÔMˆ Ø× Ò Ñ Ô Ğ İĞ,Ô3Ñ4Ô4ğ eñ eˆEØ$1 MĞ2CÔ2NÑ$OÔ$Oğ
eñ
eĞ İĞ'Ñ(Ô(¨AÒ-Ğ-Ø İ #Ğ%5Ğ 6‘
�˜˜Aİœ¥2¤8¨FÑ#3Ô#3½2¼:ÈfĞUÑUÔU�İ”Y�rœx¨™|œ|µ2´:ÀfĞMÑMÔM�İ”I˜a¥r¤z¸&ĞAÑAÔA×KÒKÈAÑNÔN�Ø"&§,¢,¨v¸e ,Ñ"DÔ"D‘�˜Ø2Ğ2°6Ñ:Ô:�İ ¨Ñ*Ô*¨T¯\ª\¸'À2ÀuÈfÑ-UÔ-UÑU�Ø—
’
˜dŸiši™kœkÑ*Ô*Ğ*ğ ×#Ò#Ñ%Ô%Ğ%Ø—
’
‘”�Ø—’Ñ Ô Ğ ØÔ!×7Ò7¸¸d¼nÔ>UÕW_ÕagÔanØ>NĞPT×PYÒPYÑP[ÔP[Ğ]bĞ=cğ 8ñ eô eğ eñ eñ!
eğ&
Ô×/Ò/°°d´nÔ6MÍxÕY_ÔYdĞlrĞktĞ/ÑuÔuĞuİ�6‰{Œ{�S ™[œ[Ñ(¨&Ğ0Ğ0r6 rz c óÌ — | |j ¦ « |¦ « }||z }t j ||z ¦ « | ¦ « d z S ©Nr ©r? ÚshapeÚtorJ r˜ r@ ©r3 Úy_hatÚyr rz Ú masked_y_hats r5 rŒ zAlphaZeroNet.pi_loss† óU € Ø—
’
˜eœkÑ*Ô*×-Ò-¨fÑ5Ô5ˆØ ˜u‘}ˆ İ”�q˜<Ñ'Ñ(Ô(Ğ(¨1¯6ª6©8¬8°A¬;Ñ6Ğ6r6 c ó\ — | ¦ « D ]}| ¦ « Œd S rg )r� Ú
share_memory_)r3 Úparams r5 Úto_shared_memoryz AlphaZeroNet.to_shared_memory‹ s: € Ø—_’_Ñ&Ô&ğ "ğ "ˆEØ × Ò Ñ !Ô !Ğ !Ğ !ğ "ğ "r6 c ór — | j | | j j t t
j ¦ « d S rg )r r’ Úrun_at_training_endr” r• r ÚALLrh s r5 r± z AlphaZeroNet.run_at_training_end� s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr6 rg ©T)rQ )rZ N)r” Ú
__module__Ú __qualname__ÚintÚfloatr r ÚboolrE rJ Úno_gradrO rY ÚstrrV r2 ri Ú staticmethodr rq r
r“ rz rŒ r¯ r± Ú
__classcell__©r4 s @r5 r
r
sÒ ø€ € € € € à59ğ +ğ + Cğ +°sğ +ÀUğ +ĞY\ğ +Ğqtğ +Ø *Ğ 2¨dğ +ğ +ğ +ğ +ğ +ğ +ğBğ ğ ğ ğ ğ ğ* €R„Z�\„\ğCğ Cğ Cñ „\ğCğ
Rğ R¨3ğ Rğ Rğ Rğ Rğ ğ ¨ğ ğ ğ ğ ğ "ğ "ğ "ğ "ğ
4ğ 4ğ 4ğ ğpğ p ğ pÀ Ğ@SÈtğ pğ pğ pñ „\ğpğ1 yğ 1À_ğ 1ğ 1ğ 1ğ 1ğ:7¨r¬yğ 7ğ 7ğ 7ğ 7ğ
"ğ "ğ "ğoğ oğ oğ oğ oğ oğ or6 r
c óT ‡ — e Zd Zddgddddfdedededed ee d
ed ed ed
ededee depddededef ˆ fd„
Zd$dedefd„Z e
j ¦ « d%d„¦ « Z de
j
fd„Zd„ Zed&d ed ddepdfd „¦ « Zd ed eeee f fd!„Zd ed dfd"„Zd#„ Zˆ xZS )'ÚOriginalAlphaZerNetworké Né Fr r r r r Ú support_sizeÚstate_linear_layersÚpi_linear_layersÚv_linear_layersÚlinear_head_hidden_sizeÚ latent_sizer Ú
num_blocksr7 Ú is_dynamicsc ó •‡— t t | ¦ « ¦ « || _ ‰| _ || _ || _ || _ || _ |
| _ || _
| | _ || _ || _
|| _ | | _ |
| _ d | _ | �| n
t% ¦ « | _ t) j |‰dd¬¦ « | _ t) j ‰¦ « | _ t) j |¦ « | _ t) j ˆfd„t9 |
¦ « D ¦ « ¦ « | _ t= ||d |‰| |
¦ « | _ |r tA |d ‰| ||
¦ « | _! d S tE ||d ‰||
¦ « | _! d S )Nr r r c ó0 •— g | ]}t ‰‰¦ « ‘ŒS © )ÚOriginalAlphaZeroBlock)Ú.0Ú_r s €r5 ú
<listcomp>z4OriginalAlphaZerNetwork.__init__.<locals>.<listcomp>° s% ø€ Ğ$sĞ$sĞ$sĞ\]Õ%;¸LÈ,Ñ%WÔ%WĞ$sĞ$sĞ$sr6 r é )#r r¿ r r r r r r r rÈ r7 rÇ rÉ rà rÄ rÅ rÆ rš r r r r r r r r- r Ú
ModuleListrƒ ÚblocksÚ ValueHeadÚ
value_headÚ StateHeadÚ policy_headÚ
PolicyHead)r3 r r r r r r rà rÄ rÅ rÆ rÇ r rÈ r7 rÉ r4 s ` €r5 r z OriginalAlphaZerNetwork.__init__• s¡ øø€ õ Õ% tÑ,Ô,×5Ò5Ñ7Ô7Ğ7Ø&ˆÔØ (ˆÔØ ˆŒØ&ˆÔØ!2ˆÔ Ø (ˆÔØ$ˆŒØ ˆŒ Ø&ˆÔØ&ˆÔØ#6ˆÔ Ø 0ˆÔ Ø .ˆÔ Ø'>ˆÔ$Ø ˆŒØ,8Ğ,D˜L˜LÍ+É-Ì-ˆÔå”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”z 'Ñ*Ô*ˆŒ İ”mĞ$sĞ$sĞ$sĞ$sÕafĞgqÑarÔarĞ$sÑ$sÔ$sÑtÔtˆŒ İ# FĞ,=¸aÔ,@À,ĞP\Ğ^mØ$;ñ=ô =ˆŒà ğ Cİ (Ğ):¸1Ô)=¸|È[ĞZmØ)@ñ Bô BˆDÔ Ğ Ğ õ *¨+Ğ7HÈÔ7KÈ\Ğ[kØ*Añ Cô CˆDÔ Ğ Ğ r6 Úreturn_supportc ó� — |s| d¦ « }t j | | |¦ « ¦ « ¦ « }| j D ]
} ||¦ « }Œ| |¦ « }| |¦ « }| |¦ « }|sœt j
|¦ « }t j | j | j dz dt j
|j ¬¦ « d¦ « }t j ||z d¬¦ « }t! |¦ « }|| d¦ « fS ||fS )Nr rx r r: )r< r= r> r r rÓ r rÕ r× rJ rK Úaranger r‹ rz r˜ r ) r3 rD r7 rÙ ÚblockÚ val_h_outputÚ pol_h_outputÚ
support_rangeÚoutputs r5 rE z OriginalAlphaZerNetwork.forwardº s6 € Øğ Ø— ’ ˜A‘”ˆAİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆØ”[ğ ğ ˆEØ��a‘”ˆAˆAØ �LŠL˜‰OŒOˆØ—’ qÑ)Ô)ˆ Ø×'Ò'¨Ñ*Ô*ˆ Ø ğ 5å œ6 ,Ñ/Ô/ˆLİ œI tÔ'8Ğ&8¸$Ô:KÈaÑ:OĞQRÕZ\ÔZdØ-.¬Xğ 7ñ 7ô 7ß7@²yÀ±|´|ğ
å”V˜L¨=Ñ 8¸aĞ@Ñ@Ô@ˆFİ.¨vÑ6Ô6ˆFØ ×!1Ò!1°!Ñ!4Ô!4Ğ4Ğ 4ؘ\Ğ)Ğ)r6 Tc ó> — | ||¬¦ « \ }}t j |¦ « }| ¦ « ¦ « ¦ « | ¦ « ¦ « ¦ « fS rG rI rC s r5 rO z OriginalAlphaZerNetwork.predictÍ rP r6 rz c óÌ — | |j ¦ « |¦ « }||z }t j ||z ¦ « | ¦ « d z S r£ r¤ r§ s r5 rŒ z OriginalAlphaZerNetwork.pi_lossÓ r« r6 c óÔ — t | j | j | j | j | j | j | j | j | j | j
| j | j | j
| j | j ¬¦ « S )N)r rÈ r7 rÉ )r¿ r r r r r r rà rÄ rÅ rÆ rÇ r rÈ r7 rÉ rh s r5 ri z+OriginalAlphaZerNetwork.make_fresh_instanceØ sn € İ& tÔ'7¸Ô9JÈDÌNĞ\`Ô\lØ'+Ô'=¸tÔ?PØ'+Ô'?ÀÔAVĞX\ÔXlØ'+Ô'CØ'+Ô'7Ø48Ô4EØ26´/È$Ì+ĞcgÔcsğ
uñ uô uğ ur6 rj Ú game_managerc ó¾ — t |j |j |j |j |j ||j |j |j |j |j
|j |j |j
¬¦ « S )N) r rà rÄ rÅ rÆ rÈ r7 r rÇ )r¿ rl rm rn ro rp rà rÄ rÅ rÆ rÈ r7 r Únet_latent_size)Úclsrj rä r s r5 rq z(OriginalAlphaZerNetwork.make_from_configá si € å& vÔ'AÀ6ÔCZĞ\bÔ\nØ'-Ô'=¸vÔ?^Ø4@Ø;AÔ;UØ8>Ô8OØ7=Ô7MØ?EÔ?]Ø28Ô2CÈFÌMØ4:Ô4GĞU[ÔUkğmñ mô mğ mr6 Úmuzero_alphazero_configrZ c ó> — g }| j €Ct j | ¦ « |j |j ¬¦ « | _ | ¦ « t |j ¦ « D �]N} ||j
¦ « D �]9}t |¦ « dk rŒ| ||¦ « \ }}}|
| ¦ « ¦ « t j | ¦ « | ¦ « | ¦ « dœ¦ « | j ¦ « | ¦ « | j ¦ « | j | | j j t0 t2 j || ¦ « |f¬¦ « �Œ;�ŒP| j | | j j t0 t2 j |f¬¦ « t9 |¦ « t |¦ « z |fS )N©rw Ú weight_decayr )Ú
combined_lossÚloss_vÚloss_pir{ ) rš rJ r r€ r� rw Úl2r‚ rƒ r„ r… r† Úcalculate_lossr� r� ÚwandbÚlogr� r� r‘ r r’ r“ r” r• r r– r— r˜ ) r3 rr rè r™ r› rœ r¡ Úv_lossrŒ s r5 r“ z!OriginalAlphaZerNetwork.train_netí sõ € ØˆØ Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<ĞAXÔA[Ø8OÔ8Rğ +ñ Tô TˆDŒNà× Ò Ñ Ô Ğ İĞ2Ô9Ñ:Ô:ğ eñ eˆEØ$1 MĞ2IÔ2TÑ$UÔ$Uğ
eñ
eĞ İĞ'Ñ(Ô(¨AÒ-Ğ-Ø Ø(,×(;Ò(;Ğ<LĞNeÑ(fÔ(fÑ%��f˜gØ—
’
˜dŸiši™kœkÑ*Ô*Ğ*İ” ¨D¯IªI©K¬KÀ6Ç;Â;Á=Ä=Ğ]d×]iÒ]iÑ]kÔ]kĞlĞlÑmÔmĞmØ”×(Ò(Ñ*Ô*Ğ*Ø—
’
‘”�Ø”×#Ò#Ñ%Ô%Ğ%ØÔ!×7Ò7¸¸d¼nÔ>UÕW_ÕagÔanØ>NĞPT×PYÒPYÑP[ÔP[Ğ]bĞ=cğ 8ñ eô eğ eñ eñ
eğ
Ô×/Ò/°°d´nÔ6MÍxÕY_ÔYdĞlrĞktĞ/ÑuÔuĞuİ�6‰{Œ{�S ™[œ[Ñ(¨&Ğ0Ğ0r6 c ó — | j €Ct j | ¦ « |j |j ¬¦ « | _ | d¬¦ « t |j ¦ « D ]“} ||j
d¬¦ « D ]~}t |¦ « dk rŒ| ||¦ « \ }}}t j | ¦ « | ¦ « | ¦ « dœ¦ « ŒŒ”d S )Nrê T)Úis_evalr )Úeval_combined_lossÚ eval_loss_vÚ eval_loss_pi)rš rJ r r€ r� rw rï r‚ rƒ Ú eval_epochsr… r† rğ rñ rò r� )r3 rr rè r› rœ r¡ ró rŒ s r5 Úeval_netz OriginalAlphaZerNetwork.eval_net s( € Ø Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<ĞAXÔA[Ø8OÔ8Rğ +ñ Tô TˆDŒNà× Ò dĞ Ñ+Ô+Ğ+İĞ2Ô>Ñ?Ô?ğ wğ wˆEØ$1 MĞ2IÔ2TĞ^bĞ$cÑ$cÔ$cğ
wğ
wĞ İĞ'Ñ(Ô(¨AÒ-Ğ-Ø Ø(,×(;Ò(;Ğ<LĞNeÑ(fÔ(fÑ%��f˜gİ” Ø+/¯9ª9©;¬;ÀvÇ{Â{Á}Ä}Ğfm×frÒfrÑftÔftĞuĞuñwô wğ wğ wğ
wğ wğ wr6 c ó| — ddl m} t j t j ¦ « rdnd¦ « }t
|� \ }}}t j t j |¦ « t j
|¬¦ « }t j t j |¦ « t j
|¬¦ « }t j |t j
|¬¦ « d¦ « }| ||j
¬¦ « \ }} ||¦ « }
t | |¦ « } | |||
|¦ « } | | z }
|
| | fS )Nr ru rX rM rx r rH )r} rv rJ rz rX r~ r‡ rˆ r‰ rŠ r‹ r< rE r7 r rŒ )r3 rœ rè rv rz r� r. r/ r� rŸ r ró rŒ r¡ s r5 rğ z&OriginalAlphaZerNetwork.calculate_loss s € ØLĞLĞLĞLĞLĞLİ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİĞ -Ğ.‰
ˆ��Aİ”�2œ8 FÑ+Ô+µ2´:ÀfĞMÑMÔMˆİ
ŒY•r”x ‘|”|2¬:¸fĞ
EÑ
EÔ
Eˆİ ŒI�a�rœz°&Ğ 9Ñ 9Ô 9× CÒ CÀAÑ FÔ FˆØ Ÿ,š, vĞ6MÔ6T˜,ÑUÔU‰ˆ�Ø*Ğ*¨6Ñ2Ô2ˆİ˜& !Ñ$Ô$ˆØ—,’,˜w¨¨E°6Ñ:Ô:ˆØ˜Ñ ˆØ�V˜WĞ$Ğ$r6 )FFr³ rg )r” r´ rµ r¶ r· Úlistr r¸ r rE rJ r¹ rO rz rŒ ri Ú classmethodr rq Útupler“ rú rğ r¼ r½ s @r5 r¿ r¿ “ s3 ø€ € € € € ğ
,-¨a¨&Ø59ÈQĞ_dØ%*ğ #Cğ #C Cğ #C°sğ #CÀUğ #CĞY\ğ #CØ$(¨¤Iğ#CØ=@ğ#Cà&)ğ#Cà=@ğ#CàSVğ#Càqtğ#Cğ # 3œiğ#Cğ +Ğ 2¨dğ #Cğ HKğ #Cğ Y]ğ #Cğ
#ğ #Cğ #Cğ #Cğ #Cğ #Cğ #CğJ*ğ * ğ *¸tğ *ğ *ğ *ğ *ğ& €R„Z�\„\ğCğ Cğ Cñ „\ğCğ
7¨r¬yğ 7ğ 7ğ 7ğ 7ğ
uğ uğ uğ ğ mğ m fğ m¸Dğ mĞP[ĞPcĞ_cğ mğ mğ mñ „[ğ mğ1Àğ 1È5ĞQVĞX\Ğ]bÔXcĞQcÔKdğ 1ğ 1ğ 1ğ 1ğ, w¸vğ wÈ$ğ wğ wğ wğ wğ %ğ %ğ %ğ %ğ %ğ %ğ %r6 r¿ c ó. ‡ — e Zd Zdedefˆ fd„ Zd„ Zˆ xZS )rÍ r r c ó. •— t t | ¦ « ¦ « t j ||dd¬¦ « | _ t j |¦ « | _ t j ||dd¬¦ « | _ t j |¦ « | _ d S )Nr r r )
r rÍ r r r r r r r! r" )r3 r r r4 s €r5 r z OriginalAlphaZeroBlock.__init__ sz ø€ İ
Õ$ dÑ+Ô+×4Ò4Ñ6Ô6Ğ6İ”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒˆˆr6 c óü — |}t j | | |¦ « ¦ « ¦ « }| | |¦ « ¦ « }||z
}t j |¦ « S rg )r= r> r r r" r! )r3 rD Úx_skips r5 rE z OriginalAlphaZeroBlock.forward' s^ € ؈İ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆØ �HŠH�T—Z’Z ‘]”]Ñ #Ô #ˆØ ˆV‰ ˆİŒv�a‰yŒyĞr6 ©r” r´ rµ r¶ r rE r¼ r½ s @r5 rÍ rÍ sZ ø€ € € € € ğ0 Cğ 0°sğ 0ğ 0ğ 0ğ 0ğ 0ğ 0ğğ ğ ğ ğ ğ ğ r6 rÍ c ó> ‡ — e Zd Zdedededededef ˆ fd„ Zd„ Zˆ xZS ) rÔ r7 r r r Ú
num_layersÚlinear_hidden_sizec ó •— t t | ¦ « ¦ « t j |dd¦ « | _ t j d¦ « | _ t j |d¦ « | _ t dd||¦ « | _ |r<t j dd|z dz ¦ « | _ t j
d¬¦ « | _ d S t j dd¦ « | _ t j ¦ « | _ d S )Nr rT rÑ r: )r rÔ r r r Úconvr Úbnr' r( Ú
HeadLinearÚfcr+ Ú
LogSoftmaxÚactÚTanh)r3 r7 r r r r r r4 s €r5 r zValueHead.__init__0 sĞ ø€ å
�i˜Ñ Ô ×'Ò'Ñ)Ô)Ğ)İ”I˜l¨A¨qÑ1Ô1ˆŒ İ”. Ñ#Ô#ˆŒİ”9Ğ .°Ñ4Ô4ˆŒİ ˜S # zĞ3EÑFÔFˆŒØ ğ !İ”y a¨,Ñ&6¸Ñ&:Ñ;Ô;ˆDŒHİ”}¨Ğ+Ñ+Ô+ˆDŒHˆHˆHå”y aÑ(Ô(ˆDŒHİ”w‘y”yˆDŒHˆHˆHr6 c ó¸ — t j | | |¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j | |¦ « ¦ « }t j | |¦ « ¦ « }| |¦ « }| |¦ « S ©Nr r9 )
r= r> r r r? r@ r( r r+ r
©r3 rD s r5 rE zValueHead.forward> s• € İ
ŒF�4—7’7˜4Ÿ9š9 Q™<œ<Ñ(Ô(Ñ )Ô )ˆØ
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆİ
ŒF�4—8’8˜A‘;”;Ñ Ô ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ �HŠH�Q‰KŒKˆØ�xŠx˜‰{Œ{Ğr6 )r” r´ rµ r¸ r¶ r rE r¼ r½ s @r5 rÔ rÔ / sy ø€ € € € € ğ !˜tğ !¸ğ !È3ğ !Ğ^ağ !Ğorğ !Ø%(ğ !ğ !ğ !ğ !ğ !ğ !ğ ğ ğ ğ ğ ğ ğ r6 rÔ c
ó: ‡ — e Zd Zdededededef
ˆ fd„ Zd„ Zˆ xZS )rØ r Úlinear_input_size_policyr r r c óT •— t t | ¦ « ¦ « t j |dd¦ « | _ t j d¦ « | _ t j |d¦ « | _ t dd||¦ « | _ t j d|¦ « | _ d S )NrÑ r rT )
r rØ r r r r r r r' r( r
r r+ )r3 r r r r r r4 s €r5 r zPolicyHead.__init__H s† ø€ å
�j˜$Ñ Ô ×(Ò(Ñ*Ô*Ğ*İ”I˜l¨A¨qÑ1Ô1ˆŒ İ”. Ñ#Ô#ˆŒİ”9Ğ 5°sÑ;Ô;ˆŒİ ˜S # zĞ3EÑFÔFˆŒİ”9˜S +Ñ.Ô.ˆŒˆˆr6 c óº — t j | | |¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j | |¦ « ¦ « }t j | |¦ « ¦ « }| |¦ « }t j |d¬¦ « S )Nr r9 r r: )
r= r> r r r? r@ r( r r+ rA r s r5 rE zPolicyHead.forwardQ sš € İ
ŒF�4—7’7˜4Ÿ9š9 Q™<œ<Ñ(Ô(Ñ )Ô )ˆØ
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆİ
ŒF�4—8’8˜A‘;”;Ñ Ô ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ �HŠH�Q‰KŒKˆİŒ}˜Q AĞ&Ñ&Ô&Ğ&r6 r r½ s @r5 rØ rØ G sr ø€ € € € € ğ/ Cğ /À3ğ /ĞVYğ /Ğgjğ /Ø%(ğ/ğ /ğ /ğ /ğ /ğ /ğ'ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'r6 rØ c
óF ‡ — e Zd Zdededee dedef
ˆ fd„ Zd„ Zˆ xZS )rÖ r Ú out_channelsrÇ r r c óD •— t t | ¦ « ¦ « t j |dd¦ « | _ t j d¦ « | _ t ||||¦ « | _ t j
||d |d z |z ¦ « | _ d S )Né r r ) r rÖ r r r r r r r
r r' Úfc3)r3 r r rÇ r r r4 s €r5 r zStateHead.__init__[ s… ø€ å
�i˜Ñ Ô ×'Ò'Ñ)Ô)Ğ)İ”I˜l¨A¨qÑ1Ô1ˆŒ İ”. Ñ#Ô#ˆŒİ Ğ .° ¸jĞJ\Ñ]Ô]ˆŒİ”9˜\¨;°q¬>¸KȼNÑ+JÈ\Ñ+YÑZÔZˆŒˆˆr6 c óD — t j | | |¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j | |¦ « ¦ « }| |¦ « }|S r )r= r> r r r? r@ r r r s r5 rE zStateHead.forwardc sq € İ
ŒF�4—7’7˜4Ÿ9š9 Q™<œ<Ñ(Ô(Ñ )Ô )ˆØ
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ �HŠH�Q‰KŒKˆØˆr6 )r” r´ rµ r¶ rü r rE r¼ r½ s @r5 rÖ rÖ Z s� ø€ € € € € ğ[¨#ğ [¸Sğ [ÈtĞTWÌyğ [Ğfiğ [Ø%(ğ[ğ [ğ [ğ [ğ [ğ [ğğ ğ ğ ğ ğ ğ r6 rÖ c ó6 ‡ — e Zd Zdedededefˆ fd„ Zd„ Zˆ xZS )r
r r r Ú hidden_sizec ó •‡— t t | ¦ « ¦ « t j |‰¦ « | _ t j ˆfd„t |¦ « D ¦ « � | _ t j ‰|¦ « | _ d S )Nc ó: •— g | ]}t j ‰‰¦ « ‘ŒS rÌ )r r' )rÎ rÏ r s €r5 rĞ z'HeadLinear.__init__.<locals>.<listcomp>o s% ø€ Ğ!aĞ!aĞ!aÈ!¥"¤)¨K¸Ñ"EÔ"EĞ!aĞ!aĞ!ar6 )
r r
r r r' r( Ú
Sequentialrƒ r r+ )r3 r r r r r4 s `€r5 r zHeadLinear.__init__l sv øø€ İ
�j˜$Ñ Ô ×(Ò(Ñ*Ô*Ğ*İ”9˜[¨+Ñ6Ô6ˆŒİ”-Ğ!aĞ!aĞ!aĞ!aÍuĞU_ÑO`ÔO`Ğ!aÑ!aÔ!aĞbˆŒİ”9˜[¨,Ñ7Ô7ˆŒˆˆr6 c óÈ — t j | |¦ « ¦ « }t j | |¦ « ¦ « }| |¦ « S rg )r= r> r( r r+ r s r5 rE zHeadLinear.forwardr sD € İ
ŒF�4—8’8˜A‘;”;Ñ Ô ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ�xŠx˜‰{Œ{Ğr6 r r½ s @r5 r
r
k si ø€ € € € € ğ8 Cğ 8°sğ 8Èğ 8ĞZ]ğ 8ğ 8ğ 8ğ 8ğ 8ğ 8ğ ğ ğ ğ ğ ğ ğ r6 r
)%r0 rb rc rN r‰ ÚtorchrJ Útorch.nnr Útorch.nn.functionalÚ
functionalr= rñ r Útorch.nn.modules.moduler Ú mu_alpha_zero.General.networkr Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr Úmu_alpha_zero.configr r Úmu_alpha_zero.mem_bufferr
Úmu_alpha_zero.MuZero.utilsr ÚModuler
r¿ rÍ rÔ rØ rÖ r
rÌ r6 r5 ú<module>r. sj ğØ
€
€
€
Ø € € € Ø € € € à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ø € € € Ø (Ğ (Ğ (Ğ (Ğ (Ğ (Ø %Ğ %Ğ %Ğ %Ğ %Ğ %à @Ğ @Ğ @Ğ @Ğ @Ğ @Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø .Ğ .Ğ .Ğ .Ğ .Ğ .Ø @Ğ @Ğ @Ğ @Ğ @Ğ @ğ{oğ {oğ {oğ {oğ {o�2”9Ğ 4ñ {oô {oğ {oğ|I%ğ I%ğ I%ğ I%ğ I%˜bœiĞ)?ñ I%ô I%ğ I%ğX
ğ
ğ
ğ
ğ
˜RœUœ\ñ
ô
ğ
ğ ğ ğ ğ ğ �”” ñ ô ğ ğ0'ğ 'ğ 'ğ 'ğ '�”” ñ 'ô 'ğ 'ğ&ğ ğ ğ ğ �”” ñ ô ğ ğ"
ğ
ğ
ğ
ğ
�”ñ
ô
ğ
ğ
ğ
r6 | 31,519 | Python | .py | 95 | 330.715789 | 1,929 | 0.273636 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,521 | trainer.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Network/__pycache__/trainer.cpython-311.pyc | §
°#Ûeu> ã óJ — d dl mZ d dlmZ edd¬¦ « d dlmZ d dlmZ d dlZ d d l
m Z d d
l m Z d d l
mZ d d lmZmZmZ d d
lmZ d dlmZ d dlmZ d dlmZmZ d dl m Z d dl m Z d dl m!Z! d dl"m#Z# d dl$m%Z% d dl&m'Z' d dl(m)Z) d dl*m+Z+ d dl,m-Z- G d„ d¦ « Z.dS ) é )Úset_start_method)Únot_zeroÚspawnT)Úforce)Údeepcopy)ÚTypeN)Ú
SummaryWriter)Útqdm)ÚArena)Ú RandomPlayerÚPlayerÚ NetPlayer)Ú McSearchTree)Ú AlphaZeroNet)Ú CheckPointer)ÚLoggingMessageTemplatesÚLogger)Úbuild_all_from_config)Ú GeneralArena)Ú
AlphaZeroGame)ÚGeneralMemoryBuffer)ÚGeneralNetwork)Ú
SearchTree)Ú JavaManager)ÚConfig)Ú MemBufferc ó" — e Zd Z d0dededej dedede d e
d
e d e d ej
j pdd
epddeddfd„Ze d1dee dee
dee dededed e de d
epdfd„¦ « Ze d2dededed e
d
e d e de d
epddepddepdfd„¦ « Ze d1dededed e
d e de depdfd „¦ « Zdefd „Zd edee fd „Z d edee fd „Z d!„ Z d"„ Z d3d#„Z!d$„ Z"d%„ Z#d&„ Z$d'ed(efd)„Z%d'ed(ed*efd+„Z&d,ed-ed.e'd*efd/„Z(dS )4ÚTrainerTNÚnetworkÚgameÚ optimizerÚmemoryÚmuzero_alphazero_configÚ checkpointerÚ search_treeÚ
net_playerÚheadlessÚopponent_network_overrideÚarena_overrideÚ java_managerÚreturnc óÎ — || _ | | _ |
| _ || _ || _ || _ || _ | €| j ¦ « n| | _ || _ || _
|
| _ t d¦ « | _
| € t | j | j | j ¦ « n| | _ || _ t# | j j | j j ¬¦ « | _ g | _ g | _ d S )NzLogs/AlphaZero)ÚlogdirÚtoken)r# Údevicer' Ú game_managerÚmctsr& r Úmake_fresh_instanceÚopponent_networkr! r" r* r Úsummary_writerr Úarenar$ r Úlog_dirÚpushbullet_tokenÚloggerÚarena_win_frequenciesÚlosses)Úselfr r r! r" r# r$ r% r& r/ r' r( r) r* s úV/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Network/trainer.pyÚ__init__zTrainer.__init__" sû € ğ (?ˆÔ$Ø ˆŒ Ø ˆŒ
Ø ˆÔØ ˆŒ Ø$ˆŒØ ˆŒ ØF_ĞFg ¤ × @Ò @Ñ BÔ BĞ Bğ nGˆÔ Ø"ˆŒØ ˆŒ Ø (ˆÔİ +Ğ,<Ñ =Ô =ˆÔà+9Ğ+Aõ ˜4Ô,¨dÔ.JØ œ;ñ(ô (ğ (ØGUğ
Œ
à (ˆÔİ DÔ$@Ô$HØ#'Ô#?Ô#PğRñ Rô RˆŒ à%'ˆÔ"؈Œ ˆ ˆ ó FÚ net_classÚ
tree_classÚnet_player_classÚcheckpoint_pathÚcheckpoint_dirÚcheckpointer_verbosec
óö — t j t j ¦ « rdnd¦ « }
t ||¬¦ « } | |¦ « \ } }
}}}}t
j |¦ « } || ¦ « |¦ « }| d j d |_
| |¦ « }| ¦ « }t j
| ¦ « |¬¦ « } || ¦ « fi ||dœ¤�}| | ¦ « | |¦ « | |
¦ « | |||||| |||
|| ¬¦ « S ) NÚcudaÚcpu©Úverbosez
fc1.weighté )Úlr©r Úmonte_carlo_tree_search)r' r) )Úthr/ rF Ú is_availabler Úload_checkpoint_from_pathr Ú from_argsr2 ÚshapeÚaz_net_linear_input_sizeÚmake_from_configÚoptimÚAdamÚ
parametersÚload_state_dict)Úclsr? r@ rA rB rC r r' rD r) r/ r$ Ú network_dictÚoptimizer_dictr" rK ÚargsÚ
opponent_dictÚconfÚtreer r3 r! r& s r< Úfrom_checkpointzTrainer.from_checkpoint> s– € õ ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ# NĞ<PĞQÑQÔQˆ àHT×HnÒHnØ ñI ô I ÑEˆ �n f¨b°$¸
åÔ Ñ%Ô%ˆØˆz˜$×2Ò2Ñ4Ô4°dÑ;Ô;ˆØ(4°\Ô(BÔ(HÈÔ(KˆÔ%Ø×,Ò,¨TÑ2Ô2ˆØ"×6Ò6Ñ8Ô8Ğİ”H—M’M '×"4Ò"4Ñ"6Ô"6¸2�MÑ>Ô>ˆ à%Ğ% d×&>Ò&>Ñ&@Ô&@ğ _ğ _Ø4;ĞX\Ğ(]Ğ(]ğ_ğ _ˆ
à× Ò Ñ-Ô-Ğ-Ø×(Ò(¨ Ñ7Ô7Ğ7Ø×!Ò! .Ñ1Ô1Ğ1؈s�7˜D )¨V°T¸<ÈÈzĞ[aĞltØ"0ğ2ñ 2ô 2ğ 2r> Úmemory_overridec óø — t j t j ¦ « rdnd¦ « } t || ¦ « \ } }
}| €|n| }t |j |¬¦ « } | |||
|||||| |||
¬¦ « S )NrF rG rH )r' r) r* )rN r/ rF rO r r rC )rY r# r r r% r& r' rD r) ra r* r/ Ú_r! Úmemr" r$ s r< ÚcreatezTrainer.createY s¡ € õ ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ 1Ğ2IÈ6Ñ RÔ Rшˆ9�cØ'Ğ/��°_ˆİ#Ğ$;Ô$JĞThĞiÑiÔiˆ ؈s�7˜D )¨VĞ5LÈlĞ\gĞisØØ $°^ĞR^ğ`ñ `ô `ğ `r> Úpathc
ó| — t j t j ¦ « rdnd¦ « }t ||¦ « \ } }
} t |j |¬¦ « } | t j |¦ « ¦ « t |
¦ « fi | |dœ¤�}
| | ||
| || ||
|||¬¦ « S )NrF rG rH rL )r' r* ) rN r/ rF rO r r rC rX Úloadr r2 )rY rf r# r r% r' rD r* r/ Únetr! r" r$ r& s r< Úfrom_state_dictzTrainer.from_state_dictj sÔ € õ ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ!6Ğ7NĞPVÑ!WÔ!WÑ ˆˆY˜İ#Ğ$;Ô$JĞThĞiÑiÔiˆ Ø ×Ò�BœG D™MœMÑ*Ô*Ğ*İ ˜t× 7Ò 7Ñ 9Ô 9ĞvĞvÈĞitĞ=uĞ=uĞvĞvˆ
؈s�3˜˜i¨Ğ1HÈ,ĞXcĞeoĞqwØ $Ø ,ğ.ñ .ô .ğ .r> c óš — | j | j ¦ « | j t j | j j ¦ « ¦ « | j | j
¦ « ¦ « | j j }| j j }| j j
}| j j }| j
¦ « | t# |¦ « dd¦ « D �] }|| j j k r d| j _ t) j ¦ « 5 | j j }| j t j |¦ « ¦ « | j j dk r˜| j | |¦ « | |¦ « | j | j ||¦ « \ }}} t; | j t< ¦ « r*| j j r | j ¦ « | _ n0| j ! | j
| j || j ¦ « \ }}} | j t j" ||| tF ¦ « ¦ « | j $ d¦ « d d d ¦ « n # 1 swxY w Y | j% & | j
¦ « | j t j' d| j% ( ¦ « ¦ « ¦ « | j% ) | j ¦ « | j t j* d| j% ( ¦ « ¦ « ¦ « | j
+ ¦ « | j t j, |¦ « ¦ « | j
- | j | j ¦ « \ }
} | j t j. |
¦ « ¦ « | j $ d|
› dt_ | ¦ « › d ta | ¦ « › �¦ « | j1 2 | ¦ « | j% 3 | j1 ¦ « | j% 4 | j
| j | j5 | j j6 || j d
¬ ¦ « | j j7 } | j j8 }
| 9 || ¦ « \ }}}}| : |||
|¦ « | ; || |¦ « �Œ | j j | j j | j j< | j j8 | j j
| j j= | j j> | j j7 d œ}| j t j? |¦ « ¦ « | j $ t j@ ¦ « ¦ « | j
S )
NzTraining Progressr rJ zFinished self-play.ztemp checkpointzopponent networkz Mean loss: z , Max loss: z , Min loss: Úlatest_trained_net)Úname)ÚnumItersÚnumSelfPlayGamesÚtauÚupdateThresholdÚ
mcSimulationsÚcÚmaxDepthÚ numPitGames)Ar3 Útor/ r8 Úlogr ÚTRAINING_STARTr# Ú num_itersrX r Ú
state_dictÚepochsÚnum_simulationsÚself_play_gamesÚevalÚ
make_tqdm_barÚrangeÚzero_tau_afterÚ arena_taurN Úno_gradÚ num_workersÚSELF_PLAY_STARTr1 Úparallel_self_playÚmake_n_networksÚ make_n_treesr" Ú
isinstancer Úis_diskr2 Ú self_playÚ
SELF_PLAY_ENDr Úpushbullet_logr$ Úsave_temp_net_checkpointÚSAVEDÚ
get_temp_pathÚload_temp_net_checkpointÚLOADEDÚtrainÚNETWORK_TRAINING_STARTÚ train_netÚNETWORK_TRAINING_ENDÚmaxÚminr: ÚextendÚ save_lossesÚsave_checkpointr! rK Ú
num_pit_gamesÚupdate_thresholdÚ run_pittingÚ check_modelÚrun_pitting_randomrp rs Ú max_depthÚ TRAINING_ENDÚTRAINING_END_PSB)r; ry r{ r| r} ÚiÚn_jobsÚwins_p1Úwins_p2Ú
game_drawsÚ mean_lossr: Ú num_gamesr� Úp1_winsÚp2_winsÚdrawsÚ
wins_totalÚimportant_argss r< r“ z
Trainer.trainw sn € Ø Ô × Ò ¤Ñ-Ô-Ğ-Ø Œ �ŠÕ/Ô>¸tÔ?[Ô?eÑfÔfÑgÔgĞgØ Ô ×-Ò-¨d¬l×.EÒ.EÑ.GÔ.GÑHÔHĞHàÔ0Ô:ˆ ØÔ-Ô4ˆØ Ô6ÔFˆØ Ô6ÔFˆØ Œ ×ÒÑÔĞà×#Ò#¥E¨)Ñ$4Ô$4Ğ6IÈ1ÑMÔMğ ( Cñ ( CˆAØ�DÔ0Ô?Ò?Ğ?Ø9:�Ô,Ô6İ”‘ ” ğ
Bğ
BØ Ô5ÔA�Ø” —’Õ 7Ô GÈÑ XÔ XÑYÔYĞYàÔ/Ô;¸aÒ?Ğ?Ø37´9×3OÒ3OĞPT×PdÒPdĞekÑPlÔPlØPT×PaÒPaĞbhÑPiÔPiØPTÔP[Ğ]aÔ]hØP_ØPVñ 4Xô 4XÑ0�G˜W jõ
" $¤+yÑ9Ô9ğ H¸d¼kÔ>Qğ HØ&*¤k×&EÒ&EÑ&GÔ&G˜œ øà37´9×3FÒ3FÀtÄ|ĞUYÔU`ĞbqØGKÄ{ñ4Tô 4TÑ0�G˜W jà” —’Õ 7Ô EÀgÈwĞXbÕdlÑ mÔ mÑnÔnĞnØ” ×*Ò*Ğ+@ÑAÔAĞAğ!
Bğ
Bğ
Bñ
Bô
Bğ
Bğ
Bğ
Bğ
Bğ
Bğ
Bøøøğ
Bğ
Bğ
Bğ
Bğ$
Ô × 6Ò 6°t´|Ñ DÔ DĞ DØ ŒK�OŠOÕ 3Ô 9Ğ:KÈTÔM^×MlÒMlÑMnÔMnÑ oÔ oÑ pÔ pĞ pØ Ô × 6Ò 6°tÔ7LÑ MÔ MĞ MØ ŒK�OŠOÕ 3Ô :Ğ;MÈtÔO`×OnÒOnÑOpÔOpÑ qÔ qÑ rÔ rĞ rØ ŒL× Ò Ñ Ô Ğ Ø ŒK�OŠOÕ 3Ô JÈ6Ñ RÔ RÑ SÔ SĞ SØ $¤ × 6Ò 6°t´{ÀDÔD`Ñ aÔ aÑ ˆI�vØ ŒK�OŠOÕ 3Ô HÈÑ SÔ SÑ TÔ TĞ TØ ŒK× &Ò &Ğ'r°YĞ'rĞ'rÍCĞPVÉKÌKĞ'rĞ'rÕehĞioÑepÔepĞ'rĞ'rÑ sÔ sĞ sØ ŒK× Ò ˜vÑ &Ô &Ğ &Ø Ô × )Ò )¨$¬+Ñ 6Ô 6Ğ 6Ø Ô × -Ò -¨d¬l¸DÔ<QĞSWÔSaØ.2Ô.JÔ.MÈqĞRVÔRnØ3Gğ
.ñ
Iô
Iğ
Iğ Ô4ÔBˆIØ #Ô ;Ô LĞ Ø26×2BÒ2BÀ?ĞT]Ñ2^Ô2^Ñ /ˆG�W˜e ZØ × Ò ˜W jĞ2BÀAÑ FÔ FĞ FØ × #Ò # O°YÀÑ BÔ BĞ BÑ Bğ Ô4Ô>Ø $Ô <Ô LØÔ/Ô3Ø #Ô ;Ô LØ !Ô 9Ô IØÔ-Ô/Ø Ô4Ô>Ø Ô7ÔEğ
ğ
ˆğ
Œ �ŠÕ/Ô<¸^ÑLÔLÑMÔMĞMØ Œ ×"Ò"Õ#:Ô#KÑ#MÔ#MÑNÔNĞNØŒ|Ğs Ä&E JÊJ ÊJ Únc ó: ‡ — ˆ fd„t |¦ « D ¦ « S )z«
Make n identical copies of self.network using deepcopy.
:param n: The number of copies to make.
:return: A list of n identical networks.
c ó8 •— g | ]}t ‰j ¦ « ‘ŒS © )r r )Ú.0rc r; s €r< ú
<listcomp>z+Trainer.make_n_networks.<locals>.<listcomp> s# ø€ Ğ9Ğ9Ğ9¨1•˜œ Ñ&Ô&Ğ9Ğ9Ğ9r> )r€ )r; r° s ` r< r‡ zTrainer.make_n_networks» s% ø€ ğ :Ğ9Ğ9Ğ9µ°a±´Ğ9Ñ9Ô9Ğ9r> c óŠ — g }t |¦ « D ]0}| j ¦ « }| |¦ « Œ1|S )zŠ
Make n new search trees.
:param n: The number of trees to create.
:return: A list of n new search trees.
)r€ r1 r2 Úappend)r; r° Útreesr¤ r_ s r< rˆ zTrainer.make_n_treesÄ sK € ğ ˆİ�q‘”ğ ğ ˆAà”9×0Ò0Ñ2Ô2ˆDØ �LŠL˜Ñ Ô Ğ Ğ Øˆ r> c ón — t | j ¦ « t t | j ¦ « ¦ « z S ©N)Úsumr9 r Úlen©r; s r< Ú get_arena_win_frequencies_meanz&Trainer.get_arena_win_frequencies_meanÑ s+ € İ�4Ô-Ñ.Ô.µ½#¸dÔ>XÑ:YÔ:YÑ1ZÔ1ZÑZĞZr> c ód — | j ¦ « }| j ¦ « }| j ¦ « }t j || j | j j ||| j ¦ « dœ|¦ « t d |¦ « ¦ « d S )N)r! r" rK ri Úopponent_state_dictr\ z Saved latest model data to {}) r rz r3 r! rN Úsaver" r# rK Úto_dictÚprintÚformat)r; rf rz rÀ Úoptimizer_state_dicts r< Ú save_latestzTrainer.save_latestÔ s¯ € Ø”\×,Ò,Ñ.Ô.ˆ
Ø "Ô 3× >Ò >Ñ @Ô @ĞØ #œ~× 8Ò 8Ñ :Ô :Ğ İ
ŒØ-Ø”kØÔ.Ô1Ø Ø#6ØÔ0×8Ò8Ñ:Ô:ğ
ğ
ğ ñ ô ğ õ Ğ-×4Ò4°TÑ:Ô:Ñ;Ô;Ğ;Ğ;Ğ;r> c óD — | j j rt ||||¬¦ « S |S )N)ÚdescÚpositionÚleave)r# Ú show_tqdmr
)r; ÚiterablerÈ rÉ rÊ s r< r zTrainer.make_tqdm_barâ s, € Ø Ô 'Ô 1ğ ݘ t°hÀeĞLÑLÔLĞ LàˆOr> c ó — | j S rº )r r½ s r< Ú get_networkzTrainer.get_networkè s
€ ØŒ|Ğr> c ó — | j S rº )r1 r½ s r< Úget_treezTrainer.get_treeë s
€ ØŒyĞr> c ó — | j S rº )r# r½ s r< Úget_argszTrainer.get_argsî s € ØÔ+Ğ+r> r| rª c
óÈ — | j ¦ « | j ¦ « | j ¦ « }| | j ¦ « | j ¦ « }| | j ¦ « | j t j |j
|j
|¦ « ¦ « | j ||||d¬¦ « \ }}}t ||z ¦ « }| j t j |j
|j
||||¦ « ¦ « | j ||z ¦ « ||||fS )NF)Únum_mc_simulationsÚ
one_player)r r~ r3 r& r2 Ú set_networkr8 rw r Ú
PITTING_STARTrm r5 Úpitr Ú PITTING_ENDr9 r· ) r; r| rª Úp1Úp2r« r¬ r r® s r< r� zTrainer.run_pittingñ sN € Ø Œ ×ÒÑÔĞØ Ô ×"Ò"Ñ$Ô$Ğ$Ø
Œ_×
0Ò
0Ñ
2Ô
2ˆØ
�Š�t”|Ñ$Ô$Ğ$Ø
Œ_×
0Ò
0Ñ
2Ô
2ˆØ
�Š�tÔ,Ñ-Ô-Ğ-Ø Œ �ŠÕ/Ô=¸b¼gÀrÄwĞPYÑZÔZÑ[Ô[Ğ[Ø"&¤*§.¢.°°R¸ĞWfØ<Ağ #1ñ #Cô #CÑ ˆ�˜%å ˜g¨Ñ /Ñ0Ô0ˆ
Ø Œ �ŠÕ/Ô;¸B¼GÀRÄWÈgØ<CÀZĞQVñXô Xñ Yô Yğ Yà Ô"×)Ò)¨'°IÑ*=Ñ>Ô>Ğ>ؘ ¨
Ğ2Ğ2r> r¤ c ó¶ — || j j z dk rd S | j ¦ « t j ¦ « 5 t
| j ¦ « fi i ¤�}| j ¦ « }|
| j ¦ « | j t j |j |j |¦ « ¦ « | j ||||¬¦ « \ }}}t% ||z ¦ « } | j t j |j |j ||| |¦ « ¦ « d d d ¦ « d S # 1 swxY w Y d S )Nr )rÔ )r# Úrandom_pit_freqr r~ rN rƒ r r0 r2 r& rÖ r8 rw r r× rm r5 rØ r rÙ )
r; r| rª r¤ Ú
random_playerrÚ Úp1_wins_randomÚp2_wins_randomÚ draws_randomr® s
r< r zTrainer.run_pitting_random s¬ € Ø ˆtÔ+Ô;Ñ ;¸qÒ @Ğ @Ø ˆFØ Œ ×ÒÑÔĞİ
ŒZ‰\Œ\ğ _ğ _İ (¨Ô):×)NÒ)NÑ)PÔ)PĞ WĞ WĞTVĞ WĞ WˆMà” ×4Ò4Ñ6Ô6ˆBØ �NŠN˜4œ<Ñ (Ô (Ğ (Ø ŒK�OŠOÕ 3Ô AÀ"Ä'È=ÔK]Ğ_hÑ iÔ iÑ jÔ jĞ jØ;?¼:¿>º>È"ÈmĞ]fØ]lğ <Jñ <nô <nÑ 8ˆN˜N¨Lå! .°>Ñ"AÑBÔBˆJØ ŒK�OŠOİ'Ô3°B´G¸]Ô=OĞQ_Ø4BÀJĞP\ñ^ô ^ñ
_ô
_ğ
_ğ _ğ _ğ _ñ _ô _ğ _ğ _ğ _ğ _ğ _ğ _ğ _øøøğ _ğ _ğ _ğ _ğ _ğ _s ÁC?EÅEÅEr« r® r� c óú — ||z |k r²| j t j ||z |¦ « ¦ « | j | j | j | j | j j
|| j ¦ « | j t j d| j ¦ « ¦ « ¦ « n“| j t j
||z |¦ « ¦ « | j | j ¦ « | j t j d| j ¦ « ¦ « ¦ « | j t j |¦ « ¦ « d S )Nzaccepted model checkpointzprevious version checkpoint)r8 rw r Ú MODEL_ACCEPTr$ r› r r3 r! r# rK r� Úget_checkpoint_dirÚ MODEL_REJECTr‘ r’ r� r� ÚITER_FINISHED_PSB)r; r« r® r� r¤ s r< rŸ zTrainer.check_model s� € Ø �ZÑ Ğ"2Ò 2Ğ 2Ø ŒK�OŠOÕ 3Ô @ÀÈ:ÑAUØAQñ Sô Sñ
Tô
Tğ
Tà Ô × -Ò -¨d¬l¸DÔ<QĞSWÔSaØ.2Ô.JÔ.MÈqØ.2Ô.Jñ
Lô
Lğ
Lğ
ŒK�OŠOÕ 3Ô 9Ğ:UØ:>Ô:K×:^Ò:^Ñ:`Ô:`ñ bô bñ
cô
cğ
cğ
cğ
ŒK�OŠOÕ 3Ô @ÀÈ:ÑAUØAQñ Sô Sñ
Tô
Tğ
Tà Ô × 6Ò 6°t´|Ñ DÔ DĞ DØ ŒK�OŠOÕ 3Ô :Ğ;XØ;?Ô;L×;ZÒ;ZÑ;\Ô;\ñ ^ô ^ñ
_ô
_ğ
_à Œ ×"Ò"Õ#:Ô#LÈQÑ#OÔ#OÑPÔPĞPĞPĞPr> )TNNN)TFN)TFNNN)T))Ú__name__Ú
__module__Ú __qualname__r r rN rU r r r r r
ÚboolÚnnÚModuler r r= Ú classmethodr Ústrr` re rj r r“ ÚintÚlistr‡ r rˆ r¾ rÆ r rÎ rĞ rÒ r� r ÚfloatrŸ r³ r> r< r r ! sÁ € € € € € ğ
+/ØCGØ8<Ø-1ğğ ğ °mğ Ø œHğØ.Ağà*0ğà@Lğğ )ğğ 7=ğğ $(ğ ğ
-/¬E¬LĞ,@¸Dğ ğ ".Ğ!5°ğ
ğ +ğğ 7;ğğ ğ ğ ğ8 ğ ?CØ5:Ø?Cğ 2ğ 2¨¨^Ô(<ğ 2È$ÈzÔJZğ 2Ø*.¨v¬,ğ2à),ğ2à>Ağ2ğ ,ğ2ğ 8<ğ2ğ /3ğ 2ğ
)5Ğ(<¸ğ 2ğ 2ğ 2ñ „[ğ2ğ4 ğ !%Ø,1Ø6:Ø>BØ37ğ`ğ `¨Vğ `¸=ğ `ĞSağ `Ø &ğ`à!ğ`ğ ğ`ğ &*ğ `ğ
,Ğ 3¨tğ `ğ !4Ğ ;°tğ
`ğ )Ğ 0¨Dğ`ğ `ğ `ñ „[ğ`ğ à)-Ø`dğ
.ğ
. 3ğ
.Àğ
.È}ğ
.Ğkuğ
.Ø"&ğ
.à.2ğ
.àJUĞJ]ĞY]ğ
.ğ
.ğ
.ñ „[ğ
.ğB �|ğ B ğ B ğ B ğ B ğH: ğ :¨¨lÔ);ğ :ğ :ğ :ğ :ğ ˜cğ d¨<Ô&8ğ ğ ğ ğ ğ[ğ [ğ [ğ <ğ <ğ <ğ ğ ğ ğ ğ ğ ğ ğğ ğ ğ,ğ ,ğ ,ğ3¨3ğ 3¸3ğ 3ğ 3ğ 3ğ 3ğ _°#ğ _À#ğ _È#ğ _ğ _ğ _ğ _ğ"Q 3ğ Q°Cğ QÈ5ğ QĞUXğ Qğ Qğ Qğ Qğ Qğ Qr> r )/Útorch.multiprocessingr Úmu_alpha_zero.General.utilsr Úcopyr Útypingr ÚtorchrN Útorch.utils.tensorboardr r
Ú#mu_alpha_zero.AlphaZero.Arena.arenar Ú%mu_alpha_zero.AlphaZero.Arena.playersr r
r Ú+mu_alpha_zero.AlphaZero.MCTS.az_search_treer Ú$mu_alpha_zero.AlphaZero.Network.nnetr Ú$mu_alpha_zero.AlphaZero.checkpointerr Ú mu_alpha_zero.AlphaZero.loggerr r Ú mu_alpha_zero.AlphaZero.utilsr Úmu_alpha_zero.General.arenar Ú mu_alpha_zero.General.az_gamer Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr Ú!mu_alpha_zero.General.search_treer Ú-mu_alpha_zero.MuZero.JavaGateway.java_managerr Úmu_alpha_zero.configr Úmu_alpha_zero.mem_bufferr r r³ r> r< ú<module>r sù ğØ 2Ğ 2Ğ 2Ğ 2Ğ 2Ğ 2à 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0à Ğ � Ğ %Ñ %Ô %Ğ %Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø Ğ Ğ Ğ Ğ Ğ Ø 5Ğ 5Ğ 5Ğ 5Ğ 5Ğ 5Ø PĞ PĞ PĞ PĞ PĞ PĞ PĞ PĞ PĞ PØ DĞ DĞ DĞ DĞ DĞ DØ =Ğ =Ğ =Ğ =Ğ =Ğ =Ø =Ğ =Ğ =Ğ =Ğ =Ğ =Ø JĞ JĞ JĞ JĞ JĞ JĞ JĞ JØ ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 7Ğ 7Ğ 7Ğ 7Ğ 7Ğ 7Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø EĞ EĞ EĞ EĞ EĞ EØ 'Ğ 'Ğ 'Ğ 'Ğ 'Ğ 'Ø .Ğ .Ğ .Ğ .Ğ .Ğ .ğ @Qğ @Qğ @Qğ @Qğ @Qñ @Qô @Qğ @Qğ @Qğ @Qr> | 21,360 | Python | .py | 95 | 223.126316 | 1,279 | 0.33802 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,522 | asteroids.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/asteroids.py | import numpy as np
import torch as th
from gymnasium import make
from mu_alpha_zero.General.mz_game import MuZeroGame
class Asteroids(MuZeroGame):
def get_state_for_passive_player(self, state: np.ndarray, player: int):
return state
def __init__(self):
self.env = make("ALE/Asteroids-v5")
self.done = False
def reset(self) -> np.ndarray or th.Tensor:
obs, _ = self.env.reset()
return obs
def get_noop(self) -> int:
return 0
def get_num_actions(self) -> int:
return self.env.action_space.n
def game_result(self, player: int or None) -> bool or None:
return self.done
def make_fresh_instance(self):
return Asteroids()
def get_next_state(self, action: int, player: int or None, frame_skip: int = 4) -> (
np.ndarray or th.Tensor, int, bool):
obs, rew, done, _, _ = self.env.step(action)
self.done = done
return obs, rew, done
def frame_skip_step(self, action: int, player: int or None, frame_skip: int = 4) -> (
np.ndarray or th.Tensor, int, bool):
obs, rew, done = self.get_next_state(action, player)
for i in range(frame_skip - 1):
obs, rew, done = self.get_next_state(action, player)
return obs, rew, done
def render(self):
pass
def get_random_valid_action(self, state: np.ndarray, **kwargs):
return self.env.action_space.sample()
def get_invalid_actions(self, state: np.ndarray, player: int):
return np.ones((self.get_num_actions()))
| 1,576 | Python | .py | 38 | 34.078947 | 89 | 0.630335 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,523 | tictactoe_game.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/tictactoe_game.py | import io
import random
import sys
import numpy as np
import pygame as pg
import torch as th
from PIL import Image
from mu_alpha_zero.General.az_game import AlphaZeroGame
class TicTacToeGameManager(AlphaZeroGame):
"""
This class is the game manager for the game of Tic Tac Toe and its variants.
"""
def __init__(self, board_size: int, headless: bool, num_to_win=None) -> None:
# TODO: Implement the possibility to play over internet using sockets.
self.player = 1
self.enemy_player = -1
self.board_size = board_size
self.board = self.initialize_board()
self.headless = headless
self.num_to_win = self.init_num_to_win(num_to_win)
self.screen = self.create_screen(headless)
def play(self, player: int, index: tuple) -> None:
self.board[index] = player
def init_num_to_win(self, num_to_win: int | None) -> int:
if num_to_win is None:
num_to_win = self.board_size
if num_to_win > self.board_size:
raise Exception("Num to win can't be greater than board size")
return num_to_win
def initialize_board(self):
board = np.zeros((self.board_size, self.board_size), dtype=np.int8)
return board
def get_random_valid_action(self, observations: np.ndarray, **kwargs) -> list:
valid_moves = self.get_valid_moves(observations)
if len(valid_moves) == 0:
raise Exception("No valid moves")
return random.choice(valid_moves)
def create_screen(self, headless):
if headless:
return
pg.init()
pg.display.set_caption("Tic Tac Toe")
board_rect_size = self.board_size * 100
screen = pg.display.set_mode((board_rect_size, board_rect_size))
return screen
def full_iterate_array(self, arr: np.ndarray, func: callable) -> list:
"""
This function iterates over all rows, columns and the two main diagonals of supplied array,
applies the supplied function to each of them and returns the results in a list.
:param arr: a 2D numpy array.
:param func: a callable function that takes a 1D array as input and returns a result.
:return: A list of results.
"""
results = []
results.append("row")
for row in arr:
results.append(func(row.reshape(-1)))
results.append("col")
for col in arr.T:
results.append(func(col.reshape(-1)))
diags = [np.diag(arr, k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
flipped_diags = [np.diag(np.fliplr(arr), k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
diags.extend(flipped_diags)
for idx, diag in enumerate(diags):
if idx == 0:
results.append("diag_left")
elif idx == len(diags) // 2:
results.append("diag_right")
results.append(func(diag.reshape(-1)))
# results.append("diag_left")
# results.append(func(arr.diagonal().reshape(-1)))
# results.append("diag_right")
# results.append(func(np.fliplr(arr).diagonal().reshape(-1)))
return results
def eval_board(self, board: np.ndarray, check_end: bool = True) -> int | None:
if self.is_board_full(board):
return 0
score = 0
for i in range(self.num_to_win, self.num_to_win // 2, -1):
current_won = self.check_partial_win(-1, self.num_to_win, board=board)
opp_won = self.check_partial_win(1, self.num_to_win, board=board)
if current_won:
score += 1 * 2
if opp_won:
score -= 1
if check_end:
return None if self.game_result(-1, board) is None else score
return score
def full_iterate_array_all_diags(self, arr: np.ndarray, func: callable):
results = []
for row in arr:
results.append(func(row.reshape(-1)))
for col in arr.T:
results.append(func(col.reshape(-1)))
diags = [np.diag(arr, k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
flipped_diags = [np.diag(np.fliplr(arr), k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
diags.extend(flipped_diags)
for diag in diags:
# if diag.size < self.num_to_win:
# continue
results.append(func(diag.reshape(-1)))
return results
def extract_all_vectors(self, board: np.ndarray):
vectors = board.tolist() + board.T.tolist()
vectors.extend([np.diag(board, k=i) for i in range(-board.shape[0] + 1, board.shape[1])])
vectors.extend([np.diag(np.fliplr(board), k=i) for i in range(-board.shape[0] + 1, board.shape[1])])
# pad vectors with -3's to have the same length as the board size and stack them
vectors = [np.pad(vector, (0, self.board_size - len(vector)), constant_values=-3) for vector in vectors]
return np.array(vectors)
def check_win(self, player: int, board=None) -> bool:
# if self.num_to_win == self.board_size:
# return self.check_full_win(player, board=board)
# else:
return self.check_partial_win(player, self.num_to_win, board=board)
def check_full_win(self, player: int, board=None) -> bool:
"""
This function checks if the supplied player has won the game with a full win (num_to_win == board_size).
:param player: The player to check for (1 or -1).
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
"""
if board is None:
board = self.get_board()
matches = self.full_iterate_array(board, lambda part: np.all(part == player))
for match in matches:
if not isinstance(match, str) and match:
return True
return False
def check_partial_win(self, player: int, n: int, board=None) -> bool:
"""
This function checks if the supplied player has won the game with a partial win (num_to_win < board_size).
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
"""
if board is None:
board = self.get_board()
matches = self.full_iterate_array_all_diags(board,
lambda part:
np.convolve((part == player), np.ones(n, dtype=np.int8),
"valid"))
for match in matches:
if np.any(match == n):
return True
return False
def check_partial_win_vectorized(self, player: int, n: int, board=None) -> bool:
if board is None:
board = self.get_board()
vectors = th.tensor(self.extract_all_vectors(board)).unsqueeze(0)
weight = th.ones((1, 1, 1, n), dtype=th.long)
vectors_where_player = th.where(vectors == player, 1, 0).long()
res = th.nn.functional.conv2d(vectors_where_player, weight=weight)
return th.any(res == n).item()
def check_partial_win_to_index(self, player: int, n: int, board=None) -> dict[tuple, str] | dict:
"""
This variation of check_partial_win returns the index of the first partial win found.
The index being the index of the first piece in the winning sequence.
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: A dictionary containing the index and the position of the winning sequence.
"""
if board is None:
board = self.get_board()
indices = self.full_iterate_array(board,
lambda part:
np.convolve((part == player), np.ones(n, dtype=int),
"valid"))
indices = [x.tolist() if not isinstance(x, str) else x for x in indices]
pos = "row"
for vector in indices:
if isinstance(vector, str):
pos = vector
continue
for el in vector:
if el == n:
vector_index = indices.index(vector)
pos_index = indices.index(pos)
# num_strings_before = len([x for index,x in enumerate(indices) if isinstance(x,str) and index < pos_index])
element_index = vector_index - (pos_index + 1) if vector_index > pos_index else indices.index(
vector, pos_index)
diag_idx = {
"index": ((self.board_size - 1) - element_index, 0)} if element_index < self.board_size else {
"index": (0, element_index - self.board_size)}
diag_idx["pos"] = pos
match pos:
case "row":
return {"index": (element_index, 0), "pos": pos}
case "col":
return {"index": (0, element_index), "pos": pos}
case "diag_left":
return diag_idx
case "diag_right":
return diag_idx
# case "diag_right":
return {"index": None, "pos": None}
def return_index_if_valid(self, index: tuple, return_on_fail: tuple = ()) -> tuple:
if index[0] < 0 or index[0] >= self.board_size or index[1] < 0 or index[1] >= self.board_size:
return return_on_fail
return index
def make_fresh_instance(self):
return TicTacToeGameManager(self.board_size, self.headless, num_to_win=self.num_to_win)
def get_previous(self, index: tuple, pos: str, n: int):
if np.array(index).all() == 0:
return index
match pos:
case "row":
return self.return_index_if_valid((index[0], index[1] - n), return_on_fail=index)
case "col":
return self.return_index_if_valid((index[0] - n, index[1]), return_on_fail=index)
case "diag_left":
return self.return_index_if_valid((index[0] - n, index[1] - n), return_on_fail=index)
case "diag_right":
return self.return_index_if_valid((index[0] - n, index[1] + n), return_on_fail=index)
def get_next(self, index: tuple, pos: str, n: int):
if np.array(index).all() == self.board_size - 1:
return index
match pos:
case "row":
return self.return_index_if_valid((index[0], index[1] + n), return_on_fail=index)
case "col":
return self.return_index_if_valid((index[0] + n, index[1]), return_on_fail=index)
case "diag_left":
return self.return_index_if_valid((index[0] + n, index[1] + n), return_on_fail=index)
case "diag_right":
return self.return_index_if_valid((index[0] + n, index[1] - n), return_on_fail=index)
def is_board_full(self, board=None) -> bool:
if board is None:
board = self.get_board()
return np.all(board != 0)
def get_board(self):
return self.board.copy()
def reset(self):
self.board = self.initialize_board()
return self.board.copy()
def get_board_size(self):
return self.board_size
def render(self) -> bool:
if self.headless:
return False
self.screen.fill((0, 0, 0))
self._draw_board()
for row in range(self.board_size):
for col in range(self.board_size):
if self.board[row][col] == self.player:
self._draw_circle(col * 100 + 50, row * 100 + 50)
elif self.board[row][col] == self.enemy_player:
self._draw_cross(col * 100 + 50, row * 100 + 50)
pg.event.pump()
pg.display.flip()
return True
def _draw_circle(self, x, y) -> None:
if self.headless:
return
pg.draw.circle(self.screen, "green", (x, y), 40, 1)
def _draw_cross(self, x, y) -> None:
if self.headless:
return
pg.draw.line(self.screen, "red", (x - 40, y - 40), (x + 40, y + 40), 1)
pg.draw.line(self.screen, "red", (x + 40, y - 40), (x - 40, y + 40), 1)
def _draw_board(self):
for x in range(0, self.board_size * 100, 100):
for y in range(0, self.board_size * 100, 100):
pg.draw.rect(self.screen, (255, 255, 255), pg.Rect(x, y, 100, 100), 1)
def is_empty(self, index: tuple) -> bool:
return self.board[index] == 0
def get_valid_moves(self, observation: np.ndarray, player: int or None = None) -> list:
"""
Legal moves are the empty spaces on the board.
:param observation: A 2D numpy array representing the current state of the game.
:param player: The player to check for. Since the game is symmetric, this is ignored.
:return: A list of legal moves.
"""
legal_moves = []
observation = observation.reshape(self.board_size, self.board_size)
for row in range(self.board_size):
for col in range(self.board_size):
if observation[row][col] == 0:
legal_moves.append([row, col])
return legal_moves
def pygame_quit(self) -> bool:
if self.headless:
return False
pg.quit()
return True
def get_click_coords(self):
if self.headless:
return
mouse_pos = (x // 100 for x in pg.mouse.get_pos())
if pg.mouse.get_pressed()[0]: # left click
return mouse_pos
def get_human_input(self, board: np.ndarray):
if self.headless:
return
while True:
self.check_pg_events()
if self.get_click_coords() is not None:
x, y = self.get_click_coords()
if board[y][x] == 0:
return y, x
# time.sleep(1 / 60)
def check_pg_events(self):
if self.headless:
return
for event in pg.event.get():
if event.type == pg.QUIT:
self.pygame_quit()
sys.exit(0)
def network_to_board(self, move):
"""
Converts an integer move from the network to a board index.
:param move: An integer move selected from the network probabilities.
:return: A tuple representing the board index (int,int).
"""
return np.unravel_index(move, self.board.shape)
@staticmethod
def get_canonical_form(board, player) -> np.ndarray:
return board * player
def get_next_state(self, board: np.ndarray, action: int or tuple, player: int) -> np.ndarray:
if isinstance(action, int):
action = self.network_to_board(action)
board_ = board.copy()
board_[action] = player
return board_
def set_headless(self, val: bool):
self.headless = val
def get_invalid_actions(self, state: np.ndarray, player: int):
mask = np.where(state == 0, 1, 0)
return mask
def __str__(self):
return str(self.board).replace('1', 'X').replace('-1', 'O')
#
# if __name__ == "__main__":
# sample_arr = np.array([[-1,1,-1,1,1],[-1,1,1,-1,-1],[1,-1,1,1,0],[-1,1,1,-1,0],[0,0,-1,1,1]])
# game_manager = GameManager(5, True,num_to_win=3)
# res = game_manager.check_partial_win_vectorized(1,3,sample_arr)
# res = game_manager.game_result(-1,sample_arr)
# print(res)
# print(game_manager.check_partial_win(1,3,sample_arr))
| 16,109 | Python | .py | 340 | 36.097059 | 128 | 0.567581 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,524 | tictactoe_game.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/__pycache__/tictactoe_game.cpython-311.pyc | §
Uxf"D ã ó~ — d dl Z d dlZd dlZd dlmZ d dlZd dlZ d dl
Z d dl Z
d dlmZ d dlmZ G d„ de¦ « ZdS )é N)ÚImage)Ú
AlphaZeroGamec óŞ — e Zd ZdZd?dededdfd„Zdededdfd „Zd
edz defd „Z d „ Z
d
e j de
fd„Zd„ Zde j dede
fd„Zd@de j dededz fd„Zde j defd„Zde j fd„Zd?dedefd„Zd?dedefd„Zd?dededefd „Zd?dededefd „Zd?dededeeef ez fd „ZdAded edefd!„Z d"„ Z ded#edefd$„Z ded#edefd%„Z d?defd&„Z d'„ Z!d(„ Z"d)„ Z#defd*„Z$dBd+„Z%dBd,„Z&d-„ Z'dedefd.„Z(d?d/e j depdde
fd0„Z)defd1„Z*d2„ Z+de j fd3„Z,d4„ Z-d5„ Z.d6„ Z/e0de j fd7„¦ « Z1de j d8epedede j fd9„Z2d:efd;„Z3d<e j defd=„Z4d>„ Z5dS )CÚTicTacToeGameManagerzV
This class is the game manager for the game of Tic Tac Toe and its variants.
NÚ
board_sizeÚheadlessÚreturnc óØ — d| _ d| _ || _ | ¦ « | _ || _ | |¦ « | _ | |¦ « | _ d S )Né éÿÿÿÿ)
ÚplayerÚ enemy_playerr Úinitialize_boardÚboardr Úinit_num_to_winÚ
num_to_winÚ
create_screenÚscreen)Úselfr r r s úP/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Game/tictactoe_game.pyÚ__init__z TicTacToeGameManager.__init__ s` € àˆŒ Ø ˆÔØ$ˆŒØ×*Ò*Ñ,Ô,ˆŒ
Ø ˆŒ
Ø ×.Ò.¨zÑ:Ô:ˆŒØ×(Ò(¨Ñ2Ô2ˆŒ ˆ ˆ ó r
Úindexc ó — || j |< d S ©N©r )r r
r s r ÚplayzTicTacToeGameManager.play s € Ø "ˆŒ
�5ÑĞĞr r c óL — |€| j }|| j k rt d¦ « ‚|S )Nz+Num to win can't be greater than board size)r Ú Exception)r r s r r z$TicTacToeGameManager.init_num_to_win! s1 € Ø Ğ Ø œ ˆJØ ˜œÒ 'Ğ 'İĞ IÑJÔJĞ JØĞr c ó^ — t j | j | j ft j ¬¦ « }|S )N©Údtype)ÚnpÚzerosr Úint8©r r s r r z%TicTacToeGameManager.initialize_board( s& € İ”˜$œ/¨4¬?Ğ;Å2Ä7ĞKÑKÔKˆØˆ r Ú observationsc ó˜ — | |¦ « }t |¦ « dk rt d¦ « ‚t j |¦ « S )Nr zNo valid moves)Úget_valid_movesÚlenr ÚrandomÚchoice)r r' ÚkwargsÚ valid_movess r Úget_random_valid_actionz,TicTacToeGameManager.get_random_valid_action, sG € Ø×*Ò*¨<Ñ8Ô8ˆ İ ˆ{Ñ Ô ˜qÒ Ğ İĞ ,Ñ-Ô-Ğ -İŒ}˜[Ñ)Ô)Ğ)r c óÈ — |rd S t j ¦ « t j d¦ « | j dz }t j ||f¦ « }|S )Nz Tic Tac Toeéd )ÚpgÚinitÚdisplayÚ set_captionr Úset_mode)r r Úboard_rect_sizer s r r z"TicTacToeGameManager.create_screen2 s[ € Ø ğ Ø ˆFå
Œ‰ Œ ˆ İ
Œ
× Ò ˜}Ñ-Ô-Ğ-Ø œ/¨CÑ/ˆİ”×$Ò$ o°Ğ%GÑHÔHˆØˆ
r ÚarrÚfuncc ó^ ‡— g }| d¦ « ‰D ]3}| || d¦ « ¦ « ¦ « Œ4| d¦ « ‰j D ]3}| || d¦ « ¦ « ¦ « Œ4ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }| |¦ « t
|¦ « D ]}\ }} |dk r| d¦ « n+|t |¦ « d z k r| d
¦ « | || d¦ « ¦ « ¦ « Œ~|S ) an
This function iterates over all rows, columns and the two main diagonals of supplied array,
applies the supplied function to each of them and returns the results in a list.
:param arr: a 2D numpy array.
:param func: a callable function that takes a 1D array as input and returns a result.
:return: A list of results.
Úrowr Úcolc ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS ©)Úk©r# Údiag©Ú.0Úir8 s €r ú
<listcomp>z;TicTacToeGameManager.full_iterate_array.<locals>.<listcomp>M ó( ø€ ĞSĞSĞS q•”˜ Ğ"Ñ"Ô"ĞSĞSĞSr r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> ©r# rA ÚfliplrrB s €r rE z;TicTacToeGameManager.full_iterate_array.<locals>.<listcomp>N ó0 ø€ ĞfĞfĞf¸!�œ¥¤¨3¡ ¤ °1Ğ5Ñ5Ô5ĞfĞfĞfr Ú diag_lefté Ú
diag_right)ÚappendÚreshapeÚTÚrangeÚshapeÚextendÚ enumerater* )
r r8 r9 Úresultsr; r< ÚdiagsÚ
flipped_diagsÚidxrA s
` r Úfull_iterate_arrayz'TicTacToeGameManager.full_iterate_array< s¶ ø€ ğ ˆØ�Š�uÑ Ô Ğ Øğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1à�Š�uÑ Ô Ğ Ø”5ğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1àSĞSĞSĞSE°3´9¸Q´<°-À!Ñ2CÀSÄYÈqÄ\Ñ,RÔ,RĞSÑSÔSˆØfĞfĞfĞf½uÀcÄiĞPQÄlÀ]ĞUVÑEVĞX[ÔXaĞbcÔXdÑ?eÔ?eĞfÑfÔfˆ
Ø
� Š �]Ñ#Ô#Ğ#İ" 5Ñ)Ô)ğ 3ğ 3‰IˆC�Ø�aŠxˆxØ—’˜{Ñ+Ô+Ğ+Ğ+Ø�˜E™
œ
a™Ò'Ğ'Ø—’˜|Ñ,Ô,Ğ,Ø �NŠN˜4˜4 § ¢ ¨RÑ 0Ô 0Ñ1Ô1Ñ 2Ô 2Ğ 2Ğ 2ğ ˆr Tr Ú check_endc óB — | |¦ « rdS d}t | j | j dz d¦ « D ]J}| d| j |¬¦ « }| d| j |¬¦ « }|r|dz
}|r|dz }ŒK|r| d|¦ « €d n|S |S )Nr rL r r r )Ú
is_board_fullrQ r Úcheck_partial_winÚ game_result)r r rZ ÚscorerD Ú current_wonÚopp_wons r Ú
eval_boardz TicTacToeGameManager.eval_board] sÎ € Ø × Ò ˜eÑ $Ô $ğ Ø�1؈İ�t”¨¬¸1Ñ(<¸bÑAÔAğ ğ ˆAØ ×0Ò0°°T´_ÈEĞ0ÑRÔRˆKØ×,Ò,¨Q°´ÀuĞ,ÑMÔMˆGØğ
ؘ‘�Øğ
ؘ‘
�øØ ğ JØ ×+Ò+¨B°Ñ6Ô6Ğ>�4�4ÀEĞ I؈ r c ó\ ‡— g }‰D ]3}| || d¦ « ¦ « ¦ « Œ4‰j D ]3}| || d¦ « ¦ « ¦ « Œ4ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }| |¦ « |D ]3}| || d¦ « ¦ « ¦ « Œ4|S )Nr c ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS r> r@ rB s €r rE zETicTacToeGameManager.full_iterate_array_all_diags.<locals>.<listcomp>u rF r r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> rH rB s €r rE zETicTacToeGameManager.full_iterate_array_all_diags.<locals>.<listcomp>v rJ r )rN rO rP rQ rR rS ) r r8 r9 rU r; r< rV rW rA s ` r Ú full_iterate_array_all_diagsz1TicTacToeGameManager.full_iterate_array_all_diagsl s? ø€ àˆØğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1à”5ğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1àSĞSĞSĞSE°3´9¸Q´<°-À!Ñ2CÀSÄYÈqÄ\Ñ,RÔ,RĞSÑSÔSˆØfĞfĞfĞf½uÀcÄiĞPQÄlÀ]ĞUVÑEVĞX[ÔXaĞbcÔXdÑ?eÔ?eĞfÑfÔfˆ
Ø
� Š �]Ñ#Ô#Ğ#Øğ 3ğ 3ˆDğ
�NŠN˜4˜4 § ¢ ¨RÑ 0Ô 0Ñ1Ô1Ñ 2Ô 2Ğ 2Ğ 2àˆr c óÊ ‡ ‡— ‰ ¦ « ‰j ¦ « z }| ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « ¦ « | ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « ¦ « ˆ fd„|D ¦ « }t j |¦ « S )Nc ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS r> r@ ©rC rD r s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>� s( ø€ Ğ`Ğ`Ğ`°�œ ¨Ğ+Ñ+Ô+Ğ`Ğ`Ğ`r r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> rH ri s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>‚ s2 ø€ ĞkĞkĞk¸1�œ¥¤ ¨%Ñ 0Ô 0°AĞ6Ñ6Ô6ĞkĞkĞkr c
ól •— g | ]0}t j |d ‰j t |¦ « z
fd¬¦ « ‘Œ1S )r éıÿÿÿ)Úconstant_values)r# Úpadr r* )rC Úvectorr s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>„ s> ø€ ĞpĞpĞpĞ^d•2”6˜& 1 d¤o½¸F¹ ¼ Ñ&CĞ"DĞVXĞYÑYÔYĞpĞpĞpr )ÚtolistrP rS rQ rR r# Úarray)r r Úvectorss `` r Úextract_all_vectorsz(TicTacToeGameManager.extract_all_vectors sŞ øø€ Ø—,’,‘.”. 5¤7§>¢>Ñ#3Ô#3Ñ3ˆØ�ŠĞ`Ğ`Ğ`Ğ`µU¸E¼KȼN¸?ÈQÑ;NĞPUÔP[Ğ\]ÔP^Ñ5_Ô5_Ğ`Ñ`Ô`ÑaÔaĞaØ�ŠĞkĞkĞkĞkÅÀuÄ{ĞSTÄ~ÀoĞXYÑFYĞ[`Ô[fĞghÔ[iÑ@jÔ@jĞkÑkÔkÑlÔlĞlàpĞpĞpĞpĞhoĞpÑpÔpˆİŒx˜Ñ Ô Ğ r c ó< — | || j |¬¦ « S )Nr )r] r )r r
r s r Ú check_winz TicTacToeGameManager.check_win‡ s! € ğ ×%Ò% f¨d¬oÀUĞ%ÑKÔKĞKr c ó¤ ‡— |€| ¦ « }| |ˆfd„¦ « }|D ] }t |t ¦ « s|r dS Œ dS )aC
This function checks if the supplied player has won the game with a full win (num_to_win == board_size).
:param player: The player to check for (1 or -1).
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
Nc ó4 •— t j | ‰k ¦ « S r )r# Úall)Úpartr
s €r ú<lambda>z5TicTacToeGameManager.check_full_win.<locals>.<lambda>– s ø€ ½b¼fÀTÈVÂ^Ñ>TÔ>T€ r TF)Ú get_boardrY Ú
isinstanceÚstr)r r
r ÚmatchesÚmatchs ` r Úcheck_full_winz#TicTacToeGameManager.check_full_win� sm ø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×)Ò)¨%Ğ1TĞ1TĞ1TĞ1TÑUÔUˆØ ğ ğ ˆEİ ˜e¥SÑ)Ô)ğ
¨eğ
Ø�t�tøàˆur Únc óª ‡‡— |€| ¦ « }| |ˆˆfd„¦ « }|D ] }t j |‰k ¦ « r dS Œ dS )aˆ
This function checks if the supplied player has won the game with a partial win (num_to_win < board_size).
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
Nc ót •— t j | ‰k t j ‰t j ¬¦ « d¦ « S ©Nr! Úvalid)r# ÚconvolveÚonesr% ©ry r� r
s €€r rz z8TicTacToeGameManager.check_partial_win.<locals>.<lambda>© s: ø€ İ46´KÀÈÂ ÕRTÔRYĞZ[ÕceÔcjĞRkÑRkÔRkØ@Gñ5Iô 5Iğ r TF)r{ rf r# Úany)r r
r� r r~ r s `` r r] z&TicTacToeGameManager.check_partial_win� s‹ øø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×3Ò3°Eğ5Iğ 5Iğ 5Iğ 5Iğ 5IñJô Jˆğ
ğ ğ ˆEİŒv�e˜q’jÑ!Ô!ğ
Ø�t�tğ
ğ ˆur c óâ — |€| ¦ « }t j | |¦ « ¦ « d¦ « }t j ddd|ft j ¬¦ « }t j ||k dd¦ « ¦ « }t j j
||¬¦ « }t j ||k ¦ « ¦ « S )Nr r r! )Úweight)
r{ ÚthÚtensorrs Ú unsqueezer‡ ÚlongÚwhereÚnnÚ
functionalÚconv2dr‰ Úitem)r r
r� r rr r‹ Úvectors_where_playerÚress r Ú check_partial_win_vectorizedz1TicTacToeGameManager.check_partial_win_vectorized³ s € Ø ˆ=Ø—N’NÑ$Ô$ˆEİ”)˜D× 4Ò 4°UÑ ;Ô ;Ñ<Ô<×FÒFÀqÑIÔIˆİ”˜!˜Q 1˜ R¬WĞ5Ñ5Ô5ˆİ !œx¨°6Ò(9¸1¸aÑ @Ô @× EÒ EÑ GÔ GĞ İŒeÔ ×%Ò%Ğ&:À6Ğ%ÑJÔJˆİŒv�c˜Q’hÑ Ô ×$Ò$Ñ&Ô&Ğ&r c óP ‡‡— |€| ¦ « }| |ˆˆfd„¦ « }d„ |D ¦ « }d}|D ]ß}t |t ¦ « r|}Œ|D ]Â}|‰k rº| |¦ « }| |¦ « } || k r|| dz z
n| || ¦ « }
|
| j k rd| j dz
|
z
dfin
dd|
| j z
fi} || d< |xdk r |
df|d œc c S xd
k r d|
f|d œc c S xd k r | c c S d k r| c c S ŒÃŒàddd œS )
aß
This variation of check_partial_win returns the index of the first partial win found.
The index being the index of the first piece in the winning sequence.
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: A dictionary containing the index and the position of the winning sequence.
Nc ój •— t j | ‰k t j ‰t ¬¦ « d¦ « S r„ )r# r† r‡ Úintrˆ s €€r rz zATicTacToeGameManager.check_partial_win_to_index.<locals>.<lambda>È s3 ø€ İ*,¬+°t¸v²~ÍÌĞPQÕY\ĞH]ÑH]ÔH]Ø6=ñ+?ô +?ğ r c ód — g | ]-}t |t ¦ « s| ¦ « n|‘Œ.S © )r| r} rp ©rC Úxs r rE zCTicTacToeGameManager.check_partial_win_to_index.<locals>.<listcomp>Ë s3 € ĞPĞPĞPÀ1¥Z°µ3Ñ%7Ô%7Ğ>�1—8’8‘:”:�:¸QĞPĞPĞPr r; r r r Úpos)r rŸ r< rK rM )r{ rY r| r} r r ) r r
r� r ÚindicesrŸ ro ÚelÚ vector_indexÚ pos_indexÚ
element_indexÚdiag_idxs `` r Úcheck_partial_win_to_indexz/TicTacToeGameManager.check_partial_win_to_index¼ sõ øø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×)Ò)¨%ğ+?ğ +?ğ +?ğ +?ğ +?ñ@ô @ˆğ QĞPÈĞPÑPÔPˆàˆØ ğ ,ğ ,ˆFݘ&¥#Ñ&Ô&ğ
Ø �ØØ ğ
,ğ
,�ؘ’7�7Ø#*§=¢=°Ñ#8Ô#8�LØ '§
¢
¨cÑ 2Ô 2�IàFRĞU^ÒF^ĞF^ L°IÀ±MÑ$BĞ$BĞdk×dqÒdqØ ñe+ô e+�Mğ Q^Ğ`dÔ`oÒPoĞPo˜ 4¤?°QÑ#6¸-Ñ"GÈĞ!Kğ Mğ Mà ! ]°T´_Ñ%DĞ!EğvGğ ğ '*�H˜U‘OØ Ø "˜UšU˜U˜UØ.;¸QĞ-?ÈĞ#LĞ#LĞ LĞ LĞ LĞ LĞ LØ "˜UšU˜U˜UØ./° Ğ-?ÈĞ#LĞ#LĞ LĞ LĞ LĞ LĞ LØ (˜[š[˜[˜[Ø#+˜O˜O˜O˜O˜OØ )š\˜\Ø#+˜O˜O˜O˜O˜Oøğ'
,ğ. dĞ+Ğ+Ğ+r rœ Úreturn_on_failc ó~ — |d dk s.|d | j k s |d dk s|d | j k r|S |S ©Nr r ©r )r r r§ s r Úreturn_index_if_validz*TicTacToeGameManager.return_index_if_validë sI € Ø �Œ8�aŠ<ˆ<˜5 œ8 t¤Ò6Ğ6¸%À¼(ÀQº,¸,È%ĞPQÌ(ĞVZÔVeÒJeĞJeØ!Ğ !؈ r c óD — t | j | j | j ¬¦ « S )N)r )r r r r ©r s r Úmake_fresh_instancez(TicTacToeGameManager.make_fresh_instanceğ s € İ# D¤O°T´]ÈtÌĞ_Ñ_Ô_Ğ_r rŸ c óà — t j |¦ « ¦ « dk r|S |xdk r) | |d |d |z
f|¬¦ « S xdk r) | |d |z
|d f|¬¦ « S xdk r, | |d |z
|d |z
f|¬¦ « S dk r+| |d |z
|d |z f|¬¦ « S d S )Nr r; r ©r§ r< rK rM )r# rq rx r« ©r r rŸ r� s r Ú get_previousz!TicTacToeGameManager.get_previousó s € İ
Œ8�E‰?Œ?× Ò Ñ Ô AÒ %Ğ %؈LØØ�’��Ø×1Ò1°5¸´8¸UÀ1¼Xȹ\Ğ2JĞ[`Ğ1ÑaÔaĞaØ�’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄĞ2JĞ[`Ğ1ÑaÔaĞaØ �’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeØ ’ � Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeğ � r c óğ — t j |¦ « ¦ « | j dz
k r|S |xdk r) | |d |d |z f|¬¦ « S xdk r) | |d |z |d f|¬¦ « S xdk r, | |d |z |d |z f|¬¦ « S dk r+| |d |z |d |z
f|¬¦ « S d S )Nr r; r r° r< rK rM )r# rq rx r r« r± s r Úget_nextz TicTacToeGameManager.get_next s € İ
Œ8�E‰?Œ?× Ò Ñ Ô D¤O°aÑ$7Ò 7Ğ 7؈LØØ�’��Ø×1Ò1°5¸´8¸UÀ1¼Xȹ\Ğ2JĞ[`Ğ1ÑaÔaĞaØ�’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄĞ2JĞ[`Ğ1ÑaÔaĞaØ �’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeØ ’ � Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeğ � r c ó^ — |€| ¦ « }t j |dk ¦ « S ©Nr )r{ r# rx r& s r r\ z"TicTacToeGameManager.is_board_full
s* € Ø ˆ=Ø—N’NÑ$Ô$ˆEİŒv�e˜q’jÑ!Ô!Ğ!r c ó4 — | j ¦ « S r )r Úcopyr s r r{ z TicTacToeGameManager.get_board s € ØŒz�ŠÑ Ô Ğ r c óf — | ¦ « | _ | j ¦ « S r )r r r¸ r s r ÚresetzTicTacToeGameManager.reset s' € Ø×*Ò*Ñ,Ô,ˆŒ
ØŒz�ŠÑ Ô Ğ r c ó — | j S r rª r s r Úget_board_sizez#TicTacToeGameManager.get_board_size s
€ ،Рr c óB — | j rdS | j d¦ « | ¦ « t | j ¦ « D ]–}t | j ¦ « D ]}| j | | | j k r#| |dz dz |dz dz ¦ « ŒA| j | | | j k r"|
|dz dz |dz dz ¦ « Œ€Œ—t j
¦ « t j ¦ « dS )NF)r r r r1 é2 T)r r ÚfillÚ _draw_boardrQ r r r
Ú _draw_circler Ú _draw_crossr2 ÚeventÚpumpr4 Úflip)r r; r< s r ÚrenderzTicTacToeGameManager.render s € Ø Œ=ğ Ø�5à Œ ×Ò˜Ñ#Ô#Ğ#Ø ×ÒÑÔĞݘœ Ñ)Ô)ğ Eğ EˆCİ ˜Tœ_Ñ-Ô-ğ
Eğ
E�Ø”:˜c”? 3Ô'¨4¬;Ò6Ğ6Ø×%Ò% c¨C¡i°"¡n°c¸C±iÀ"±nÑEÔEĞEĞEà”Z ”_ SÔ)¨TÔ->Ò>Ğ>Ø×$Ò$ S¨3¡Y°¡^°S¸3±YÀ±^ÑDÔDĞDøğ
Eõ Œ�
Š
‰Œˆİ
Œ
�ŠÑÔĞ؈tr c ól — | j rd S t j | j d||fdd¦ « d S )NÚgreené( r )r r2 ÚdrawÚcircler ©r r� Úys r rÁ z!TicTacToeGameManager._draw_circle. s9 € Ø Œ=ğ Ø ˆFİ
Œ�Š�t”{ G¨a°¨V°R¸Ñ;Ô;Ğ;Ğ;Ğ;r c óø — | j rd S t j | j d|dz
|dz
f|dz |dz fd¦ « t j | j d|dz |dz
f|dz
|dz fd¦ « d S )NÚredrÉ r )r r2 rÊ Úliner rÌ s r r z TicTacToeGameManager._draw_cross3 s‡ € Ø Œ=ğ Ø ˆFİ
Œ� Š �T”[ %¨!¨b©&°!°b±&Ğ)9¸AÀ¹FÀAÈÁFĞ;KÈQÑOÔOĞOİ
Œ� Š �T”[ %¨!¨b©&°!°b±&Ğ)9¸AÀ¹FÀAÈÁFĞ;KÈQÑOÔOĞOĞOĞOr c óî — t d| j dz d¦ « D ]Z}t d| j dz d¦ « D ]>}t j | j dt j ||dd¦ « d¦ « Œ?Œ[d S )Nr r1 )éÿ rÒ rÒ r )rQ r r2 rÊ Úrectr ÚRectrÌ s r rÀ z TicTacToeGameManager._draw_board9 sŒ € İ�q˜$œ/¨CÑ/°Ñ5Ô5ğ Wğ WˆAݘ1˜dœo°Ñ 3°SÑ9Ô9ğ
Wğ
W�İ”— ’ ˜Tœ[¨/½2¼7À1ÀaÈÈcÑ;RÔ;RĞTUÑVÔVĞVĞVğ
Wğ Wğ Wr c ó$ — | j | dk S r¶ r )r r s r Úis_emptyz TicTacToeGameManager.is_empty> s € ØŒz˜%Ô AÒ%Ğ%r Ú observationc óø — g }| | j | j ¦ « }t | j ¦ « D ]B}t | j ¦ « D ]+}|| | dk r| ||g¦ « Œ,ŒC|S )a
Legal moves are the empty spaces on the board.
:param observation: A 2D numpy array representing the current state of the game.
:param player: The player to check for. Since the game is symmetric, this is ignored.
:return: A list of legal moves.
r )rO r rQ rN )r r× r
Ú legal_movesr; r< s r r) z$TicTacToeGameManager.get_valid_movesA s� € ğ ˆ Ø!×)Ò)¨$¬/¸4¼?ÑKÔKˆ ݘœ Ñ)Ô)ğ 3ğ 3ˆCİ ˜Tœ_Ñ-Ô-ğ
3ğ
3�Ø ˜sÔ# CÔ(¨AÒ-Ğ-Ø ×&Ò&¨¨S zÑ2Ô2Ğ2øğ
3ğ Ğr c ó> — | j rdS t j ¦ « dS )NFT)r r2 Úquitr s r Ú pygame_quitz TicTacToeGameManager.pygame_quitP s! € Ø Œ=ğ Ø�5İ
Œ‰ Œ ˆ ؈tr c ó´ — | j rd S d„ t j ¦ « D ¦ « }t j ¦ « d r|S d S )Nc 3 ó K — | ] }|d z V — Œ
dS )r1 Nrœ r� s r ú <genexpr>z8TicTacToeGameManager.get_click_coords.<locals>.<genexpr>Y s& è è € Ğ:Ğ: !�Q˜#‘XĞ:Ğ:Ğ:Ğ:Ğ:Ğ:r r )r r2 ÚmouseÚget_posÚ get_pressed)r Ú mouse_poss r Úget_click_coordsz%TicTacToeGameManager.get_click_coordsV s] € Ø Œ=ğ Ø ˆFØ:Ğ:¥r¤x×'7Ò'7Ñ'9Ô'9Ğ:Ñ:Ô:ˆ İ
Œ8× Ò Ñ !Ô ! !Ô $ğ Ø Ğ ğ ğ r c ó — | j rd S | ¦ « | ¦ « �-| ¦ « \ }}|| | dk r||fS ŒV)NTr )r Úcheck_pg_eventsrä )r r r� rÍ s r Úget_human_inputz$TicTacToeGameManager.get_human_input] sp € Ø Œ=ğ Ø ˆFğ Ø × Ò Ñ "Ô "Ğ "Ø×$Ò$Ñ&Ô&Ğ2Ø×,Ò,Ñ.Ô.‘��1ؘ”8˜A”; !Ò#Ğ#Ø ˜a˜4�Kğ r c óÔ — | j rd S t j ¦ « D ]?}|j t j k r(| ¦ « t j d¦ « Œ@d S r¶ ) r r2 rà ÚgetÚtypeÚQUITrÜ ÚsysÚexit)r rà s r ræ z$TicTacToeGameManager.check_pg_eventsi sc € Ø Œ=ğ Ø ˆFİ”X—\’\‘^”^ğ ğ ˆEØŒz�RœWÒ$Ğ$Ø× Ò Ñ"Ô"Ğ"İ”˜‘ ” � øğ ğ r c ó@ — t j || j j ¦ « S )zÜ
Converts an integer move from the network to a board index.
:param move: An integer move selected from the network probabilities.
:return: A tuple representing the board index (int,int).
©r# Ú
unravel_indexr rR )r Úmoves r Únetwork_to_boardz%TicTacToeGameManager.network_to_boardq s € õ Ô d¤jÔ&6Ñ7Ô7Ğ7r c óŞ ‡ — ‰ j rd S t j d¬¦ « t | ¦ « � \ }}t |¦ « ˆ fd„|D ¦ « }t
j ||¬¦ « t j d¬¦ « t j d¦ « t j
d¦ « t j d ¦ « t j
¦ « }t j |d
d ¬ ¦ « | d
¦ « t! j |¦ « }t$ j ‰ j d¦ « }t! j d‰ j ¦ « |¦ « }|j |j z } t! j dt5 |j |j ¦ « | f¦ « }
|
|d¦ « |
|d
|j f¦ « |
|¦ « d S )N)é é
)Úfigsizec óª •— g | ]O}t j |‰j j ¦ « d › dt j |‰j j ¦ « d › �‘ŒPS )r ú;r rï )rC r� r s €r rE zKTicTacToeGameManager.save_screenshot_with_probabilities.<locals>.<listcomp> sf ø€ ğ ğ ğ Ğop•RÔ% a¨¬Ô)9Ñ:Ô:¸1Ô=ĞjĞjÅÔ@PĞQRĞTXÔT^ÔTdÑ@eÔ@eĞfgÔ@hĞjĞjğ ğ ğ r )r� rÍ éZ )ÚrotationÚMoveÚ ProbabilityzAction probabilitiesÚpngÚtight)ÚformatÚ bbox_inchesr ÚRGBAÚRGB)r r ) r ÚpltÚfigureÚzipÚitemsÚprintÚsnsÚbarplotÚxticksÚxlabelÚylabelÚtitleÚioÚBytesIOÚsavefigÚseekr Úopenr2 ÚimageÚtostringr Ú frombytesÚget_sizeÚheightÚnewÚmaxÚwidthÚpasteÚsave) r Ú action_probsÚpathÚlabelsÚ
probabilitiesÚbufÚplot_imgÚsurface_bufferÚ surface_imgÚ total_heightÚ combined_imgs ` r Ú"save_screenshot_with_probabilitiesz7TicTacToeGameManager.save_screenshot_with_probabilitiesy sÍ ø€ Ø Œ=ğ Ø ˆFİ Œ
˜8Ğ$Ñ$Ô$Ğ$İ # \×%7Ò%7Ñ%9Ô%9Ğ :Ñ ˆ�
İ
ˆmÑ Ô Ğ ğğ ğ ğ Øğñ ô ˆå Œ �f
Ğ.Ñ.Ô.Ğ.İ Œ
˜BĞ Ñ Ô Ğ İ Œ
�6ÑÔĞİ Œ
�=Ñ!Ô!Ğ!İ Œ Ğ(Ñ)Ô)Ğ)åŒj‰lŒlˆİ Œ �C °7Ğ;Ñ;Ô;Ğ;Ø �Š�‰ Œ ˆ İ”:˜c‘?”?ˆõ œ×*Ò*¨4¬;¸Ñ?Ô?ˆİ”o f¨d¬k×.BÒ.BÑ.DÔ.DÀnÑUÔUˆ 𠔨Ô);Ñ;ˆ İ ”y ¨X¬^¸[Ô=NÑ)OÔ)OĞQ]Ğ(^Ñ_Ô_ˆ Ø×Ò˜8 VÑ,Ô,Ğ,Ø×Ò˜;¨¨H¬OĞ(<Ñ=Ô=Ğ=Ø×Ò˜$Ñ Ô Ğ Ğ Ğ r c ó — | |z S r rœ )r r
s r Úget_canonical_formz'TicTacToeGameManager.get_canonical_form— s € à�v‰~Ğ r Úactionc óŒ — t |t ¦ « r| |¦ « }| ¦ « }|||< |S r )r| rš rò r¸ )r r r* r
Úboard_s r Úget_next_statez#TicTacToeGameManager.get_next_state› sB € İ �f�cÑ "Ô "ğ 3Ø×*Ò*¨6Ñ2Ô2ˆFØ—’‘ ” ˆØ ˆˆv‰Øˆ
r Úvalc ó — || _ d S r )r )r r. s r Ú set_headlessz!TicTacToeGameManager.set_headless¢ s
€ ؈Œ
ˆ
ˆ
r Ústatec ó: — t j |dk dd¦ « }|S r© )r# r� )r r1 r
Úmasks r Úget_invalid_actionsz(TicTacToeGameManager.get_invalid_actions¥ s € İŒx˜ š
A qÑ)Ô)ˆØˆ r c óz — t | j ¦ « dd¦ « dd¦ « S )NÚ1ÚXz-1ÚO)r} r Úreplacer s r Ú__str__z TicTacToeGameManager.__str__© s0 € İ�4”:‰Œ×&Ò& s¨CÑ0Ô0×8Ò8¸¸sÑCÔCĞCr r )T)rœ )r N)6Ú__name__Ú
__module__Ú __qualname__Ú__doc__rš Úboolr Útupler r r r# ÚndarrayÚlistr/ r ÚcallablerY rb rf rs ru r€ r] r— Údictr} r¦ r« r® r² r´ r\ r{ rº r¼ rÆ rÁ r rÀ rÖ r) rÜ rä rç ræ rò r' Ú staticmethodr) r- r0 r4 r: rœ r r r r s8 € € € € € ğğ ğ3ğ 3 3ğ 3°$ğ 3ÈDğ 3ğ 3ğ 3ğ 3ğ#˜3ğ # uğ #°ğ #ğ #ğ #ğ #ğ¨#°©*ğ ¸ğ ğ ğ ğ ğğ ğ ğ*°B´Jğ *ÈTğ *ğ *ğ *ğ *ğ ğ ğ ğ b¤jğ ¸ğ ÀTğ ğ ğ ğ ğB
ğ
¤
ğ
°tğ
ÀsÈTÁzğ
ğ
ğ
ğ
ğ °´
ğ À(ğ ğ ğ ğ ğ&!¨¬ğ !ğ !ğ !ğ !ğLğ L ğ L°Dğ Lğ Lğ Lğ Lğ ğ Sğ ¸ğ ğ ğ ğ ğ ğ ¨ğ °ğ ÀDğ ğ ğ ğ ğ,'ğ '°3ğ '¸3ğ 'Ètğ 'ğ 'ğ 'ğ 'ğ-,ğ -,°ğ -,¸ğ -,ÈTĞRWĞY\ĞR\ÔM]Ğ`dÑMdğ -,ğ -,ğ -,ğ -,ğ^ğ ¨5ğ À%ğ ĞQVğ ğ ğ ğ ğ
`ğ `ğ `ğ f %ğ f¨cğ f°cğ fğ fğ fğ fğ f˜eğ f¨#ğ f°#ğ fğ fğ fğ fğ"ğ "¨4ğ "ğ "ğ "ğ "ğ
!ğ !ğ !ğ!ğ !ğ !ğ ğ ğ ğ˜ğ ğ ğ ğ ğ$<ğ <ğ <ğ <ğ
Pğ Pğ Pğ Pğ Wğ Wğ Wğ
&˜eğ &¨ğ &ğ &ğ &ğ &ğ
ğ
¨2¬:ğ
¸s¸{Àdğ
ĞVZğ
ğ
ğ
ğ
ğ ˜Tğ ğ ğ ğ ğ ğ ğ ğ R¤Zğ ğ ğ ğ ğ ğ ğ ğ8ğ 8ğ 8ğ ğ ğ ğ< ğ ¨R¬Zğ ğ ğ ñ „\ğ ğ B¤J𠸸 ¸uğ Ècğ ĞVXÔV`ğ ğ ğ ğ ğ ğ ğ ğ ğ ğ¨¬ğ ¸Sğ ğ ğ ğ ğDğ Dğ Dğ Dğ Dr r )r r+ rì Úmatplotlib.pyplotÚpyplotr Únumpyr# Úpygamer2 Úseabornr ÚtorchrŒ ÚPILr Ú mu_alpha_zero.General.az_gamer r rœ r r ú<module>rN sÎ ğØ € € € Ø
€
€
€
Ø
€
€
€
à Ğ Ğ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à 7Ğ 7Ğ 7Ğ 7Ğ 7Ğ 7ğ[Dğ [Dğ [Dğ [Dğ [D˜=ñ [Dô [Dğ [Dğ [Dğ [Dr | 31,194 | Python | .py | 118 | 260.949153 | 2,695 | 0.344221 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,525 | asteroids.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/__pycache__/asteroids.cpython-311.pyc | §
Ô`�f( ã óJ — d dl Zd dlZd dlmZ d dlmZ G d„ de¦ « ZdS )é N)Úmake)Ú
MuZeroGamec ó6 — e Zd Zdej defd„Zd„ Zdej pej fd„Z
defd„Z defd„Z depd de
pd fd
„Zd „ Zdd
edepd dedej pej ee
ffd„Zdd
edepd dedej pej ee
ffd„Zd„ Zdej fd„Zdej defd„Zd S )Ú AsteroidsÚstateÚplayerc ó — |S ©N© ©Úselfr r s úK/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Game/asteroids.pyÚ get_state_for_passive_playerz&Asteroids.get_state_for_passive_player
s € ؈ ó c ó< — t d¦ « | _ d| _ d S )NzALE/Asteroids-v5F)r ÚenvÚdone©r
s r Ú__init__zAsteroids.__init__
s € İĞ*Ñ+Ô+ˆŒØˆŒ ˆ ˆ r Úreturnc ó> — | j ¦ « \ }}|S r
)r Úreset)r
ÚobsÚ_s r r zAsteroids.reset s € Ø”— ’ Ñ!Ô!‰ˆˆQ؈
r c ó — dS )Nr r r s r Úget_noopzAsteroids.get_noop s € ؈qr c ó$ — | j j j S r
)r Ú action_spaceÚnr s r Úget_num_actionszAsteroids.get_num_actions s € ØŒxÔ$Ô&Ğ&r Nc ó — | j S r
)r )r
r s r Ú game_resultzAsteroids.game_result s
€ ØŒyĞr c ó — t ¦ « S r
)r r s r Úmake_fresh_instancez Asteroids.make_fresh_instance s € ݉{Œ{Ğr é ÚactionÚ
frame_skipc óZ — | j |¦ « \ }}}}}|| _ |||fS r
)r Ústepr )r
r& r r' r Úrewr r s r Úget_next_statezAsteroids.get_next_state! s3 € à #œxŸ}š}¨VÑ 4Ô 4Ñ ˆˆS�$˜˜1؈Œ Ø�C˜ˆ~Ğ r c ó� — | ||¦ « \ }}}t |dz
¦ « D ] }| ||¦ « \ }}}Œ |||fS )Né )r+ Úrange)r
r& r r' r r* r Úis r Úframe_skip_stepzAsteroids.frame_skip_step' sc € à ×,Ò,¨V°VÑ<Ô<‰ˆˆS�$İ�z A‘~Ñ&Ô&ğ Ağ AˆAØ !× 0Ò 0°¸Ñ @Ô @‰NˆC��d�dØ�C˜ˆ~Ğ r c ó — d S r
r r s r ÚrenderzAsteroids.render. s € Ø ˆr c ó> — | j j ¦ « S r
)r r Úsample)r
r Úkwargss r Úget_random_valid_actionz!Asteroids.get_random_valid_action1 s € ØŒxÔ$×+Ò+Ñ-Ô-Ğ-r c óN — t j | ¦ « ¦ « S r
)ÚnpÚonesr r s r Úget_invalid_actionsz Asteroids.get_invalid_actions4 s € İŒw˜×,Ò,Ñ.Ô.Ñ0Ô0Ğ0r )r% )Ú__name__Ú
__module__Ú __qualname__r8 ÚndarrayÚintr r ÚthÚTensorr r r Úboolr" r$ r+ r0 r2 r6 r: r r r r r sÄ € € € € € ğ°"´*ğ Àcğ ğ ğ ğ ğğ ğ ğ�r”zĞ. R¤Yğ ğ ğ ğ ğ˜#ğ ğ ğ ğ ğ' ğ 'ğ 'ğ 'ğ 'ğ # +¨ğ °$°,¸$ğ ğ ğ ğ ğğ ğ ğ ğ Sğ °#°+¸ğ È3ğ Ø ŒJĞ #˜"œ) S¨$ğX0ğ ğ ğ ğ ğ ğ cğ °3°;¸$ğ ÈCğ Ø ŒJĞ #˜"œ) S¨$ğY0ğ ğ ğ ğ ğ
ğ
ğ
ğ.¨R¬Zğ .ğ .ğ .ğ .ğ1¨¬ğ 1¸Sğ 1ğ 1ğ 1ğ 1ğ 1ğ 1r r ) Únumpyr8 Útorchr@ Ú gymnasiumr Ú mu_alpha_zero.General.mz_gamer r r r r ú<module>rG su ğØ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4ğ-1ğ -1ğ -1ğ -1ğ -1�
ñ -1ô -1ğ -1ğ -1ğ -1r | 3,983 | Python | .py | 20 | 198.1 | 830 | 0.346367 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,526 | muzero.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/muzero.py | import os
import sys
import numpy as np
from mu_alpha_zero.General.arena import GeneralArena
sys.path.append(os.path.abspath("."))
from typing import Type
import torch as th
from mu_alpha_zero.AlphaZero.Arena.players import NetPlayer
from mu_alpha_zero.trainer import Trainer
from mu_alpha_zero.General.utils import find_project_root, net_not_none
from mu_alpha_zero.General.mz_game import MuZeroGame
from mu_alpha_zero.General.memory import GeneralMemoryBuffer
from mu_alpha_zero.General.network import GeneralNetwork
from mu_alpha_zero.MuZero.MZ_Arena.arena import MzArena
from mu_alpha_zero.MuZero.MZ_MCTS.mz_search_tree import MuZeroSearchTree
from mu_alpha_zero.config import MuZeroConfig
from mu_alpha_zero.Hooks.hook_manager import HookManager
class MuZero:
"""
Class for managing the training and creation of a MuZero model.
Attributes:
game_manager (MuZeroGame): The game manager instance.
net (GeneralNetwork): The neural network used by MuZero for prediction.
trainer (Trainer): The trainer object responsible for training the model.
device (torch.device): The device (CPU or CUDA) used for computations.
Methods:
__init__(self, game_manager: MuZeroGame)
Initializes a new instance of the MuZero class.
create_new(self, args: dict, network_class: Type[GeneralNetwork], headless: bool = True,
checkpointer_verbose: bool = False)
Creates a new MuZero model using the specified arguments.
train(self)
Trains the MuZero model.
"""
def __init__(self, game_manager: MuZeroGame):
self.game_manager = game_manager
self.net = None
self.trainer = None
self.device = th.device("cuda" if th.cuda.is_available() else "cpu")
self.tree = None
self.muzero_config = None
def create_new(self, muzero_config: MuZeroConfig, network_class: Type[GeneralNetwork], memory: GeneralMemoryBuffer,
hook_manager: HookManager or None = None,
headless: bool = True,
arena_override: GeneralArena or None = None,
checkpointer_verbose: bool = False):
muzero_config.net_action_size = int(self.game_manager.get_num_actions())
if not os.path.isabs(muzero_config.pickle_dir):
muzero_config.pickle_dir = find_project_root() + "/" + muzero_config.pickle_dir
self.muzero_config = muzero_config
network = network_class.make_from_config(muzero_config, hook_manager=hook_manager).to(
self.device)
self.tree = MuZeroSearchTree(self.game_manager.make_fresh_instance(), muzero_config)
net_player = NetPlayer(self.game_manager.make_fresh_instance(),
**{"network": network, "monte_carlo_tree_search": self.tree})
arena = MzArena(self.game_manager.make_fresh_instance(), self.muzero_config,
self.device) if arena_override is None else arena_override
self.trainer = Trainer.create(self.muzero_config, self.game_manager.make_fresh_instance(), network, self.tree,
net_player,
headless=headless,
checkpointer_verbose=checkpointer_verbose, arena_override=arena,
hook_manager=hook_manager,
memory_override=memory)
self.net = self.trainer.get_network()
def from_checkpoint(self, network_class: Type[GeneralNetwork], memory: GeneralMemoryBuffer, path: str,
checkpoint_dir: str,
headless: bool = True,
hook_manager: HookManager or None = None,
logdir_override: str = None,
checkpointer_verbose: bool = False):
self.trainer = Trainer.from_checkpoint(network_class, MuZeroSearchTree, NetPlayer, path,
checkpoint_dir,
self.game_manager,
headless=headless,
hook_manager=hook_manager,
checkpointer_verbose=checkpointer_verbose,
mem=memory, log_dir_override=logdir_override)
self.net = self.trainer.get_network()
self.tree = self.trainer.get_tree()
self.args = self.trainer.get_args()
def train(self):
net_not_none(self.net)
self.trainer.train()
def train_parallel(self, use_reanalyze: bool):
net_not_none(self.net)
self.trainer.train_parallel(use_reanalyze=use_reanalyze, use_pitting=False)
def predict(self, x: np.ndarray, tau: float = 0) -> int:
net_not_none(self.net)
assert x.shape == self.muzero_config.target_resolution + (
3,), "Input shape must match target resolution with 3 channels. Got: " + str(x.shape)
self.net.eval()
pi, (v, _) = self.tree.search(self.net, x, None, self.device, tau=tau)
move = self.game_manager.select_move(pi, tau=self.muzero_config.tau)
return move
| 5,294 | Python | .py | 94 | 42.946809 | 119 | 0.620876 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,527 | utils.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/utils.py | import json
import math
from typing import Callable
import numpy as np
import optuna
import torch as th
from PIL import Image
from mu_alpha_zero.config import MuZeroConfig
def add_actions_to_obs(observations: th.Tensor, actions: th.Tensor, dim=0):
return th.cat((observations, actions), dim=dim)
def match_action_with_obs(observations: th.Tensor, action: int, config: MuZeroConfig):
if config.is_atari:
tensor_action = th.zeros((config.net_action_size,), dtype=th.float32, device=observations.device).scatter(0,
th.tensor(
action),
1)
tensor_action = tensor_action.expand((observations.shape[1], observations.shape[2]))
else:
if config.actions_are == "columns":
tensor_action = th.full((1, observations.size(1), observations.size(2)),
scale_action(action, config.net_action_size),
device=observations.device)
elif config.actions_are == "rows":
tensor_action = th.zeros((observations.shape[1],), device=observations.device).scatter(0, th.tensor(action,
device=observations.device),
1).unsqueeze(0)
tensor_action = tensor_action.expand((1, observations.shape[1], observations.shape[2]))
elif config.actions_are == "board":
# unravel to 2d
action = [action % observations.shape[1], action % observations.shape[2]]
tensor_action = th.zeros((1, observations.shape[1], observations.shape[2]), device=observations.device)[
action[0], action[1]] = 1
else:
raise ValueError("Invalid config.actions_are value.")
return add_actions_to_obs(observations, tensor_action)
def match_action_with_obs_batch(observation_batch: th.Tensor, action_batch: list[int], config: MuZeroConfig):
tensors = [match_action_with_obs(observation_batch[index], action_batch[index], config).unsqueeze(0) for index in
range(len(action_batch))]
return th.cat(tensors, dim=0)
def resize_obs(observations: np.ndarray, size: tuple[int, int], resize: bool) -> np.ndarray:
if not resize:
return observations
obs = Image.fromarray(observations)
obs = obs.resize(size)
return np.array(obs)
def scale_state(state: np.ndarray, scale: bool):
if not scale:
return state
# scales the given state to be between 0 and 1
return state / 255
def scale_action(action: int, num_actions: int):
# return action / (num_actions - 1)
return (action + 1) / num_actions
def scale_hidden_state(hidden_state: th.Tensor):
was_reshaped = False
if len(hidden_state.shape) == 3:
hidden_state = hidden_state.unsqueeze(0)
was_reshaped = True
max_ = hidden_state.view(hidden_state.size(0), hidden_state.size(1), -1).max(dim=2, keepdim=True)[0].unsqueeze(-1)
min_ = hidden_state.view(hidden_state.size(0), hidden_state.size(1), -1).min(dim=2, keepdim=True)[0].unsqueeze(-1)
max_min_dif = max_ - min_
max_min_dif[max_min_dif < 1e-5] += 1e-5
hidden_state = (hidden_state - min_) / max_min_dif
return hidden_state.squeeze(0) if was_reshaped else hidden_state
def scale_reward_value(value: th.Tensor, e: float = 0.001):
return th.sign(value) * (th.sqrt(th.abs(value) + 1) - 1) + e * value
def invert_scale_reward_value(value: th.Tensor, e: float = 0.001):
return th.sign(value) * (
((th.sqrt(1 + 4 * 0.001 * (th.abs(value) + 1 + 0.001)) - 1) / (2 * 0.001))
** 2
- 1
)
def scale_reward(reward: float):
return math.log(reward + 1, 5)
def muzero_loss(y_hat, y):
return -th.sum(y * y_hat, dim=1).unsqueeze(1)
def scalar_to_support(x: th.Tensor, support_size: int, is_atari: bool):
# x is of shape: [batch_size,1] at unroll step t
if is_atari:
x = scale_reward_value(x)
x = th.clamp(x, -support_size, support_size)
lower_p = 1 - (x - x.floor())
upper_p = x - x.floor()
support = th.zeros((x.size(0), 2 * support_size + 1), device=x.device)
support.scatter_(1, (support_size + x.floor()).type(th.int64), lower_p)
upper_index = (support_size + x.floor() + 1)
upper_index[upper_index >= 2 * support_size] = 0
support.scatter_(1, (
upper_index).type(th.int64), upper_p)
return support
def support_to_scalar(x: th.Tensor, support_size: int, is_atari: bool):
support = th.arange(-support_size, support_size + 1, 1, dtype=x.dtype, device=x.device).unsqueeze(0)
output = th.sum(x * support, dim=1)
if is_atari:
output = invert_scale_reward_value(output)
return output.unsqueeze(1)
def mz_optuna_parameter_search(n_trials: int, storage: str or None, study_name: str, game,
muzero_config: MuZeroConfig, in_memory: bool = False, direction: str = "maximize",
arena_override=None, memory_override=None):
def objective(trial: optuna.Trial):
muzero_config.num_simulations = trial.suggest_int("num_mc_simulations", 60, 800)
muzero_config.lr = trial.suggest_float("lr", 1e-5, 5e-2, log=True)
muzero_config.tau = trial.suggest_float("temp", 0.5, 2)
muzero_config.arena_tau = trial.suggest_float("arena_temp", 0, 2)
muzero_config.c = trial.suggest_float("cpuct", 0.7, 2)
muzero_config.l2 = trial.suggest_float("l2_norm", 1e-6, 6e-2)
# muzero_config.frame_buffer_size = trial.suggest_int("frame_buffer_size", 10, 40)
muzero_config.alpha = trial.suggest_float("alpha", 0.4, 2)
muzero_config.beta = trial.suggest_float("beta", 0.1, 1)
muzero_config.balance_term = trial.suggest_categorical("loss_scale", [1, 0.5])
muzero_config.num_blocks = trial.suggest_int("num_blocks", 5, 32)
muzero_config.K = trial.suggest_int("k", 2, 25)
muzero_config.epochs = trial.suggest_int("epochs", 200, 500)
muzero_config.self_play_games = 300
muzero_config.num_iters = 1
device = th.device("cuda" if th.cuda.is_available() else "cpu")
trial.net_action_size = int(game.get_num_actions())
network = MuZeroNet.make_from_config(muzero_config).to(device)
tree = MuZeroSearchTree(game.make_fresh_instance(), muzero_config)
net_player = NetPlayer(game.make_fresh_instance(), **{"network": network, "monte_carlo_tree_search": tree})
if arena_override is None:
arena = MzArena(game.make_fresh_instance(), muzero_config, device)
else:
arena = arena_override
if memory_override is None:
mem = MemBuffer(muzero_config.max_buffer_size, disk=True, full_disk=False,
dir_path=muzero_config.pickle_dir)
else:
mem = memory_override
trainer = Trainer.create(muzero_config, game.make_fresh_instance(), network, tree, net_player, headless=True,
arena_override=arena, checkpointer_verbose=False, memory_override=mem)
trainer.train()
mean = trainer.get_arena_win_frequencies_mean()
trial.report(mean, muzero_config.num_iters)
print(f"Trial {trial.number} finished with win freq {mean}.")
del trainer
del network
del tree
del net_player
return mean
from mu_alpha_zero.MuZero.MZ_Arena.arena import MzArena
from mu_alpha_zero.MuZero.MZ_MCTS.mz_search_tree import MuZeroSearchTree
from mu_alpha_zero.MuZero.Network.networks import MuZeroNet
from mu_alpha_zero.trainer import Trainer
from mu_alpha_zero.AlphaZero.Arena.players import NetPlayer
from mu_alpha_zero.mem_buffer import MemBuffer
muzero_config.show_tqdm = False
if in_memory:
study = optuna.create_study(study_name=study_name, direction=direction)
else:
if storage is None:
raise ValueError("Storage can't be None if in_memory is False.")
study = optuna.load_study(study_name=study_name, storage=storage)
study.optimize(objective, n_trials=n_trials)
with open(f"{muzero_config.checkpoint_dir}/study_params.json", "w") as file:
json.dump(study.best_params, file)
def mask_invalid_actions(invalid_actions: np.ndarray, pi: np.ndarray):
if np.sum(invalid_actions) == 0:
print("No valid actions left.")
pi = pi.reshape(-1) * invalid_actions.reshape(-1)
return pi / pi.sum()
def mask_invalid_actions_batch(get_invalid_actions: Callable, pis: th.Tensor, players: list[int]):
invalid_actions_ts = th.empty(pis.shape)
for i, player in enumerate(players):
invaid_actions = get_invalid_actions(pis[i], player)
invalid_actions_ts[i] = invaid_actions
return invalid_actions_ts
def get_opponent_action(game_state: np.ndarray, last_move: int, last_player: int, opponent: str, tree, net) -> tuple:
if opponent == "muzero":
res = tree.search(net, game_state, -last_player, th.device("cuda" if th.cuda.is_available() else "cpu"))
return res[0], res[1][0]
elif opponent == "minimax":
return get_minimax_action_connect4(game_state[:,:,0], last_move, last_player)
def get_minimax_action_connect4(game_state: np.ndarray, last_move: int, last_player: int) -> tuple[dict, None]:
from mu_alpha_zero.MuZero.MinimaxOpponent.board import Board
from mu_alpha_zero.MuZero.MinimaxOpponent.player import PlayerMM
board = Board.from_game_state(game_state, (last_player, last_move))
depth = 5
player = PlayerMM(depth, False)
move = player.findMove(board)
pi = [0.0] * game_state.shape[1]
pi[move] = 1.0
return {idx: el for idx, el in enumerate(pi)}, None
| 10,181 | Python | .py | 183 | 45.071038 | 140 | 0.624108 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,528 | lazy_arrays.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/lazy_arrays.py | import gzip
import os
import uuid
import numpy as np
class LazyArray:
def __init__(self, array: np.ndarray, directory_path: str):
self.directory_path = directory_path
self.path = f"{directory_path}/array_{uuid.uuid4()}.npy.gz"
self.persistent = False
self.save_array(array)
def load_array(self):
f = gzip.GzipFile(self.path, 'r')
arr = np.load(f)
f.close()
return arr
def save_array(self, array: np.ndarray):
file = gzip.GzipFile(self.path, 'w')
np.save(file, array)
file.close()
def __array__(self):
return self.load_array()
def remove_array(self):
os.remove(self.path)
def make_persistent(self):
self.persistent = True
| 767 | Python | .py | 25 | 23.84 | 67 | 0.617486 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,529 | utils.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/__pycache__/utils.cpython-311.pyc | §
‰Ó¢f� ã
ó$ — d dl Z d dlZd dlmZ d dlZd dlZd dlZd dl m
Z
d dl m Z d3dej
dej
fd„Zdej
defd „Zd
ej
d ee fd „Zdej d
eeef dedej fd„Zdej defd„Zdedefd„Zdej
fd„Zd4dej
defd„Zd4dej
defd „Z d efd „Z d ej
d efd!„Z d ej
d efd"„Z d5d%ed&e pdd'e d(e d)ed*e f d+„Z!d,ej d-ej fd.„Z"d/ed0ej
d1ee fd2„Z#dS )6é N)ÚCallable)ÚImage)Ú MuZeroConfigÚ observationsÚactionsc ó2 — t j | |f|¬¦ « S )N©Údim)ÚthÚcat)r r r
s úI/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/utils.pyÚadd_actions_to_obsr
s € İ
Œ6�< Ğ)¨sĞ 3Ñ 3Ô 3Ğ3ó Úactionc ó¢ — t j d| j d | j d f|t j | j ¬¦ « }t | |¦ « S )Né é ©ÚdtypeÚdevice)r ÚfullÚshapeÚfloat32r r )r r s r
Úmatch_action_with_obsr sO € İ
ŒW�a˜ Ô+¨AÔ.° Ô0BÀ1Ô0EĞFÈÕVXÔV`Ø (Ô /ğ1ñ 1ô 1€Få ˜l¨FÑ 3Ô 3Ğ3r Úobservation_batchÚ action_batchc óp ‡ — ˆ fd„|D ¦ « }t j |d¬¦ « }t ‰ |d¬¦ « S )Nc ó’ •— g | ]C}t j d d ‰j d ‰j d f|t j ‰j ¬¦ « ‘ŒDS )r r é r )r r r r r )Ú.0r r s €r
ú
<listcomp>z/match_action_with_obs_batch.<locals>.<listcomp> sm ø€ ğ Tğ Tğ TØ<Bõ Œw˜˜1Ğ /Ô 5°aÔ 8Ğ:KÔ:QĞRSÔ:TĞUĞW]ÕegÔeoØ /Ô 6ğ8ñ 8ô 8ğ Tğ Tğ Tr r r r )r r r )r r Útensorsr s ` r
Úmatch_action_with_obs_batchr# s\ ø€ ğTğ Tğ Tğ TØFRğTñ Tô T€GåŒf�W !Ğ$Ñ$Ô$€Gİ Ğ /°¸aĞ @Ñ @Ô @Ğ@r ÚsizeÚresizeÚreturnc ó„ — |s| S t j | ¦ « }| |¦ « }t j |¦ « S ©N)r Ú fromarrayr% ÚnpÚarray)r r$ r% Úobss r
Ú
resize_obsr- s> € Ø ğ ØĞİ
Œ/˜,Ñ
'Ô
'€CØ
�*Š*�TÑ
Ô
€Cİ
Œ8�C‰=Œ=Ğr ÚstateÚscalec ó — |s| S | dz S )Néÿ © )r. r/ s r
Ú scale_stater3 & s € Ø ğ ؈ à �3‰;Ğr Ú num_actionsc ó — | dz |z S ©Nr r2 )r r4 s r
Ú scale_actionr7 - s € à �Q‰J˜+Ñ %Ğ%r Ú hidden_statec ón — d}t | j ¦ « dk r| d¦ « } d}| | d¦ « | d¦ « d¦ « dd¬¦ « d d¦ « }| | d¦ « | d¦ « d¦ « dd¬¦ « d d¦ « }||z
}d ||dk < | |z
|z } |r| d¦ « n| S )
NFr r Tr éÿÿÿÿr )r
Úkeepdimçñh㈵øä>)Úlenr Ú unsqueezeÚviewr$ ÚmaxÚminÚsqueeze)r8 Ú was_reshapedÚmax_Úmin_Ú max_min_difs r
Úscale_hidden_staterG 2 s4 € Ø€Lİ
ˆ<Ô Ñ Ô !Ò#Ğ#Ø#×-Ò-¨aÑ0Ô0ˆ ؈ Ø × Ò ˜\× .Ò .¨qÑ 1Ô 1°,×2CÒ2CÀAÑ2FÔ2FÀrÑ JÔ J× NÒ NĞSTĞ]aĞ NÑ bÔ bĞcdÔ e× oÒ oĞprÑ sÔ s€DØ × Ò ˜\× .Ò .¨qÑ 1Ô 1°,×2CÒ2CÀAÑ2FÔ2FÀrÑ JÔ J× NÒ NĞSTĞ]aĞ NÑ bÔ bĞcdÔ e× oÒ oĞprÑ sÔ s€Dؘ‘+€KØ$(€K� ˜qÒ Ñ!Ø 4Ñ'¨;Ñ6€LØ&2Ğ Dˆ<× Ò Ñ "Ô "Ğ "¸ ĞDr çü©ñÒMbP?ÚvalueÚec ó� — t j | ¦ « t j t j | ¦ « dz ¦ « dz
z || z z S r6 ©r ÚsignÚsqrtÚabs©rI rJ s r
Úscale_reward_valuerQ ? s: € İ
Œ7�5‰>Œ>�RœW¥R¤V¨E¡]¤]°QÑ%6Ñ 7Ô 7¸!Ñ ;Ñ <¸qÀ5¹yÑ HĞHr c ó¨ — t j | ¦ « t j ddt j | ¦ « dz dz z z ¦ « dz
dz dz dz
z S )Nr gü©ñÒMbp?rH gü©ñÒMb`?r rL rP s r
Úinvert_scale_reward_valuerS C sZ € İ
Œ7�5‰>Œ>İŒg�a˜)¥r¤v¨e¡}¤}°qÑ'8¸5Ñ'@ÑAÑAÑBÔBÀQÑFÈ9Ñ
UØñ
àñ
ñ ğ r Úrewardc ó2 — t j | dz d¦ « S )Nr é )ÚmathÚlog)rT s r
Ú scale_rewardrY L s € İ Œ8�F˜Q‘J Ñ "Ô "Ğ"r ÚxÚ support_sizec ón — t | ¦ « } t j | | |¦ « } d| | ¦ « z
z
}| | ¦ « z
}t j | d¦ « d|z dz f| j ¬¦ « }| d|| ¦ « z t j ¦ « |¦ « | d|| ¦ « z dz t j ¦ « |¦ « n# t $ r Y nw xY w|S )Nr r r )r ) rQ r ÚclampÚfloorÚzerosr$ r Úscatter_ÚtypeÚint64Ú RuntimeError)rZ r[ Úlower_pÚupper_pÚsupports r
Úscalar_to_supportrg P s! € å˜1Ñ Ô €Aİ
Œ��\�M <Ñ0Ô0€AØ�1�q—w’w‘y”y‘=Ñ!€GØ�!—'’'‘)”)‰m€GİŒh˜Ÿš˜q™ œ 1 |Ñ#3°aÑ#7Ğ8ÀÄĞJÑJÔJ€GØ ×Ò�Q˜ ¨¯ª© ¬ Ñ1×7Ò7½¼ÑAÔAÀ7ÑKÔKĞKğ
Ø×Ò˜Ø ˜qŸwšw™yœyÑ(¨1Ñ,¯dªdµ2´8©n¬n¸gñ Gô Gğ Gğ Gøå ğ
ğ
ğ
à ˆğ
øøøğ €Ns ÃA D% Ä%
D2Ä1D2c óğ — t j | |dz d| j | j ¬¦ « d¦ « }t j | |z d¬¦ « }t
|¦ « }| d¦ « S )Nr r r r )r Úaranger r r> ÚsumrS )rZ r[ rf Úoutputs r
Úsupport_to_scalarrl a sq € İŒi˜ ˜
|°aÑ'7¸À!Ä'ĞRSÔRZĞ[Ñ[Ô[×eÒeĞfgÑhÔh€Gİ
ŒV�A˜‘K QĞ
'Ñ
'Ô
'€Fİ
& vÑ
.Ô
.€FØ × Ò ˜AÑ Ô Ğ r FÚmaximizeÚn_trialsÚstorageÚ
study_nameÚ
muzero_configÚ in_memoryÚ directionc óê ‡‡‡‡‡ ‡
‡‡‡‡— dt j fˆ ˆ
ˆˆˆˆˆˆˆˆf
d„ } ddlmŠ ddlmŠ ddlmŠ
ddlm Š ddl
m Š dd l m
Š d
‰_ |rt j ||¬ ¦ « }
n'|€t! d ¦ « ‚t j ||¬
¦ « }
|
| | ¬¦ « t' ‰j › d�d¦ « 5 } t+ j |
j | ¦ « d d d ¦ « d S # 1 swxY w Y d S )NÚtrialc óî •
— | ddd¦ « ‰_ | dddd¬¦ « ‰_ | d d
d ¦ « ‰_ | d d
d ¦ « ‰_ | ddd ¦ « ‰_ | ddd¦ « ‰_ | ddd ¦ « ‰_ | ddd¦ « ‰_ |
ddd
g¦ « ‰_ | ddd¦ « ‰_ | d d d ¦ « ‰_
| d d d ¦ « ‰_ d!‰_ d‰_ t# j t" j ¦ « rd"nd#¦ « }t+ ‰ ¦ « ¦ « | _ ‰
‰¦ « |¦ « } ‰ ‰ ¦ « ‰¦ « } ‰
‰ ¦ « fi ||d$œ¤�}‰€ ‰ ‰ ¦ « ‰|¦ « }n‰}‰€ ‰ ‰j dd%‰j ¬&¦ « }n‰}‰ ‰‰ ¦ « |||d|d%|¬'¦ « }| ¦ « | ¦ « }| |‰j ¦ « tC d(| j" › d)|› d*�¦ « ~~~~|S )+NÚnum_mc_simulationsé< i Úlrr< gš™™™™™©?T)rX Útempg à?r Ú
arena_tempr Úcpuctgffffffæ?Úl2_normg�íµ ÷Æ°>g¸ …ëQ¸®?Úalphagš™™™™™Ù?Úbetagš™™™™™¹?r Ú
loss_scaleÚ
num_blocksrV é Úké ÚepochséÈ iô i, ÚcudaÚcpu)ÚnetworkÚmonte_carlo_tree_searchF)ÚdiskÚ full_diskÚdir_path)ÚheadlessÚarena_overrideÚcheckpointer_verboseÚmemory_overridezTrial z finished with win freq ú.)#Ú suggest_intÚnum_simulationsÚ
suggest_floatry ÚtauÚ arena_tauÚcÚl2r~ r Úsuggest_categoricalÚ balance_termr� ÚKr… Úself_play_gamesÚ num_itersr r r‡ Ú is_availableÚintÚget_num_actionsÚnet_action_sizeÚmake_from_configÚtoÚmake_fresh_instanceÚmax_buffer_sizeÚ
pickle_dirÚcreateÚtrainÚ get_arena_win_frequencies_meanÚreportÚprintÚnumber)ru r r‰ ÚtreeÚ
net_playerÚarenaÚmemÚtrainerÚmeanÚ MemBufferÚ MuZeroNetÚMuZeroSearchTreeÚMzArenaÚ NetPlayerÚTrainerr� Úgamer‘ rq s €€€€€€€€€€r
Ú objectivez-mz_optuna_parameter_search.<locals>.objectivek s ø€ Ø(-×(9Ò(9Ğ:NĞPRĞTWÑ(XÔ(Xˆ
Ô%Ø ×.Ò.¨t°T¸4ÀTĞ.ÑJÔJˆ
ÔØ !× /Ò /°¸¸QÑ ?Ô ?ˆ
ÔØ"'×"5Ò"5°lÀAÀqÑ"IÔ"Iˆ
Ô Ø ×-Ò-¨g°s¸AÑ>Ô>ˆ
ŒØ ×.Ò.¨y¸$ÀÑEÔEˆ
Ôà #× 1Ò 1°'¸3ÀÑ BÔ Bˆ
ÔØ "× 0Ò 0°¸¸aÑ @Ô @ˆ
ÔØ%*×%>Ò%>¸|ÈaĞQTÈXÑ%VÔ%Vˆ
Ô"Ø#(×#4Ò#4°\À1ÀbÑ#IÔ#Iˆ
Ô Ø ×+Ò+¨C°°BÑ7Ô7ˆ
ŒØ $× 0Ò 0°¸3ÀÑ DÔ Dˆ
Ô à(+ˆ
Ô%Ø"#ˆ
Ô å”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ # D×$8Ò$8Ñ$:Ô$:Ñ ;Ô ;ˆÔ Ø×,Ò,¨]Ñ;Ô;×>Ò>¸vÑFÔFˆØ Ğ × 8Ò 8Ñ :Ô :¸MÑJÔJˆØ �Y˜t× 7Ò 7Ñ 9Ô 9ĞsĞsÈĞmqĞ=rĞ=rĞsĞsˆ
Ø Ğ !Ø�G˜D× 4Ò 4Ñ 6Ô 6¸
ÀvÑNÔNˆEˆEà"ˆEØ Ğ "Ø�)˜MÔ 9ÀĞPUØ%2Ô%=ğ?ñ ?ô ?ˆCˆCğ "ˆCØ—.’. °×0HÒ0HÑ0JÔ0JÈGĞUYĞ[eĞptØ05ÈEĞcfğ !ñ hô hˆà�
Š
‰ŒˆØ×5Ò5Ñ7Ô7ˆØ
� Š �T˜=Ô2Ñ3Ô3Ğ3İ
ĞD�u”|ĞDĞD¸TĞDĞDĞDÑEÔEĞEØ Ø Ø Ø Øˆ r r )r· )r¶ )rµ )r¹ )r¸ )r´ F)rp rs z,Storage can't be None if in_memory is False.)rp ro )rn z/study_params.jsonÚw)ÚoptunaÚTrialÚ#mu_alpha_zero.MuZero.MZ_Arena.arenar· Ú+mu_alpha_zero.MuZero.MZ_MCTS.mz_search_treer¶ Ú%mu_alpha_zero.MuZero.Network.networksrµ Úmu_alpha_zero.trainerr¹ Ú%mu_alpha_zero.AlphaZero.Arena.playersr¸ Úmu_alpha_zero.mem_bufferr´ Ú show_tqdmÚ create_studyÚ
ValueErrorÚ
load_studyÚoptimizeÚopenÚcheckpoint_dirÚjsonÚdumpÚ best_params)rn ro rp rº rq rr rs r� r‘ r» ÚstudyÚfiler´ rµ r¶ r· r¸ r¹ s `` `` @@@@@@r
Úmz_optuna_parameter_searchrÑ h sÈ øøøøøøøøøø€ ğ*�œ ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğX <Ğ;Ğ;Ğ;Ğ;Ğ;ØLĞLĞLĞLĞLĞLØ?Ğ?Ğ?Ğ?Ğ?Ğ?Ø-Ğ-Ğ-Ğ-Ğ-Ğ-Ø?Ğ?Ğ?Ğ?Ğ?Ğ?Ø2Ğ2Ğ2Ğ2Ğ2Ğ2à #€MÔØğ JİÔ#¨zÀYĞOÑOÔOˆˆà ˆ?İ Ğ KÑLÔLĞ LİÔ!¨ZÀĞIÑIÔIˆØ ‡N‚N�9 x€NÑ0Ô0Ğ0İ
� Ô-ĞAĞAĞAÀ3Ñ GÔ Gğ +È4İ Œ �%Ô# TÑ*Ô*Ğ*ğ+ğ +ğ +ñ +ô +ğ +ğ +ğ +ğ +ğ +ğ +ğ +øøøğ +ğ +ğ +ğ +ğ +ğ +s à C(Ã(C,Ã/C,Úinvalid_actionsÚpic óÔ — t j | ¦ « dk rt d¦ « | d¦ « | d¦ « z }|| ¦ « z S )Nr zNo valid actions left.r: )r* rj r¬ Úreshape)rÒ rÓ s r
Úmask_invalid_actionsrÖ ª sZ € İ „vˆoÑ Ô !Ò#Ğ#İ
Ğ&Ñ'Ô'Ğ'Ø �Š�B‰ Œ ˜/×1Ò1°"Ñ5Ô5Ñ 5€BØ
�—’‘”‰=Ğr Úget_invalid_actionsÚpisÚplayersc ó� — t j |j ¦ « }t |¦ « D ] \ }} | || |¦ « }|||< Œ |S r( )r Úemptyr Ú enumerate)r× rØ rÙ Úinvalid_actions_tsÚiÚplayerÚinvaid_actionss r
Úmask_invalid_actions_batchrá ± sY € İœ #¤)Ñ,Ô,Ğİ ˜wÑ'Ô'ğ /ğ /‰ ˆˆ6Ø,Ğ,¨S°¬V°VÑ<Ô<ˆØ .И1Ñ Ğ Ø Ğ r )r )rH )Frm NN)$rÌ rW Útypingr Únumpyr* r½ Útorchr ÚPILr Úmu_alpha_zero.configr ÚTensorr r r Úlistr# ÚndarrayÚtupleÚboolr- r3 r7 rG ÚfloatrQ rS rY rg rl ÚstrrÑ rÖ rá r2 r r
ú<module>rî s ğØ € € € Ø € € € Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ø
€
€
€
Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à -Ğ -Ğ -Ğ -Ğ -Ğ -ğ4ğ 4 R¤Yğ 4¸¼ğ 4ğ 4ğ 4ğ 4ğ4¨¬ ğ 4¸3ğ 4ğ 4ğ 4ğ 4ğ A°2´9ğ AÈDĞQTÌIğ Ağ Ağ Ağ Ağ˜RœZğ ¨u°S¸#°X¬ğ Èğ ĞQSÔQ[ğ ğ ğ ğ ğ�r”zğ ¨$ğ ğ ğ ğ ğ&˜ğ &¨3ğ &ğ &ğ &ğ &ğ
E R¤Yğ
Eğ
Eğ
Eğ
EğIğ I˜bœiğ I¨Eğ Iğ Iğ Iğ Iğğ R¤Yğ °5ğ ğ ğ ğ ğ#˜ğ #ğ #ğ #ğ #ğ˜œğ °#ğ ğ ğ ğ ğ" ˜œğ °#ğ ğ ğ ğ ğ gqØDHğ?+ğ ?+¨ğ ?+°s°{¸dğ ?+ĞPSğ ?+Ø.:ğ?+ØGKğ?+Ø`cğ?+ğ ?+ğ ?+ğ ?+ğD¨"¬*ğ ¸"¼*ğ ğ ğ ğ ğ °Hğ À2Ä9ğ ĞW[Ğ\_ÔW`ğ ğ ğ ğ ğ ğ r | 13,692 | Python | .py | 53 | 257.226415 | 2,296 | 0.333675 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,530 | pickler.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/__pycache__/pickler.cpython-311.pyc | §
føf,
ã ó< — d dl Z d dlZd dlmZ G d„ d¦ « ZdS )é N)ÚLockc ób — e Zd ZdZdefd„Zdedefd„Zdefd„Zde fd „Z
dd e d ed
e fd„Z d„ Z dS )Ú DataPicklerzQ
Class for efficiently storing and retrieving data from the file system.
Ú
pickle_dirc óH — d| _ | |¦ « | _ d S )Nr )Úprocessed_countÚ _DataPickler__init_pickle_dirr ©Úselfr s úK/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/pickler.pyÚ__init__zDataPickler.__init__ s# € Ø ˆÔ Ø ×0Ò0°Ñ<Ô<ˆŒˆˆó Úreturnc ó2 — t j |d¬¦ « |S )NT)Úexist_ok)ÚosÚmakedirsr
s r Ú__init_pickle_dirz DataPickler.__init_pickle_dir s € İ
Œ �J¨Ğ.Ñ.Ô.Ğ.ØĞr Úbufferc ó– ‡— t t |d ¦ « ¦ « D ]”Šˆfd„|D ¦ « }t | j › d‰› d�dd¬¦ « 5 }t j ||¦ « | ¦ « t j | ¦ « ¦ « d d d ¦ « n # 1 swxY w Y Œ•| xj
dz
c_
d S ) Nr c ó •— g | ]
}|‰ ‘Œ S © r )Ú.0ÚitemÚindexs €r ú
<listcomp>z-DataPickler.pickle_buffer.<locals>.<listcomp> s ø€ Ğ3Ğ3Ğ3 D�D˜”KĞ3Ğ3Ğ3r ú/item_ú.pklÚabéZ ©Útimeouté ) ÚrangeÚlenr r ÚpickleÚdumpÚflushr ÚfsyncÚfilenor )r r ÚdataÚfr s @r Ú
pickle_bufferzDataPickler.pickle_buffer s ø€ İ�3˜v aœy™>œ>Ñ*Ô*ğ %ğ %ˆEØ3Ğ3Ğ3Ğ3¨FĞ3Ñ3Ô3ˆDݘœ Ğ;Ğ;°Ğ;Ğ;Ğ;¸TÈ2ĞNÑNÔNğ
%ĞRSİ” ˜D !Ñ$Ô$Ğ$Ø—’‘ ” � İ”˜Ÿš™œÑ$Ô$Ğ$ğ
%ğ
%ğ
%ñ
%ô
%ğ
%ğ
%ğ
%ğ
%ğ
%ğ
%øøøğ
%ğ
%ğ
%ğ
%øğ
Ğ Ô Ñ!Ğ Ô Ğ Ğ s ÁAB-Â-B1 Â4B1 r c óğ — | j › d|› d�}g }t |dd¬¦ « 5 } | t j |¦ « ¦ « n# t
$ r Y nw xY wŒ: d d d ¦ « n # 1 swxY w Y |S )Nr r Úrbr r! )r r Úextendr& ÚloadÚEOFError)r r Úfiler+ r, s r Ú
load_indexzDataPickler.load_index sŞ € Ø”/Ğ4Ğ4¨Ğ4Ğ4Ğ4ˆØˆİ
�$˜ bĞ
)Ñ
)Ô
)ğ ¨Qğ
ğØ—K’K¥¤ ¨A¡¤Ñ/Ô/Ğ/Ğ/øİ ğ ğ ğ Ø�Eğøøøğ
ğ ğ ğ ğ ñ ô ğ ğ ğ ğ ğ ğ øøøğ ğ ğ ğ ğ ˆ s4 ¢A+¥'A
Á A+Á
AÁA+ÁAÁA+Á+A/Á2A/r# Ú
batch_sizeÚindexesÚKc ó6 — d„ t j | j ¦ « D ¦ « }| d„ ¬¦ « d„ t t |¦ « ¦ « D ¦ « }t
|¦ « D �]\ }}t | j › d|› �dd¬¦ « 5 }d } t || ¦ « |k rn¿ t j |¦ « }
|D ]ƒ} | t |
¦ « z | cxk r| k rdn Œ"| t |
¦ « z |z
| k r| || t |
¦ « z | z
z
z } | | z } |
| | |z … } ||
| ¦ « Œ„| t |
¦ « z
} n# t $ r Y nw xY wŒÙd d d ¦ « n # 1 swxY w Y �Œd „ t |� D ¦ « S ) Nc ó< — g | ]}| d ¦ « ¯|‘ŒS ©r ©Úendswith©r Úxs r r z(DataPickler.load_all.<locals>.<listcomp>* ó) € ĞNĞNĞN�q¸1¿:º:ÀfÑ;MÔ;MĞN�ĞNĞNĞNr c ó„ — t | d¦ « d d¦ « d ¦ « S )NÚ_r# ú.r )ÚintÚsplit)r> s r ú<lambda>z&DataPickler.load_all.<locals>.<lambda>+ s/ € ¥ Q§W¢W¨S¡\¤\°!¤_×%:Ò%:¸3Ñ%?Ô%?ÀÔ%BÑ!CÔ!C€ r )Úkeyc ó — g | ]}g ‘ŒS r r )r rA s r r z(DataPickler.load_all.<locals>.<listcomp>, s € Ğ.Ğ.Ğ.�q�Ğ.Ğ.Ğ.r ú/r/ r r! r Tc óŒ — g | ]A}|d |d |d d |d d |d d f|d f‘ŒBS )r r# é é r r= s r r z(DataPickler.load_all.<locals>.<listcomp>@ sL € ĞTĞTĞTÀA��1”�q˜”t˜a œd 1œg q¨¤t¨A¤w°°!´°Q´Ğ 8¸!¸A¼$Ğ?ĞTĞTĞTr )
r Úlistdirr Úsortr$ r% Ú enumerater r& r1 r0 r2 Úzip)
r r5 r6 r7 Úfilesr+ Úir3 r, Úlast_lenÚ file_datar Ú
data_points
r Úload_allzDataPickler.load_all) s € ØNĞN�BœJ t¤Ñ 7Ô 7ĞNÑNÔNˆØ
�
Š
ĞCĞCˆ
ÑDÔDĞDØ.Ğ.�E¥# e¡*¤*Ñ -Ô -Ğ.Ñ.Ô.ˆİ Ñ'Ô'ğ ñ ‰GˆAˆtݘœ Ğ1Ğ1¨4Ğ1Ğ1°4ÀĞDÑDÔDğ
ÈØ �ğ ݘ4 œ7‘|”| zÒ1Ğ1Ø ğ İ$*¤K°¡N¤N˜ Ø%,ğ ;ğ ;˜EØ '#¨i©.¬.Ñ 8¸5Ğ LĞ LÒ LĞ LÀHÒ LĞ LĞ LĞ LĞ LØ#+c°)©n¬nÑ#<¸qÑ#@À5Ò#HĞ#HØ$)¨Q°8½cÀ)¹n¼nÑ3LĞPUÑ2UÑ-VÑ$V EØ %¨Ñ 1 Ø-6°u¸UÀQ¹Y°Ô-G
Ø $ Q¤§¢¨zÑ :Ô :Ğ :øà ¥C¨ ¡N¤NÑ2˜˜øİ#ğ ğ ğ Ø ˜ğ øøøğ ğ
ğ
ğ
ñ
ô
ğ
ğ
ğ
ğ
ğ
ğ
øøøğ
ğ
ğ
ğ
ùğ$ UĞTÍÈdÈĞTÑTÔTĞTs7 Â E;Â1B,E Å E;Å
E+Å(E;Å*E+Å+E;Å;E? ÆE? c ó — d„ t j | j ¦ « D ¦ « }|D ] }t j | j › d|› �¦ « Œ!d| _ d S )Nc ó< — g | ]}| d ¦ « ¯|‘ŒS r: r; r= s r r z)DataPickler.clear_dir.<locals>.<listcomp>C r? r rH r )r rL r Úremover )r rP r3 s r Ú clear_dirzDataPickler.clear_dirB sa € ØNĞN�BœJ t¤Ñ 7Ô 7ĞNÑNÔNˆØğ 3ğ 3ˆDİ ŒI˜œ Ğ1Ğ1¨4Ğ1Ğ1Ñ 2Ô 2Ğ 2Ğ 2Ø ˆÔ Ğ Ğ r N)r# )
Ú__name__Ú
__module__Ú __qualname__Ú__doc__Ústrr
r Úlistr- rC r4 rU rY r r r r r sÖ € € € € € ğğ ğ= 3ğ =ğ =ğ =ğ =ğ¨Cğ °Cğ ğ ğ ğ ğ" Dğ "ğ "ğ "ğ "ğ
ğ
ğ
ğ
ğ
ğUğ U 3ğ U°ğ U¸#ğ Uğ Uğ Uğ Uğ2!ğ !ğ !ğ !ğ !r r )r r& Ú portalockerr r r r r ú<module>ra s[ ğØ € € € Ø
€
€
€
à Ğ Ğ Ğ Ğ Ğ ğ?!ğ ?!ğ ?!ğ ?!ğ ?!ñ ?!ô ?!ğ ?!ğ ?!ğ ?!r | 6,669 | Python | .py | 39 | 169.666667 | 1,389 | 0.31398 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,531 | board.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MinimaxOpponent/board.py | ####################### BOARD CLASS ###########################
# The Board class is the data structure that holds the Connect 4 boards and the game operations
import numpy as np
# The Connect 4 board is 7 cells wide and 6 cells tall
# The underlying data structure is a 2-d list
# The first dimension is the column; the second dimension is the row
# Every cell in the above list contains either a 0 or a 1. Player 1 is represented by 0 tiles, and Player
##############################################################################
class Board:
#static class variables - shared across all instances
HEIGHT = 6
WIDTH = 7
def __init__(self, orig=None, hash=None):
# copy
if(orig):
self.board = [list(col) for col in orig.board]
self.numMoves = orig.numMoves
self.lastMove = orig.lastMove
return
# creates from hash - NOTE: Does not understand move order
elif(hash):
self.board = []
self.numMoves = 0
self.lastMove = None
# convert to base 3
digits = []
while hash:
digits.append(int(hash % 3))
hash //= 3
col = []
for item in digits:
# 2 indicates new column
if item == 2:
self.board.append(col)
col = []
# otherwise directly append base number
else:
col.append(item)
self.numMoves += 1
return
# create new
else:
self.board = [[] for x in range(self.WIDTH)]
self.numMoves = 0
self.lastMove = None
return
@classmethod
def from_game_state(cls, game_state:np.ndarray, last_move:tuple):
clss = cls()
clss.board = [[int(x) if x == 1 else 0 for x in reversed(col.tolist()) if x != 0] for col in game_state.T]
clss.numMoves = np.sum(game_state != 0)
clss.lastMove = last_move
return clss
########################################################################
# Mutations
########################################################################
# Puts a piece in the appropriate column and checks to see if it was a winning move
# Pieces are either 1 or 0; automatically decided
def makeMove(self, column):
# update board data
piece = self.numMoves % 2
self.lastMove = (piece, column)
self.numMoves += 1
self.board[column].append(piece)
########################################################################
# Observations
########################################################################
# Generates a list of the valid children of the board
# A child is of the form (move_to_make_child, child_object)
def children(self):
children = []
for i in range(7):
if len(self.board[i]) < 6:
child = Board(self)
child.makeMove(i)
children.append((i, child))
return children
# Returns:
# -1 if game is not over
# 0 if the game is a draw
# 1 if player 1 wins
# 2 if player 2 wins
def isTerminal(self):
for i in range(0,self.WIDTH):
for j in range(0,self.HEIGHT):
try:
if self.board[i][j] == self.board[i+1][j] == self.board[i+2][j] == self.board[i+3][j]:
return self.board[i][j] + 1
except IndexError:
pass
try:
if self.board[i][j] == self.board[i][j+1] == self.board[i][j+2] == self.board[i][j+3]:
return self.board[i][j] + 1
except IndexError:
pass
try:
if not j + 3 > self.HEIGHT and self.board[i][j] == self.board[i+1][j + 1] == self.board[i+2][j + 2] == self.board[i+3][j + 3]:
return self.board[i][j] + 1
except IndexError:
pass
try:
if not j - 3 < 0 and self.board[i][j] == self.board[i+1][j - 1] == self.board[i+2][j - 2] == self.board[i+3][j - 3]:
return self.board[i][j] + 1
except IndexError:
pass
if self.isFull():
return 0
return -1
# Returns a unique decimal number for each board object based on the
# underlying data
def hash(self):
power = 0
hash = 0
for column in self.board:
# add 0 or 1 depending on piece
for piece in column:
hash += piece * (3 ** power)
power += 1
# add a 2 to indicate end of column
hash += 2 * (3 ** power)
power += 1
return hash
########################################################################
# Utilities
########################################################################
# Return true iff the game is full
def isFull(self):
return self.numMoves == 42
# Prints out a visual representation of the board
# X's are 1's and 0's are 0s
def print(self):
print("")
print("+" + "---+" * self.WIDTH)
for rowNum in range(self.HEIGHT - 1, -1, -1):
row = "|"
for colNum in range(self.WIDTH):
if len(self.board[colNum]) > rowNum:
row += " " + ('X' if self.board[colNum][rowNum] else 'O') + " |"
else:
row += " |"
print(row)
print("+" + "---+" * self.WIDTH)
print(self.lastMove[1])
print(self.numMoves)
| 6,165 | Python | .py | 144 | 29.840278 | 147 | 0.440224 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,532 | player.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MinimaxOpponent/player.py | import math
from mu_alpha_zero.MuZero.MinimaxOpponent.board import Board
## Partnered with Nick Konich
class Player:
def __init__(self, depthLimit, isPlayerOne):
self.isPlayerOne = isPlayerOne
self.depthLimit = depthLimit
# Returns a heuristic for the board position
# Good positions for 0 pieces are positive and good positions for 1 pieces
# are be negative
def heuristic(self, board):
heur = 0
state = board.board
for i in range(0, board.WIDTH):
for j in range(0, board.HEIGHT):
# check horizontal streaks
try:
# add player one streak scores to heur
if state[i][j] == state[i + 1][j] == 0:
heur += 10
if state[i][j] == state[i + 1][j] == state[i + 2][j] == 0:
heur += 100
if state[i][j] == state[i + 1][j] == state[i + 2][j] == state[i + 3][j] == 0:
heur += 10000
# subtract player two streak score to heur
if state[i][j] == state[i + 1][j] == 1:
heur -= 10
if state[i][j] == state[i + 1][j] == state[i + 2][j] == 1:
heur -= 100
if state[i][j] == state[i + 1][j] == state[i + 2][j] == state[i + 3][j] == 1:
heur -= 10000
except IndexError:
pass
# check vertical streaks
try:
# add player one vertical streaks to heur
if state[i][j] == state[i][j + 1] == 0:
heur += 10
if state[i][j] == state[i][j + 1] == state[i][j + 2] == 0:
heur += 100
if state[i][j] == state[i][j + 1] == state[i][j + 2] == state[i][j + 3] == 0:
heur += 10000
# subtract player two streaks from heur
if state[i][j] == state[i][j + 1] == 1:
heur -= 10
if state[i][j] == state[i][j + 1] == state[i][j + 2] == 1:
heur -= 100
if state[i][j] == state[i][j + 1] == state[i][j + 2] == state[i][j + 3] == 1:
heur -= 10000
except IndexError:
pass
# check positive diagonal streaks
try:
# add player one streaks to heur
if not j + 3 > board.HEIGHT and state[i][j] == state[i + 1][j + 1] == 0:
heur += 100
if not j + 3 > board.HEIGHT and state[i][j] == state[i + 1][j + 1] == state[i + 2][j + 2] == 0:
heur += 100
if not j + 3 > board.HEIGHT and state[i][j] == state[i + 1][j + 1] == state[i + 2][j + 2] \
== state[i + 3][j + 3] == 0:
heur += 10000
# add player two streaks to heur
if not j + 3 > board.HEIGHT and state[i][j] == state[i + 1][j + 1] == 1:
heur -= 100
if not j + 3 > board.HEIGHT and state[i][j] == state[i + 1][j + 1] == state[i + 2][j + 2] == 1:
heur -= 100
if not j + 3 > board.HEIGHT and state[i][j] == state[i + 1][j + 1] == state[i + 2][j + 2] \
== state[i + 3][j + 3] == 1:
heur -= 10000
except IndexError:
pass
# check negative diagonal streaks
try:
# add player one streaks
if not j - 3 < 0 and state[i][j] == state[i + 1][j - 1] == 0:
heur += 10
if not j - 3 < 0 and state[i][j] == state[i + 1][j - 1] == state[i + 2][j - 2] == 0:
heur += 100
if not j - 3 < 0 and state[i][j] == state[i + 1][j - 1] == state[i + 2][j - 2] \
== state[i + 3][j - 3] == 0:
heur += 10000
# subtract player two streaks
if not j - 3 < 0 and state[i][j] == state[i + 1][j - 1] == 1:
heur -= 10
if not j - 3 < 0 and state[i][j] == state[i + 1][j - 1] == state[i + 2][j - 2] == 1:
heur -= 100
if not j - 3 < 0 and state[i][j] == state[i + 1][j - 1] == state[i + 2][j - 2] \
== state[i + 3][j - 3] == 1:
heur -= 10000
except IndexError:
pass
return heur
class PlayerMM(Player):
def __init__(self, depthLimit, isPlayerOne):
super().__init__(depthLimit, isPlayerOne)
# returns the optimal column to move in by implementing the MiniMax algorithm
def findMove(self, board):
#return self.mmH(board, self.depthLimit, self.isPlayerOne)
#return self.minMaxHelper(board, self.depthLimit, self.isPlayerOne)
score, move = self.miniMax(board, self.depthLimit, self.isPlayerOne)
# print(self.isPlayerOne, "move made", move)
return move
# findMove helper function using miniMax algorithm
def miniMax(self, board, depth, player):
if board.isTerminal() == 0:
return -math.inf if player else math.inf, -1
elif depth == 0:
return self.heuristic(board), -1
if player:
bestScore = -math.inf
shouldReplace = lambda x: x > bestScore
else:
bestScore = math.inf
shouldReplace = lambda x: x < bestScore
bestMove = -1
children = board.children()
for child in children:
move, childboard = child
temp = self.miniMax(childboard, depth - 1, not player)[0]
if shouldReplace(temp):
bestScore = temp
bestMove = move
return bestScore, bestMove
# minimax helper function - unused
def mmH(self, board, depth, player):
if depth == 0:
boards = board.children()
scores = {}
for i in boards:
scores[i[0]] = self.heuristic(i[1])
if player:
return max(scores, key=lambda k: scores[k])
else:
return min(scores, key=lambda k: scores[k])
else:
return self.mmH(board, depth - 1, not player)
class PlayerAB(Player):
def __init__(self, depthLimit, isPlayerOne):
super().__init__(depthLimit, isPlayerOne)
# returns the optimal column to move in by implementing the Alpha-Beta algorithm
def findMove(self, board):
score, move = self.alphaBeta(board, self.depthLimit, self.isPlayerOne, -math.inf, math.inf)
return move
# findMove helper function, utilizing alpha-beta pruning within the minimax algorithm
def alphaBeta(self, board, depth, player, alpha, beta):
if board.isTerminal() == 0:
return -math.inf if player else math.inf, -1
elif depth == 0:
return self.heuristic(board), -1
if player:
bestScore = -math.inf
shouldReplace = lambda x: x > bestScore
else:
bestScore = math.inf
shouldReplace = lambda x: x < bestScore
bestMove = -1
children = board.children()
for child in children:
move, childboard = child
temp = self.alphaBeta(childboard, depth - 1, not player, alpha, beta)[0]
if shouldReplace(temp):
bestScore = temp
bestMove = move
if player:
alpha = max(alpha, temp)
else:
beta = min(beta, temp)
if alpha >= beta:
break
return bestScore, bestMove
class ManualPlayer(Player):
def findMove(self, board):
opts = " "
for c in range(board.WIDTH):
opts += " " + (str(c + 1) if len(board.board[c]) < 6 else ' ') + " "
print(opts)
col = input("Place an " + ('O' if self.isPlayerOne else 'X') + " in column: ")
col = int(col) - 1
return col
| 8,595 | Python | .py | 179 | 31.726257 | 116 | 0.454251 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,533 | mz_node.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_MCTS/mz_node.py | import math
import numpy as np
from mu_alpha_zero.AlphaZero.MCTS.az_node import AlphaZeroNode
import torch as th
class MzAlphaZeroNode(AlphaZeroNode):
def __init__(self, select_probability=0, parent=None, times_visited_init=0, current_player=1):
super().__init__(current_player, select_probability, parent, times_visited_init)
self.reward = 0
def get_best_child(self, min_q: float, max_q: float, gamma: float, multiple_players: bool, c=1.5, c2=19652):
best_utc = -float("inf")
best_child = None
best_action = None
# best_children = []
# best_actions = []
for action, child in self.children.items():
if child.select_probability == 0:
continue
child_utc = child.calculate_utc_score(min_q, max_q, gamma, multiple_players, c=c, c2=c2)
if child_utc >= best_utc:
best_utc = child_utc
best_child = child
best_action = action
# best_children.append(child)
# best_actions.append(action)
# random_idx = random.randint(0, len(best_children) - 1)
# best_child = best_children[random_idx]
# best_action = best_actions[random_idx]
return best_child, best_action
def get_value_pred(self, prediction_forward: callable):
return prediction_forward(self.state)
def expand_node(self, state: th.Tensor,action_probabilities: dict, im_reward: float) -> None:
self.state = state.clone()
self.reward = im_reward
for action, probability in enumerate(action_probabilities):
node = MzAlphaZeroNode(select_probability=probability, parent=self,
current_player=self.current_player * (-1))
self.children[action] = node
def get_immediate_reward(self, dynamics_forward: callable, action: int):
return dynamics_forward(self.state, action)
def calculate_utc_score(self, min_q: float, max_q: float, gamma: float, multiple_players: bool, c=1.5, c2=19652):
parent = self.parent()
if self.times_visited == 0:
return c * self.select_probability * math.sqrt(parent.times_visited + 1e-8)
q = self.scale_q(min_q, max_q, gamma, multiple_players)
utc = q + self.select_probability * (
(math.sqrt(parent.times_visited)) / (1 + self.times_visited)) * (
c + math.log((parent.times_visited + c2 + 1) / c2))
return utc
def scale_q(self, min_q, max_q, gamma: float, multiple_players: bool, val: float or None = None) -> float:
if val is not None:
q = val
else:
q = self.reward + (-self.get_self_value() if multiple_players else self.get_self_value())
if min_q == max_q or (min_q == float("inf") or max_q == float("-inf")) or q == 0:
return q
return (q - min_q) / (max_q - min_q)
| 2,953 | Python | .py | 56 | 42.339286 | 117 | 0.608183 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,534 | mz_search_tree.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_MCTS/mz_search_tree.py | import copy
import gc
import random
import time
import wandb
from multiprocess import set_start_method
set_start_method("spawn", force=True)
from multiprocess.pool import Pool
import numpy as np
import torch as th
from mu_alpha_zero.General.memory import GeneralMemoryBuffer
from mu_alpha_zero.General.mz_game import MuZeroGame
from mu_alpha_zero.General.search_tree import SearchTree
from mu_alpha_zero.Hooks.hook_manager import HookManager
from mu_alpha_zero.Hooks.hook_point import HookAt
from mu_alpha_zero.MuZero.MZ_MCTS.mz_node import MzAlphaZeroNode
from mu_alpha_zero.MuZero.Network.networks import MuZeroNet
from mu_alpha_zero.MuZero.lazy_arrays import LazyArray
from mu_alpha_zero.MuZero.utils import match_action_with_obs, resize_obs, scale_action, scale_state, \
scale_hidden_state, mask_invalid_actions
from mu_alpha_zero.config import MuZeroConfig
from mu_alpha_zero.mem_buffer import MuZeroFrameBuffer, SingleGameData, DataPoint
from mu_alpha_zero.shared_storage_manager import SharedStorage
from typing import Optional
class MuZeroSearchTree(SearchTree):
def __init__(self, game_manager: MuZeroGame, muzero_config: MuZeroConfig,
hook_manager: Optional[HookManager] = None):
self.game_manager = game_manager
self.muzero_config = muzero_config
self.hook_manager = hook_manager if hook_manager is not None else HookManager()
self.buffer = self.init_frame_buffer()
self.min_max_q = [float("inf"), -float("inf")]
def init_frame_buffer(self):
if self.muzero_config.enable_frame_buffer:
return MuZeroFrameBuffer(self.muzero_config.frame_buffer_size, self.game_manager.get_noop(),
self.muzero_config.net_action_size,
ignore_actions=self.muzero_config.frame_buffer_ignores_actions)
return MuZeroFrameBuffer(1, self.game_manager.get_noop(), self.muzero_config.net_action_size,
ignore_actions=self.muzero_config.frame_buffer_ignores_actions)
def play_one_game(self, network_wrapper: MuZeroNet, device: th.device, dir_path: Optional[str] = None,
calculate_avg_num_children: bool = False) -> list[SingleGameData]:
self.buffer = self.init_frame_buffer()
num_steps = self.muzero_config.num_steps
frame_skip = self.muzero_config.frame_skip
state = self.game_manager.reset()
state = resize_obs(state, self.muzero_config.target_resolution, self.muzero_config.resize_images)
state = scale_state(state, self.muzero_config.scale_state)
player = 1
self.buffer.init_buffer(state, player)
if self.muzero_config.multiple_players:
self.buffer.init_buffer(self.game_manager.get_state_for_passive_player(state, -player), -player)
data = SingleGameData()
frame = self.buffer.concat_frames(player).detach().cpu().numpy()
data.add_data_point(
DataPoint(None, None, 0, None, player, frame if dir_path is None else LazyArray(frame, dir_path),
self.game_manager.get_invalid_actions(player)))
game_length = 0
for step in range(num_steps):
game_length += 1
pi, (v, latent) = self.search(network_wrapper, state, player, device, calculate_avg_num_children=(
calculate_avg_num_children and step == 0))
move = self.game_manager.select_move(pi, tau=self.muzero_config.tau)
# _, pred_v = network_wrapper.prediction_forward(latent.unsqueeze(0), predict=True)
state, rew, done = self.game_manager.frame_skip_step(move, player, frame_skip=frame_skip)
state = resize_obs(state, self.muzero_config.target_resolution, self.muzero_config.resize_images)
state = scale_state(state, self.muzero_config.scale_state)
# data.append(
# (pi, v, (rew, move, float(pred_v[0]), player),
# ))
if self.muzero_config.multiple_players:
player = -player
self.buffer.add_frame(state, scale_action(move, self.game_manager.get_num_actions()), player)
self.buffer.add_frame(self.game_manager.get_state_for_passive_player(state, -player),
scale_action(move, self.game_manager.get_num_actions()), -player)
data.datapoints[-1].v = v
data.datapoints[-1].pi = [x for x in pi.values()]
data.datapoints[-1].move = move
frame = self.buffer.concat_frames(player).detach().cpu().numpy()
data.add_data_point(
DataPoint(np.ones((len(data.datapoints[-1].pi))) / len(data.datapoints[-1].pi), 0, rew,
random.randint(0, len(data.datapoints[-1].pi) - 1), player,
frame if dir_path is None else LazyArray(frame, dir_path),
self.game_manager.get_invalid_actions(player)))
if done:
# time.sleep(1)
# print(player)
# print(v)
break
# self.game_manager.render()
# time.sleep(0.5)
try:
wandb.log({"Game length": game_length})
except Exception:
pass
data.compute_initial_priorities(self.muzero_config)
return [data]
def search(self, network_wrapper, state: np.ndarray, current_player: Optional[int], device: th.device,
tau: Optional[float] = None, calculate_avg_num_children: bool = False, use_state_directly: bool = False):
self.min_max_q = [float("inf"), -float("inf")]
if self.buffer.__len__(current_player) == 0 and not use_state_directly:
self.buffer.init_buffer(state, current_player)
num_simulations = self.muzero_config.num_simulations
if tau is None:
tau = self.muzero_config.tau
games = [copy.deepcopy(self.game_manager)] if self.muzero_config.use_true_game_state_in_tree else []
root_node = MzAlphaZeroNode(current_player=current_player)
# print(self.buffer.buffers[current_player][-1][0][:,:,0])
game_state = state if use_state_directly else self.buffer.concat_frames(current_player)
state_ = network_wrapper.representation_forward(
game_state.permute(2, 0, 1).unsqueeze(0).to(device)).squeeze(0)
if self.muzero_config.scale_hidden_state:
state_ = scale_hidden_state(state_)
pi, v = network_wrapper.prediction_forward(state_.unsqueeze(0), predict=True)
if self.muzero_config.dirichlet_alpha > 0:
pi = pi + np.random.dirichlet([self.muzero_config.dirichlet_alpha] * self.muzero_config.net_action_size)
pi = mask_invalid_actions(self.game_manager.get_invalid_actions(current_player), pi)
pi = pi.flatten().tolist()
root_node.expand_node(state_, pi, 0)
for simulation in range(num_simulations):
current_node = root_node
path = [current_node]
action = None
while current_node.was_visited():
done = False
current_node, action = current_node.get_best_child(self.min_max_q[0], self.min_max_q[1],
self.muzero_config.gamma,
self.muzero_config.multiple_players,
c=self.muzero_config.c, c2=self.muzero_config.c2)
path.append(current_node)
if self.muzero_config.use_true_game_state_in_tree:
game = copy.deepcopy(games[-1])
_, rew, done = game.get_next_state(action, current_node.parent().current_player)
games.append(game)
# action = scale_action(action, self.game_manager.get_num_actions())
if not done:
current_node_state_with_action = match_action_with_obs(current_node.parent().state, action,
self.muzero_config)
next_state, reward = network_wrapper.dynamics_forward(current_node_state_with_action.unsqueeze(0),
predict=True)
if self.muzero_config.scale_hidden_state:
next_state = scale_hidden_state(next_state)
reward = reward[0][0]
pi, v = network_wrapper.prediction_forward(next_state.unsqueeze(0), predict=True)
if self.muzero_config.use_true_game_state_in_tree:
reward = rew
pi = mask_invalid_actions(games[-1].get_invalid_actions(current_node.current_player), pi)
pi = pi.flatten().tolist()
v = v.flatten().tolist()[0]
current_node.expand_node(next_state, pi, reward)
else:
if self.muzero_config.multiple_players:
v = -1
else:
v = rew
self.backprop(v, path, games)
action_probs = root_node.get_self_action_probabilities()
root_val_latent = (root_node.get_self_value(), root_node.get_latent())
self.hook_manager.process_hook_executes(self, self.search.__name__, __file__, HookAt.TAIL,
args=(action_probs, root_val_latent, root_node))
# if calculate_avg_num_children:
# num_nodes = self.get_num_nodes(root_node)
# wandb.log({"Number of non-leaf nodes": num_nodes})
root_node = None
return action_probs, root_val_latent
def backprop(self, v: float, path: list, games: list):
# G = v
G_node = v
gamma = self.muzero_config.gamma
for node in reversed(path):
node.times_visited += 1
if self.muzero_config.multiple_players:
# G = node.reward + gamma * (
# -G) # G should be from the perspective of the parent as the parent is selecting from the children based on what's good for them.
node.total_value += G_node
self.update_min_max_q(node.reward - node.get_self_value())
G_node = node.reward + gamma * (-G_node)
else:
node.total_value += G_node
self.update_min_max_q(node.reward + node.get_self_value())
G_node = node.reward + gamma * G_node
if self.muzero_config.use_true_game_state_in_tree and node.select_probability != 0:
games.pop()
# node.update_q(G_node)
def self_play(self, net: MuZeroNet, device: th.device, num_games: int, memory: GeneralMemoryBuffer) -> tuple[
int, int, int]:
for game in range(num_games):
game_results = self.play_one_game(net, device, calculate_avg_num_children=game == num_games - 1)
memory.add_list(game_results)
return None, None, None
def make_fresh_instance(self):
return MuZeroSearchTree(self.game_manager.make_fresh_instance(), copy.deepcopy(self.muzero_config),
hook_manager=copy.deepcopy(self.hook_manager))
def step_root(self, action: int or None):
# I am never reusing the tree in MuZero.
pass
def update_min_max_q(self, q):
self.min_max_q[0] = min(self.min_max_q[0], q)
self.min_max_q[1] = max(self.min_max_q[1], q)
def get_num_nodes(self, root_node: MzAlphaZeroNode):
num_nodes = 0
if len(root_node.children) > 0:
num_nodes = 1
for child in root_node.children.values():
nn = self.get_num_nodes(child)
num_nodes += nn
return num_nodes
@staticmethod
def parallel_self_play(nets: list, trees: list, memory: GeneralMemoryBuffer, device: th.device, num_games: int,
num_jobs: int):
with Pool(num_jobs) as p:
if memory.is_disk and memory.full_disk:
results = p.starmap(p_self_play, [
(nets[i], trees[i], copy.deepcopy(device), num_games // num_jobs, copy.deepcopy(memory), None) for i
in range(len(nets))])
else:
results = p.starmap(p_self_play, [
(nets[i], trees[i], copy.deepcopy(device), num_games // num_jobs, None, memory.dir_path) for i in
range(len(nets))])
for result in results:
memory.add_list(result)
return None, None, None
@staticmethod
def start_continuous_self_play(nets: list, trees: list, shared_storage: SharedStorage,
device: th.device, config: MuZeroConfig, num_jobs: int, num_worker_iters: int):
pool = Pool(num_jobs)
for i in range(num_jobs):
pool.apply_async(c_p_self_play, args=(
nets[i], trees[i], copy.deepcopy(device), config, i, shared_storage, num_worker_iters,
shared_storage.get_dir_path())
)
return pool
@staticmethod
def reanalyze(net, tree, device, shared_storage: SharedStorage, config: MuZeroConfig):
wandb.init(project=config.wandbd_project_name, name="Reanalyze")
net = net.to(device)
while len(shared_storage.get_buffer()) < 10_000:
time.sleep(5)
for iter_ in range(config.num_worker_iters):
data = random.choice(shared_storage.get_buffer())
net.eval()
if shared_storage.get_experimental_network_params() is not None:
net.load_state_dict(shared_storage.get_experimental_network_params())
else:
net.load_state_dict(shared_storage.get_stable_network_params())
wandb.log({"reanalyze_iteration": iter_})
tree = tree.make_fresh_instance()
for data_point in data.datapoints:
if isinstance(data_point.frame, LazyArray):
frame = data_point.frame.load_array()
else:
frame = data_point.frame
state = th.tensor(frame, device=device, dtype=th.float32)
pi, (v, _) = tree.search(net, state, data_point.player, device, use_state_directly=True)
data_point.v = v
data_point.pi = [x for x in pi.values()]
data.compute_initial_priorities(config)
def run_on_training_end(self):
self.hook_manager.process_hook_executes(self, self.run_on_training_end.__name__, __file__, HookAt.ALL)
def p_self_play(net, tree, dev, num_g, mem, dir_path: Optional[str] = None):
data = []
for game in range(num_g):
game_results = tree.play_one_game(net, dev, dir_path=dir_path, calculate_avg_num_children=game == num_g - 1)
if mem is not None:
mem.add_list(game_results)
else:
data.extend(game_results)
return data
def c_p_self_play(net, tree, device, config: MuZeroConfig, p_num: int, shared_storage: SharedStorage,
num_worker_iters: int,
dir_path: Optional[str] = None):
if p_num == 0:
wandb.init(project=config.wandbd_project_name, name="Self play")
net = net.to(device)
for iter_ in range(num_worker_iters):
# for game in range(num_g):
# if not shared_storage.get_was_pitted():
# # If the network was not yet decided on, slow down the process so the data won't get overpopulated with current params.
# time.sleep(5)
if shared_storage.get_experimental_network_params() is None:
params = shared_storage.get_stable_network_params()
else:
params = shared_storage.get_experimental_network_params()
net.eval()
net.load_state_dict(params)
game_results = tree.play_one_game(net, device, dir_path=dir_path)
shared_storage.add_list(game_results)
| 16,087 | Python | .py | 286 | 43.157343 | 151 | 0.601066 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,535 | mz_search_tree.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_MCTS/__pycache__/mz_search_tree.cpython-311.pyc | §
|Ú£fX3 ã óf — d dl Z d dlZd dlZd dlZd dlmZ edd¬¦ « d dlmZ d dlZ d dl
Z d dl m
Z
d dlmZ d d lmZ d d
lmZ d d lmZ d d lmZ d d
lmZ d dlmZ d dl m Z m Z m Z m Z m!Z!m"Z" d dl#m$Z$ d dl%m&Z&m'Z'm(Z( d dl)m*Z* G d„ de¦ « Z+d de,pdfd„Z- d de$de.de*de.de,pdf
d„Z/dS ) é N)Úset_start_methodÚspawnT)Úforce)ÚPool)ÚGeneralMemoryBuffer)Ú
MuZeroGame)Ú
SearchTree)Ú HookManager)ÚHookAt)ÚMzAlphaZeroNode)Ú MuZeroNet)Ú LazyArray)Úmatch_action_with_obsÚ
resize_obsÚ scale_actionÚ scale_stateÚscale_hidden_stateÚmask_invalid_actions)Ú MuZeroConfig)ÚMuZeroFrameBufferÚSingleGameDataÚ DataPoint)Ú
SharedStoragec óª — e Zd Zd'dededepdfd„Zd„ Z d(ded e j
d
e pdd e d e
e f
d
„Z d(dej depdd e j
depdd e f
d„Zd„ Zded e j
deded eeeef f
d„Zd„ Zdepdfd„Zd„ Zde fd „Z e d e
d e
ded e j
ded ef d „¦ « Z e d e
d e
d!e d e j
d"ed ed#efd$„¦ « Z!e d!e d"efd%„¦ « Z"d&„ Z#dS ))ÚMuZeroSearchTreeNÚ game_managerÚ
muzero_configÚ hook_managerc óÊ — || _ || _ |�|n
t ¦ « | _ | ¦ « | _ t
d¦ « t
d¦ « g| _ d S )NÚinf)r r r
r Úinit_frame_bufferÚbufferÚfloatÚ min_max_q)Úselfr r r s úZ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/MZ_MCTS/mz_search_tree.pyÚ__init__zMuZeroSearchTree.__init__ sX € Ø (ˆÔØ *ˆÔØ,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ×,Ò,Ñ.Ô.ˆŒ İ ™,œ,¨u© ¬ ¨
Ğ6ˆŒˆˆó c óö — | j j r<t | j j | j ¦ « | j j ¦ « S t d| j ¦ « | j j ¦ « S ©Né )r Úenable_frame_bufferr Úframe_buffer_sizer Úget_noopÚnet_action_size©r% s r& r! z"MuZeroSearchTree.init_frame_buffer% so € Ø Ô Ô 1ğ Iİ$ TÔ%7Ô%IÈ4ÔK\×KeÒKeÑKgÔKgØ%)Ô%7Ô%GñIô Iğ
Iå DÔ$5×$>Ò$>Ñ$@Ô$@À$ÔBTÔBdÑeÔeĞer( FÚnetwork_wrapperÚdeviceÚdir_pathÚcalculate_avg_num_childrenÚreturnc ó — | j j }| j j }| j ¦ « }t || j j | j j ¦ « }t || j j ¦ « }d}| j
||¦ « | j j r6| j
| j || ¦ « | ¦ « t ¦ « } d}
t |¦ « D �]Ì} |
dz
}
| |||||o| dk ¬¦ « \ } \ }
}| j | | j j ¬¦ « }| j |||¬¦ « \ }}}t || j j | j j ¦ « }t || j j ¦ « }t' || j ¦ « ¦ « }| j |¦ « ¦ « ¦ « ¦ « }t3 | |
||||€|nt5 ||¦ « ¦ « }| |¦ « |r nd| j j r| }| j |||¦ « | j | j || ¦ « || ¦ « �ŒÎ t; j d|
i¦ « n# t> $ r Y nw xY w| | j ¦ « | gS )Nr+ r ©r4 ©Útau)Ú
frame_skipz Game length)!r Ú num_stepsr: r Úresetr Útarget_resolutionÚ
resize_imagesr r" Ú init_bufferÚmultiple_playersÚ get_state_for_passive_playerr ÚrangeÚsearchÚ select_mover9 Úframe_skip_stepr Úget_num_actionsÚ
concat_framesÚdetachÚcpuÚnumpyr r Úadd_data_pointÚ add_frameÚwandbÚlogÚ ExceptionÚcompute_initial_priorities)r% r1 r2 r3 r4 r; r: ÚstateÚplayerÚdataÚ game_lengthÚstepÚpiÚvÚlatentÚmoveÚrewÚdoneÚframeÚ
data_points r& Ú
play_one_gamez MuZeroSearchTree.play_one_game+ s € àÔ&Ô0ˆ ØÔ'Ô2ˆ
ØÔ!×'Ò'Ñ)Ô)ˆİ˜5 $Ô"4Ô"FÈÔHZÔHhÑiÔiˆİ˜E 4Ô#5Ô#AÑBÔBˆØˆØ Œ × Ò vÑ.Ô.Ğ.Ø Ô Ô .ğ mØ ŒK× #Ò # DÔ$5×$RÒ$RĞSXĞ[aĞZaÑ$bÔ$bĞekĞdkÑ lÔ lĞ lİ Ñ Ô ˆØˆ ݘ)Ñ$Ô$ğ qñ qˆDØ ˜1Ñ ˆKØ "Ÿkšk¨/¸5À&È&Ø.Ğ<°4¸1²9ğ *ñ ?ô ?‰OˆB‘ ��FàÔ$×0Ò0°¸Ô9KÔ9OĞ0ÑPÔPˆDà #Ô 0× @Ò @ÀÀvĞZdĞ @Ñ eÔ eÑ ˆE�3˜İ ˜u dÔ&8Ô&JÈDÔL^ÔLlÑmÔmˆEİ tÔ'9Ô'EÑFÔFˆEİ dÔ&7×&GÒ&GÑ&IÔ&IÑJÔJˆDØ”K×-Ò-¨fÑ5Ô5×<Ò<Ñ>Ô>×BÒBÑDÔD×JÒJÑLÔLˆEİ" 2 q¨#¨t°VÀhĞFV¸U¸UÕ\eĞfkĞmuÑ\vÔ\vÑwÔwˆJğ
× Ò
Ñ +Ô +Ğ +Øğ
Ø�ØÔ!Ô2ğ
!Ø ˜�Ø ŒK× !Ò ! %¨¨vÑ 6Ô 6Ğ 6Ø ŒK× !Ò ! $Ô"3×"PÒ"PĞQVĞY_ĞX_Ñ"`Ô"`ĞbfĞioĞhoÑ pÔ pĞ pÑ pğ İ ŒI�} kĞ2Ñ 3Ô 3Ğ 3Ğ 3øİğ ğ ğ Ø ˆDğ øøøà ×'Ò'¨Ô(:Ñ;Ô;Ğ;؈vˆ
s Ê>K Ë
K"Ë!K"rQ Úcurrent_playerr9 c óZ — | j |¦ « dk r| j ||¦ « | j j }|€ | j j }t
|¬¦ « }| | j |¦ « ddd¦ «
d¦ « ¦ « d¦ « } t | ¦ « } |
|
d¦ « d¬¦ « \ }
} | j j dk r:|
t j | j j g| j j z ¦ « z }
t' | j ||¦ « |
¦ « }
|
¦ « ¦ « }
| | |
d¦ « t3 |¦ « D �]ç} |}
|
g}d }|
¦ « r…|
| j d | j d | j j | j j | j j | j j ¬¦ « \ }
}| ! |
¦ « |
¦ « °…tE || j # ¦ « ¦ « }tI |
% ¦ « j& |¦ « }| ' |
d¦ « d¬¦ « \ }}t |¦ « }|d d }|
|
d¦ « d¬¦ « \ }
} |
¦ « ¦ « }
| ¦ « ¦ « d } |
||
|¦ « | ( | |¦ « �Œé| ) ¦ « }| * ¦ « | + ¦ « f}| j, - | | j. j/ t` tb j2 |||f¬¦ « d }||fS ) Nr )r_ é r+ T)Úpredict)ÚcÚc2©Úargs)3r" Ú__len__r? r Únum_simulationsr9 r Úrepresentation_forwardrG ÚpermuteÚ unsqueezeÚsqueezer Úprediction_forwardÚdirichlet_alphaÚnpÚrandomÚ dirichletr/ r r Úget_invalid_actionsÚflattenÚtolistÚ expand_noderB Ú was_visitedÚget_best_childr$ Úgammar@ rc rd Úappendr rF r ÚparentrQ Údynamics_forwardÚbackpropÚ get_self_action_probabilitiesÚget_self_valueÚ
get_latentr Úprocess_hook_executesrC Ú__name__Ú__file__r ÚTAIL)r% r1 rQ r_ r2 r9 r4 rh Ú root_nodeÚstate_rV rW Ú
simulationÚ current_nodeÚpathÚactionÚ current_node_state_with_actionÚ
next_stateÚrewardÚ action_probsÚroot_val_latents r& rC zMuZeroSearchTree.searchV sñ € à Œ;× Ò ˜~Ñ .Ô .°!Ò 3Ğ 3Ø ŒK× #Ò # E¨>Ñ :Ô :Ğ :Ø Ô,Ô<ˆØ ˆ;ØÔ$Ô(ˆCå#°>ĞBÑBÔBˆ à ×7Ò7Ø ŒK× %Ò % nÑ 5Ô 5× =Ò =¸aÀÀAÑ FÔ F× PÒ PĞQRÑ SÔ SñUô UßU\ÒU\Ğ]^ÑU_ÔU_ğ å# FÑ+Ô+ˆØ ×2Ò2°6×3CÒ3CÀAÑ3FÔ3FĞPTĞ2ÑUÔU‰ˆˆAØ Ô Ô -°Ò 1Ğ 1Ø•b”i×)Ò)¨4Ô+=Ô+MĞ*NĞQUÔQcÔQsÑ*sÑtÔtÑtˆBİ
! $Ô"3×"GÒ"GÈÈ~Ñ"^Ô"^Ğ`bÑ
cÔ
cˆØ
�ZŠZ‰\Œ\×
Ò
Ñ
"Ô
"ˆØ× Ò ˜f b¨!Ñ,Ô,Ğ,İ Ñ0Ô0ğ #ñ #ˆJØ$ˆLØ �>ˆD؈FØ ×*Ò*Ñ,Ô,ğ
*Ø'3×'BÒ'BÀ4Ä>ĞRSÔCTĞVZÔVdĞefÔVgØCGÔCUÔC[ØCGÔCUÔCfØEIÔEWÔEYĞ^bÔ^pÔ^sğ (Cñ (uô (uÑ$� ˜fğ — ’ ˜LÑ)Ô)Ğ)ğ ×*Ò*Ñ,Ô,ğ
*õ " &¨$Ô*;×*KÒ*KÑ*MÔ*MÑNÔNˆFå-BÀ<×CVÒCVÑCXÔCXÔC^Ğ`fÑ-gÔ-gĞ *Ø!0×!AÒ!AĞB`×BjÒBjĞklÑBmÔBmØJNğ "Bñ "Pô "PÑ ˆJ˜å+¨JÑ7Ô7ˆJؘA”Y˜q”\ˆFØ#×6Ò6°z×7KÒ7KÈAÑ7NÔ7NĞX\Ğ6Ñ]Ô]‰EˆB�Ø—’‘ ” ×$Ò$Ñ&Ô&ˆBØ— ’ ‘ ” ×"Ò"Ñ$Ô$ QÔ'ˆAØ × $Ò $ Z°°VÑ <Ô <Ğ <Ø �MŠM˜!˜TÑ "Ô "Ğ "Ñ "à ×>Ò>Ñ@Ô@ˆ Ø$×3Ò3Ñ5Ô5°y×7KÒ7KÑ7MÔ7MĞNˆØ Ô×/Ò/°°d´kÔ6JÍHÕV\ÔVaØ6BÀOĞU^Ğ5_ğ 0ñ aô ağ ağ
ˆ ؘ_Ğ,Ğ,r( c ó² — |}| j j }t |¦ « D ]¸}| j j rN|xj |z
c_ | |j | ¦ « z
¦ « |j || z z }nL|xj |z
c_ | |j | ¦ « z ¦ « |j ||z z }|xj dz
c_ Œ¹d S r* ) r rx Úreversedr@ Ú total_valueÚupdate_min_max_qrŒ r~ Ú
times_visited)r% rW rˆ ÚG_noderx Únodes r& r| zMuZeroSearchTree.backprop‹ só € àˆØÔ"Ô(ˆİ ˜T‘N”Nğ
$ğ
$ˆDàÔ!Ô2ğ
6ğ Ğ Ô FÑ*Ğ Ô Ø×%Ò% d¤k°D×4GÒ4GÑ4IÔ4IÑ&IÑJÔJĞJØ œ u°°Ñ'8Ñ8��àĞ Ô FÑ*Ğ Ô Ø×%Ò% d¤k°D×4GÒ4GÑ4IÔ4IÑ&IÑJÔJĞJØ œ u¨v¡~Ñ5�à Ğ Ô !Ñ #Ğ Ô Ğ ğ
$ğ
$r( ÚnetÚ num_gamesÚmemoryc ó’ — t |¦ « D ]6}| ||||dz
k ¬¦ « }| |¦ « Œ7dS )Nr+ r7 ©NNN)rB r^ Úadd_list)r% r– r2 r— r˜ ÚgameÚ game_resultss r& Ú self_playzMuZeroSearchTree.self_playŸ s[ € å˜)Ñ$Ô$ğ *ğ *ˆDØ ×-Ò-¨c°6ĞVZĞ^gĞjkÑ^kÒVkĞ-ÑlÔlˆLØ �OŠO˜LÑ )Ô )Ğ )Ğ )à Ğ r( c ó° — t | j ¦ « t j | j ¦ « t j | j ¦ « ¬¦ « S )N)r )r r Úmake_fresh_instanceÚcopyÚdeepcopyr r r0 s r& r z$MuZeroSearchTree.make_fresh_instance§ sN € İ Ô 1× EÒ EÑ GÔ GÍÌ ĞW[ÔWiÑIjÔIjİ-1¬]¸4Ô;LÑ-MÔ-MğOñ Oô Oğ Or( r‰ c ó — d S ©N© )r% r‰ s r& Ú step_rootzMuZeroSearchTree.step_root« s € à ˆr( c ó’ — t | j d |¦ « | j d< t | j d |¦ « | j d< d S ©Nr r+ )Úminr$ Úmax)r% Úqs r& r’ z!MuZeroSearchTree.update_min_max_q¯ sB € İ ¤¨qÔ 1°1Ñ 5Ô 5ˆŒ�qÑİ ¤¨qÔ 1°1Ñ 5Ô 5ˆŒ�qÑĞĞr( r„ c óª — d}t |j ¦ « dk rd}|j ¦ « D ] }| |¦ « }||z
}Œ |S r¨ )ÚlenÚchildrenÚvaluesÚ
get_num_nodes)r% r„ Ú num_nodesÚchildÚnns r& r° z MuZeroSearchTree.get_num_nodes³ sd € ؈ İ ˆyÔ!Ñ "Ô " QÒ &Ğ &؈IØ Ô'×.Ò.Ñ0Ô0ğ ğ ˆEØ×#Ò# EÑ*Ô*ˆBØ ˜‰OˆIˆIàĞr( ÚnetsÚtreesÚnum_jobsc
óÊ ‡ ‡‡‡‡‡— t ‰¦ « 5 }‰j rN‰j rG| t ˆˆˆ ˆˆˆfd„t t
‰ ¦ « ¦ « D ¦ « ¦ « }nF| t ˆˆˆ ˆˆˆfd„t t
‰ ¦ « ¦ « D ¦ « ¦ « }d d d ¦ « n # 1 swxY w Y |D ]}‰ |¦ « ŒdS )Nc ó† •— g | ]=}‰| ‰| t j ‰¦ « ‰‰z t j ‰¦ « d f‘Œ>S r¤ )r¡ r¢ ©Ú.0Úir2 r˜ r´ r— r¶ rµ s €€€€€€r& ú
<listcomp>z7MuZeroSearchTree.parallel_self_play.<locals>.<listcomp>Â sX ø€ ğ 2)ğ 2)ğ 2)Øwx�T˜!”W˜e Aœh¬
°fÑ(=Ô(=¸yÈHÑ?TÕVZÔVcĞdjÑVkÔVkĞmqĞrğ2)ğ 2)ğ 2)r( c ól •— g | ]0}‰| ‰| t j ‰¦ « ‰‰z d ‰j f‘Œ1S r¤ )r¡ r¢ r3 r¹ s €€€€€€r& r¼ z7MuZeroSearchTree.parallel_self_play.<locals>.<listcomp>Æ sO ø€ ğ 2&ğ 2&ğ 2&Øqr�T˜!”W˜e Aœh¬
°fÑ(=Ô(=¸yÈHÑ?TĞVZĞ\bÔ\kĞlğ2&ğ 2&ğ 2&r( rš )r Úis_diskÚ full_diskÚstarmapÚ p_self_playrB r r› ) r´ rµ r˜ r2 r— r¶ ÚpÚresultsÚresults `````` r& Úparallel_self_playz#MuZeroSearchTree.parallel_self_play½ sp øøøøøø€ õ �(‰^Œ^ğ '˜qØŒ~ğ
' &Ô"2ğ
'ØŸ)š)¥Kğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)å �S ™YœYÑ'Ô'ğ2)ñ 2)ô 2)ñ *ô *��ğ Ÿ)š)¥Kğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&å�#˜d™)œ)Ñ$Ô$ğ2&ñ 2&ô 2&ñ 'ô '�ğ
'ğ 'ğ 'ñ 'ô 'ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'øøøğ 'ğ 'ğ 'ğ 'ğ ğ $ğ $ˆFØ �OŠO˜FÑ #Ô #Ğ #Ğ #à Ğ s –B B>Â>CÃCÚshared_storageÚconfigÚnum_worker_itersc
óğ — t |¦ « }t |¦ « D ]V}| t | | || t j |¦ « ||||| ¦ « f¬¦ « ŒW|S )Nre )r rB Ú apply_asyncÚ
c_p_self_playr¡ r¢ Ú get_dir_path) r´ rµ rÆ r2 rÇ r¶ rÈ Úpoolr» s r& Ústart_continuous_self_playz+MuZeroSearchTree.start_continuous_self_playÎ sƒ € õ �H‰~Œ~ˆİ�x‘ ” ğ ğ ˆAØ × Ò �]Ø�Q”˜˜qœ¥4¤=°Ñ#8Ô#8¸&À!À^ĞUeØ ×+Ò+Ñ-Ô-ğ2/Ğ ñ
ô
ğ
ğ
ğ
ˆ r( c óì — dt dt fd„}| |¦ « } | ¦ « t | ¦ « ¦ « |j dz k rAt j d¦ « t | ¦ « ¦ « |j dz k °At |j
¦ « D ]µ} ||j dz |¦ « }|D ]œ\ }} \ }
} } }
}t |t ¦ « r|
¦ « }t j ||¬¦ « ¦ « }| | ||
|¦ « \ }\ } }|j ||j ¬¦ « } Œ�Œ¶d S ) NÚnÚmemc ó� — | ¦ « }g }t | ¦ « D ] }| || |f¦ « Œ |S r¤ )Ú
get_bufferrB ry )rĞ rÑ r" rS r» s r& Ú get_first_nz/MuZeroSearchTree.reanalyze.<locals>.get_first_nÜ sM € Ø—^’^Ñ%Ô%ˆF؈Dݘ1‘X”Xğ
,ğ
,�Ø— ’ ˜V AœY¨˜NÑ+Ô+Ğ+Ğ+؈Kr( é é ra )r2 r8 )Úintr ÚtoÚevalr rÓ Ú
batch_sizeÚtimeÚsleeprB rÈ Ú
isinstancer Ú
load_arrayÚthÚtensorr# rC r rD r9 )r– Útreer2 rÆ rÇ rÔ Úiter_rS rV rW rZ rY Úpred_vrR r\ rQ Ú_s r& Ú reanalyzezMuZeroSearchTree.reanalyzeÚ sŠ € ğ �3ğ ¥]ğ ğ ğ ğ ğ �fŠf�V‰nŒnˆØ �Š‰
Œ
ˆ
İ�.×+Ò+Ñ-Ô-Ñ.Ô.°Ô1BÀQÑ1FÒFĞFİ ŒJ�q‰MŒMˆMõ �.×+Ò+Ñ-Ô-Ñ.Ô.°Ô1BÀQÑ1FÒFĞFå˜6Ô2Ñ3Ô3ğ Iğ IˆEØ �;˜vÔ 0°1Ñ 4°nÑEÔEˆDØ=Ağ
Iğ
IÑ9��AÑ2˜˜T 6¨6°Eİ ˜e¥YÑ/Ô/ğ /Ø !× ,Ò ,Ñ .Ô .�Eİœ %°Ğ7Ñ7Ô7×=Ò=Ñ?Ô?�Ø !Ÿ[š[¨¨e°V¸VÑ DÔ D‘
�‘F�Q˜ØÔ(×4Ò4°R¸V¼ZĞ4ÑHÔH��ğ
Iğ Iğ Ir( c ór — | j | | j j t t
j ¦ « d S r¤ )r r€ Úrun_on_training_endr� r‚ r ÚALLr0 s r& rç z$MuZeroSearchTree.run_on_training_endğ s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr( r¤ )NF)$r� Ú
__module__Ú __qualname__r r r
r' r! r
rß r2 ÚstrÚboolÚlistr r^ ro Úndarrayr× r# rC r| r Útupler� r r¦ r’ r r° Ú staticmethodrÅ r rÎ rå rç r¥ r( r& r r s« € € € € € ğ7ğ 7 Zğ 7À ğ 7Ğ\gĞ\oĞkoğ 7ğ 7ğ 7ğ 7ğfğ fğ fğ dhØ9>ğ)ğ )¨Yğ )ÀÄ ğ )ĞUXĞU`Ğ\`ğ )Ø26ğ)ØCGÈÔCWğ)ğ )ğ )ğ )ğX NSğ3-ğ 3-¨R¬Zğ 3-ÈÈĞPTğ 3-Ğ^`Ô^gğ 3-Ø�M˜Tğ3-ØFJğ3-ğ 3-ğ 3-ğ 3-ğj$ğ $ğ $ğ( ˜Yğ °´ ğ Àcğ ĞSfğ ĞkpØ ˆS�#ˆ
ôlğ ğ ğ ğ ğOğ Oğ Oğ
tğ
ğ
ğ
ğ
ğ6ğ 6ğ 6ğ ğ ğ ğ ğ ğ ğ ğ ¨dğ Ğ<Oğ ĞY[ÔYbğ Ğorğ Ø%(ğ ğ ğ ñ „\ğ ğ ğ ¨ğ °dğ ÈMğ Ø+-¬9ğ Ø>Jğ ØVYğ Ømpğ ğ ğ ñ „\ğ ğ ğI°]ğ IÈLğ Iğ Iğ Iñ „\ğIğ*oğ oğ oğ oğ or( r r3 c óÈ — g }t |¦ « D ]O}| | ||||dz
k ¬¦ « }|�| |¦ « Œ:| |¦ « ŒP|S )Nr+ )r3 r4 )rB r^ r› Úextend) r– rá ÚdevÚnum_grÑ r3 rS rœ r� s r& rÁ rÁ ô sy € Ø
€Dİ�e‘ ” ğ &ğ &ˆØ×)Ò)¨#¨s¸XĞbfĞjoĞrsÑjsÒbsĞ)ÑtÔtˆ Ø ˆ?Ø �LŠL˜ Ñ &Ô &Ğ &Ğ &à �KŠK˜ Ñ %Ô %Ğ %Ğ %Ø €Kr( rÇ Úp_numrÆ rÈ c ó¼ — |dk rt j |j d¬¦ « | |¦ « } | ¦ « t |¦ « D ]�}| ¦ « €| ¦ « } n| ¦ « } | | ¦ « | | ||¬¦ « }
|
|
¦ « Œ‚d S )Nr z Self play)ÚprojectÚname)r3 ) rM ÚinitÚwandbd_project_namerØ rÙ rB Ú get_experimental_network_paramsÚget_stable_network_paramsÚload_state_dictr^ r› ) r– rá r2 rÇ rõ rÆ rÈ r3 râ Úparamsr� s r& rË rË ÿ sà € ğ
�‚z€zİ
Œ
˜6Ô5¸KĞHÑHÔHĞHØ
�&Š&�‰.Œ.€C؇H‚H�J„J€JİĞ'Ñ(Ô(ğ .ğ .ˆğ
× 9Ò 9Ñ ;Ô ;Ğ CØ#×=Ò=Ñ?Ô?ˆFˆFà#×CÒCÑEÔEˆFØ ×Ò˜FÑ#Ô#Ğ#Ø×)Ò)¨#¨vÀĞ)ÑIÔIˆ Ø× Ò Ñ-Ô-Ğ-Ğ-ğ .ğ .r( r¤ )0r¡ ÚgcrÛ rM Ú multiprocessr Úmultiprocess.poolr rJ ro Útorchrß Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.mz_gamer Ú!mu_alpha_zero.General.search_treer Ú mu_alpha_zero.Hooks.hook_managerr
Ú mu_alpha_zero.Hooks.hook_pointr Ú$mu_alpha_zero.MuZero.MZ_MCTS.mz_noder Ú%mu_alpha_zero.MuZero.Network.networksr
Ú mu_alpha_zero.MuZero.lazy_arraysr Úmu_alpha_zero.MuZero.utilsr r r r r r Úmu_alpha_zero.configr Úmu_alpha_zero.mem_bufferr r r Ú$mu_alpha_zero.shared_storage_managerr r rë rÁ r× rË r¥ r( r& ú<module>r sN ğØ € € € Ø € € € Ø € € € à € € € Ø )Ğ )Ğ )Ğ )Ğ )Ğ )à Ğ � Ğ %Ñ %Ô %Ğ %Ø "Ğ "Ğ "Ğ "Ğ "Ğ "à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø @Ğ @Ğ @Ğ @Ğ @Ğ @Ø ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø 6Ğ 6Ğ 6Ğ 6Ğ 6Ğ 6ğ-ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -à -Ğ -Ğ -Ğ -Ğ -Ğ -Ø QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QØ >Ğ >Ğ >Ğ >Ğ >Ğ >ğUoğ Uoğ Uoğ Uoğ Uo�zñ Uoô Uoğ Uoğpğ °c°k¸Tğ ğ ğ ğ ğ +/ğ.ğ .¨\ğ .À#ğ .ĞWdğ .Ø$'ğ.à ˜K 4ğ.ğ .ğ .ğ .ğ .ğ .r( | 20,530 | Python | .py | 75 | 272.533333 | 1,754 | 0.326408 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,536 | mz_node.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_MCTS/__pycache__/mz_node.cpython-311.pyc | §
¬Õ£f^ ã ó6 — d dl Z d dlmZ G d„ de¦ « ZdS )é N)Ú
AlphaZeroNodec óŒ ‡ — e Zd Zdˆ fd„ Zddeded ed
efd „Zd efd
„Zdd„Z dede
fd„Z ddeded ed
efd„Z d ed
edefd„Z
ˆ xZS )ÚMzAlphaZeroNoder Né c ó^ •— t ¦ « ||||¦ « d| _ d S )Nr )ÚsuperÚ__init__Úreward)ÚselfÚselect_probabilityÚparentÚtimes_visited_initÚcurrent_playerÚ __class__s €úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/MZ_MCTS/mz_node.pyr zMzAlphaZeroNode.__init__ s. ø€ İ
‰Œ×Ò˜ Ğ);¸VĞEWÑXÔXĞX؈Œ ˆ ˆ ó ç ø?éÄL Úmin_qÚmax_qÚgammaÚmultiple_playersc ó¾ — t d¦ « }d }d } | j ¦ « D ],\ }
} | ||||||¬¦ « } | |k r| }| }|
} Œ-|| fS )NÚinf)ÚcÚc2)ÚfloatÚchildrenÚitemsÚcalculate_utc_score)
r r r r r r r Úbest_utcÚ
best_childÚ best_actionÚactionÚchildÚ child_utcs
r Úget_best_childz MzAlphaZeroNode.get_best_child
sƒ € ݘ%‘L”L�=ˆØˆ
؈ Ø !œ]× 0Ò 0Ñ 2Ô 2ğ %ğ %‰MˆF�EØ ×1Ò1°%¸¸uĞEUĞYZĞ_aĞ1ÑbÔbˆIؘ8Ò#Ğ#Ø$�Ø "�
Ø $� øؘ;Ğ&Ğ&r Úprediction_forwardc ó" — || j ¦ « S ©N©Ústate)r r( s r Úget_value_predz MzAlphaZeroNode.get_value_pred s € Ø!Ğ! $¤*Ñ-Ô-Ğ-r Úreturnc ó¸ — | ¦ « | _ || _ t |¦ « D ])\ }}t || | j dz ¬¦ « }|| j |< Œ*d S )Néÿÿÿÿ)r r
r )Úcloner, r
Ú enumerater r r )r r, Úaction_probabilitiesÚ im_rewardr$ Ú probabilityÚnodes r Ú expand_nodezMzAlphaZeroNode.expand_node st € à—[’[‘]”]ˆŒ
Ø ˆŒ İ#,Ğ-AÑ#BÔ#Bğ )ğ )Ñ ˆF�Kİ"°kÈ$Ø26Ô2EÈÑ2LğNñ Nô NˆDà$(ˆDŒM˜&Ñ !Ğ !ğ )ğ )r Údynamics_forwardr$ c ó$ — || j |¦ « S r* r+ )r r8 r$ s r Úget_immediate_rewardz$MzAlphaZeroNode.get_immediate_reward" s € Ø Ğ ¤
¨FÑ3Ô3Ğ3r c ól — | ¦ « }|j dk r'|| j z t j |j dz ¦ « z S | ||||¦ « }|| j t j |j ¦ « d| j z z z |t j |j |z dz |z ¦ « z z z } | S )Nr g:Œ0â�yE>r )r
Ú
times_visitedr ÚmathÚsqrtÚscale_qÚlog)
r r r r r r r r
ÚqÚutcs
r r z#MzAlphaZeroNode.calculate_utc_score% s¿ € Ø—’‘ ” ˆØ Ô 1Ò $Ğ $Ø�tÔ.Ñ.µ´¸6Ô;OĞRVÑ;VÑ1WÔ1WÑWĞ WØ �LŠL˜ eĞ,<Ñ =Ô =ˆØ�$Ô)İ”˜6Ô/Ñ0Ô0°Q¸Ô9KÑ5KÑLñNà�$œ( FÔ$8¸2Ñ$=ÀÑ$AÀRÑ#GÑHÔHÑHñJñ Jˆğ ˆ
r c óÚ — | j |r| ¦ « n| ¦ « z }||k s&|t d¦ « k s|t d¦ « k r|S ||z
||z
z S )Nr z-inf)r
Úget_self_valuer )r r r r r rA s r r? zMzAlphaZeroNode.scale_q0 su € Ø ŒKĞ5EĞ`˜D× /Ò /Ñ 1Ô 1Ğ1Ğ1È4×K^ÒK^ÑK`ÔK`Ñ aˆØ �EŠ>ˆ>˜e¥u¨U¡|¤|Ò 3Ğ 3°uÅÀfÁ
Ä
Ò7MĞ7M؈HØ�E‘ ˜e e™mÑ,Ğ,r )r Nr r )r r )r. N)Ú__name__Ú
__module__Ú __qualname__r r Úboolr' Úcallabler- r7 Úintr: r r? Ú
__classcell__)r s @r r r s# ø€ € € € € ğğ ğ ğ ğ ğ ğ
'ğ
' Eğ
'°%ğ
'Àğ
'ĞX\ğ
'ğ
'ğ
'ğ
'ğ.°ğ .ğ .ğ .ğ .ğ)ğ )ğ )ğ )ğ4°Xğ 4Àsğ 4ğ 4ğ 4ğ 4ğ ğ ¨ğ °uğ ÀUğ Ğ]ağ ğ ğ ğ ğ-¨5ğ -ÀDğ -ÈUğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -r r )r= Ú$mu_alpha_zero.AlphaZero.MCTS.az_noder r © r r ú<module>rN sT ğØ € € € Ø >Ğ >Ğ >Ğ >Ğ >Ğ >ğ/-ğ /-ğ /-ğ /-ğ /-�mñ /-ô /-ğ /-ğ /-ğ /-r | 4,303 | Python | .py | 31 | 137.774194 | 655 | 0.357126 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,537 | java_manager.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/JavaGateway/__pycache__/java_manager.cpython-311.pyc | §
Ùxf¼ ã óˆ — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dlm Z m
Z
m Z d dl m
Z
d dlmZ d dlZ G d„ d¦ « ZdS )é N)Ú JavaGateway)Úget_python_homeÚfind_project_rootÚget_site_packages_path)Ú JavaNetworks)Ú MuZeroNetc óR — e Zd Zdedefd„Zdefd„Zd„ Zdedefd„Z d efd
„Z
d „ Z d S )
Ú JavaManagerÚnetworkÚargsc óH — t ¦ « t j d< t ¦ « | _ || _ |d | _ | |¦ « | _ | ¦ « | _
t ¦ « | _ t j | j ¦ « d S )NÚ
PYTHONHOMEÚenv_id)r ÚosÚenvironr Ú java_netsÚnetr Ú prepare_argsr Ú spawn_java_pÚ java_processr Ú java_gatewayÚatexitÚregisterÚkill_java_process)Úselfr r s ú\/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/JavaGateway/java_manager.pyÚ__init__zJavaManager.__init__ s� € õ $3Ñ#4Ô#4�Œ
�<Ñ å%™ œ ˆŒØˆŒØ˜8”nˆŒ Ø×%Ò% dÑ+Ô+ˆŒ Ø × -Ò -Ñ /Ô /ˆÔİ '™MœMˆÔİŒ˜Ô.Ñ/Ô/Ğ/Ğ/Ğ/ó c ó> — d„ | ¦ « D ¦ « S )Nc óÄ — i | ]]\ }}t |t j ¦ « r>t |t ¦ « s)|t d ¦ « k ¯G|t d¦ « k ¯Z||“Œ^S )Úinfz-inf)Ú
isinstanceÚnumbersÚNumberÚboolÚfloat)Ú.0ÚkÚvs r ú
<dictcomp>z,JavaManager.prepare_args.<locals>.<dictcomp> s| € ğ xğ xğ x™˜˜Aݘ1�gœnÑ-Ô-ğxİ6@ÀÅDÑ6IÔ6IğxØNOÕSXĞY^ÑS_ÔS_ÒN_ĞN_ĞdeÕinĞouÑivÔivÒdvĞdvğ �1ØdvĞdvĞdvr )Úitems)r r s r r zJavaManager.prepare_args s0 € ğxğ x §¢¡ ¤ ğ xñ xô xğ xr c óR — t j ddddt ¦ « › d�g¦ « S )NÚjavaz-jarzH/home/skyr/IdeaProjects/JSelfPlay/target/JSeflplayLinux-1.0-SNAPSHOT.jarz-Djava.library.path=z/jep/)Ú
subprocessÚPopenr ©r s r r zJavaManager.spawn_java_p s= € İÔ Ø
�VĞ gØ
CÕ$:Ñ$<Ô$<Ğ
CĞ
CĞ
Cğ
EñFô Fğ Fr Ún_jobsÚn_gamesc óT — g }| j | j ¦ « }t j t ¦ « › d�d¬¦ « | j j |||| | j
¦ « | j ¦ « }t d„ |D ¦ « ¦ « }g }|D ])}|
t j |¦ « ¦ « Œ*t j |d¬¦ « }t# |¦ « D �]\ } }
|
¦ « } d„ | ¦ « D ¦ « } t) |
¦ « ¦ « }
t) |
¦ « ¦ « ¦ « t) |
¦ « ¦ « ¦ « t) |
¦ « ¦ « ¦ « f}|
| |
||| f¦ « �Œt/ j t ¦ « › d�¦ « |S )Nz/Arrays/T)Úexist_okc ó6 — g | ]}| ¦ « ‘ŒS © )Ú getValue3)r' Úxs r ú
<listcomp>z;JavaManager.run_parallel_java_self_play.<locals>.<listcomp>+ s € Ğ4Ğ4Ğ4¨1˜Ÿš™ œ Ğ4Ğ4Ğ4r r )Úaxisc ó� — i | ]C}t | ¦ « ¦ « t | ¦ « ¦ « “ŒDS r6 )ÚintÚgetKeyr& ÚgetValue)r' Úentrys r r* z;JavaManager.run_parallel_java_self_play.<locals>.<dictcomp>2 s< € Ğ\Ğ\Ğ\À5•#�e—l’l‘n”nÑ%Ô%¥u¨U¯^ª^Ñ-=Ô-=Ñ'>Ô'>Ğ\Ğ\Ğ\r )r Úsaver r Úmakedirsr r Ú entry_pointÚrunParallelSelfPlayÚdict_to_java_mapr r ÚsetÚappendÚnpÚloadÚ concatenateÚ enumerateÚ getValue0ÚentrySetr& Ú getValue1Ú getValue2ÚshutilÚrmtree)r r1 r2 ÚresultsÚpathÚresÚ arr_pathsÚarrsÚarrÚiÚquartedÚhmapÚpdr) Úrmpred_rs r Úrun_parallel_java_self_playz'JavaManager.run_parallel_java_self_play% s÷ € ؈،~×"Ò" 4¤8Ñ,Ô,ˆİ
Œ Õ(Ñ*Ô*Ğ4Ğ4Ğ4¸tĞDÑDÔDĞDØÔ Ô+×?Ò?ÀÈĞQUĞW[×WlÒWlĞmqÔmvÑWwÔWwØ@DÄ ñMô MˆåĞ4Ğ4°Ğ4Ñ4Ô4Ñ5Ô5ˆ ØˆØ ğ 'ğ 'ˆDØ �KŠK�œ ™
œ
Ñ &Ô &Ğ &Ğ &İŒn˜T¨Ğ*Ñ*Ô*ˆİ" 3™ œ ğ 6ñ 6‰IˆAˆgØ×$Ò$Ñ&Ô&ˆDØ\Ğ\ÈDÏMÊMÉOÌOĞ\Ñ\Ô\ˆBİ�g×'Ò'Ñ)Ô)Ñ*Ô*ˆAİ ˜g× /Ò /Ñ 1Ô 1× ;Ò ;Ñ =Ô =Ñ>Ô>ÅÀg×FWÒFWÑFYÔFY×FcÒFcÑFeÔFeÑ@fÔ@fİ ˜g× /Ò /Ñ 1Ô 1× ;Ò ;Ñ =Ô =Ñ>Ô>ğ@ˆHà �NŠN˜B 8¨S°¬VĞ4Ñ 5Ô 5Ğ 5Ñ 5İŒ
Õ*Ñ,Ô,Ğ6Ğ6Ğ6Ñ7Ô7Ğ7؈r Údictc óĞ — | j j j j ¦ « }| ¦ « D ](\ }}| |t |¦ « ¦ « Œ)|S )N)r Újvmr- ÚutilÚHashMapr+ ÚputÚstr)r r] Újava_mapr( r) s r rD z JavaManager.dict_to_java_map: s\ € ØÔ$Ô(Ô-Ô2×:Ò:Ñ<Ô<ˆØ—J’J‘L”Lğ $ğ $‰DˆAˆqØ �LŠL˜�C ™FœFÑ #Ô #Ğ #Ğ #؈r c óˆ — | j ¦ « | j ¦ « t d¦ « d S )NzJava process ended.)r ÚkillÚwaitÚprintr0 s r r z JavaManager.kill_java_process@ s@ € Ø Ô× Ò Ñ Ô Ğ Ø Ô× Ò Ñ Ô Ğ İ
Ğ#Ñ$Ô$Ğ$؈r N) Ú__name__Ú
__module__Ú __qualname__r r] r r r r< r\ rD r r6 r r r
r
s´ € € € € € ğ 0 ğ 0°ğ 0ğ 0ğ 0ğ 0ğx ğ xğ xğ xğ xğFğ Fğ Fğ
°#ğ Àğ ğ ğ ğ ğ* Tğ ğ ğ ğ ğ ğ ğ ğ ğ r r
)r# r rO r. ÚnumpyrG Úpy4j.java_gatewayr Úmu_alpha_zero.General.utilsr r r Ú.mu_alpha_zero.MuZero.JavaGateway.java_networksr Ú%mu_alpha_zero.MuZero.Network.networksr r r
r6 r r ú<module>rq sÇ ğØ €€€Ø € € € Ø
€
€
€
Ø Ğ Ğ Ğ à Ğ Ğ Ğ Ø )Ğ )Ğ )Ğ )Ğ )Ğ )Ø bĞ bĞ bĞ bĞ bĞ bĞ bĞ bĞ bĞ bØ GĞ GĞ GĞ GĞ GĞ GØ ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø
€
€
€
ğ6ğ 6ğ 6ğ 6ğ 6ñ 6ô 6ğ 6ğ 6ğ 6r | 6,645 | Python | .py | 27 | 244.888889 | 1,309 | 0.34824 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,538 | java_networks.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/JavaGateway/__pycache__/java_networks.cpython-311.pyc | §
ÙxfA ã ó@ — d dl Z d dlmZ d dlmZ G d„ d¦ « ZdS )é N)Ú MuZeroNet)Úfind_project_rootc ó — e Zd Zdedefd„ZdS )Ú JavaNetworksÚnetworks_wrapperÚreturnc óR — t ¦ « › d�}| |¦ « |S )Nz /mz_net.pth)r Ú to_pickle)Úselfr Úpaths ú]/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/JavaGateway/java_networks.pyÚsavezJavaNetworks.save s/ € İ#Ñ%Ô%Ğ2Ğ2Ğ2ˆØ×"Ò" 4Ñ(Ô(Ğ(؈ ó N)Ú__name__Ú
__module__Ú __qualname__r Ústrr © r r
r r s6 € € € € € ğ Yğ °3ğ ğ ğ ğ ğ ğ r r )ÚmathÚ%mu_alpha_zero.MuZero.Network.networksr Úmu_alpha_zero.General.utilsr r r r r
ú<module>r sd ğØ € € € à ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø 9Ğ 9Ğ 9Ğ 9Ğ 9Ğ 9ğğ ğ ğ ğ ñ ô ğ ğ ğ r | 980 | Python | .py | 3 | 325.333333 | 606 | 0.431493 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,539 | arena.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_Arena/arena.py | import random
from typing import Type
import torch as th
from mu_alpha_zero.AlphaZero.Arena.players import Player, NetPlayer, RandomPlayer
from mu_alpha_zero.General.arena import GeneralArena
from mu_alpha_zero.General.mz_game import MuZeroGame
from mu_alpha_zero.Hooks.hook_manager import HookManager
from mu_alpha_zero.Hooks.hook_point import HookAt
from mu_alpha_zero.MuZero.utils import resize_obs, scale_state, scale_action
from mu_alpha_zero.config import MuZeroConfig
import wandb
from mu_alpha_zero.shared_storage_manager import SharedStorage
class MzArena(GeneralArena):
def __init__(self, game_manager: MuZeroGame, muzero_config: MuZeroConfig, device: th.device,
hook_manager: HookManager or None = None):
self.game_manager = game_manager
self.muzero_config = muzero_config
self.hook_manager = hook_manager if hook_manager is not None else HookManager()
self.device = device
def pit(self, player1: Type[Player], player2: Type[Player], num_games_to_play: int, num_mc_simulations: int,
one_player: bool = False, start_player: int = 1):
tau = self.muzero_config.arena_tau
rewards = {1: [], -1: []}
if one_player:
num_games_per_player = num_games_to_play
else:
num_games_per_player = num_games_to_play // 2
noop_num = random.randint(0, 30)
players = {"1": player1, "-1": player2}
for player in [1, -1]:
kwargs = {"num_simulations": num_mc_simulations, "current_player": player, "device": self.device,
"tau": tau, "unravel": False}
for game in range(num_games_per_player):
self.game_manager.reset()
state, _, _ = self.game_manager.frame_skip_step(self.game_manager.get_noop(), player,
frame_skip=noop_num)
state = resize_obs(state, self.muzero_config.target_resolution, self.muzero_config.resize_images)
state = scale_state(state, self.muzero_config.scale_state)
for step in range(self.muzero_config.num_steps):
self.game_manager.render()
move = players[str(player)].choose_move(state, **kwargs)
state, reward, done = self.game_manager.frame_skip_step(move, player)
state = resize_obs(state, self.muzero_config.target_resolution, self.muzero_config.resize_images)
state = scale_state(state, self.muzero_config.scale_state)
try:
players[str(player)].monte_carlo_tree_search.buffer.add_frame(state, scale_action(move,
self.game_manager.get_num_actions()),
player)
except AttributeError:
# Player is probably not net player and doesn't have monte_carlo_tree_search.
pass
rewards[player].append(reward)
if done:
break
self.hook_manager.process_hook_executes(self, self.pit.__name__, __file__, HookAt.TAIL, (rewards, kwargs))
return sum(rewards[1]), sum(rewards[-1]), 0
def run_on_training_end(self):
self.hook_manager.process_hook_executes(self, self.run_on_training_end.__name__, __file__, HookAt.ALL)
| 3,551 | Python | .py | 58 | 45.62069 | 143 | 0.590635 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,540 | arena.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_Arena/__pycache__/arena.cpython-311.pyc | §
‹fß
ã ó¶ — d dl Z d dlmZ d dlZd dlmZmZmZ d dl m
Z
d dl m Z d dl
mZ d dlmZ d dlmZmZmZ d d lmZ d dlZd d
lmZ G d „ d e
¦ « ZdS )
é N)ÚType)ÚPlayerÚ NetPlayerÚ RandomPlayer)Ú GeneralArena)Ú
MuZeroGame)Ú HookManager)ÚHookAt)Ú
resize_obsÚ scale_stateÚ scale_action)Ú MuZeroConfig)Ú
SharedStoragec
óz — e Zd Z ddededej depdfd„Z dd e e
d
e e
d e d e d
e de f d„Z
d„ ZdS )ÚMzArenaNÚ game_managerÚ
muzero_configÚdeviceÚ hook_managerc ó^ — || _ || _ |�|n
t ¦ « | _ || _ d S ©N)r r r r r )Úselfr r r r s úR/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/MZ_Arena/arena.pyÚ__init__zMzArena.__init__ s2 € à (ˆÔØ *ˆÔØ,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ ˆŒ ˆ ˆ ó Fé Úplayer1Úplayer2Únum_games_to_playÚnum_mc_simulationsÚ
one_playerÚ start_playerc ó$ — | j j }g g dœ}|r|} n|dz } t j dd¦ « }
||dœ} dD �]ô} || | j |ddœ}
t | ¦ « D �]Ó}| j ¦ « | j | j ¦ « | |
¬¦ « \ }}}t || j j | j j ¦ « }t || j j
¦ « }t | j j ¦ « D �]%}| j ¦ « | t! | ¦ « j |fi |
¤�}| j || ¦ « \ }}}t || j j | j j ¦ « }t || j j
¦ « } | t! | ¦ « j j |t+ || j ¦ « ¦ « | ¦ « n# t. $ r Y nw xY w|| |¦ « |r n�Œ'�ŒÕ�Œö| j | | j j t: t< j ||
f¦ « tA |d ¦ « tA |d
¦ « dfS ) N)r éÿÿÿÿé r é )Ú1z-1F)Únum_simulationsÚcurrent_playerr ÚtauÚunravel)Ú
frame_skipr r$ )!r Ú arena_tauÚrandomÚrandintr Úranger ÚresetÚframe_skip_stepÚget_noopr Útarget_resolutionÚ
resize_imagesr Ú num_stepsÚrenderÚstrÚ choose_moveÚmonte_carlo_tree_searchÚbufferÚ add_framer
Úget_num_actionsÚAttributeErrorÚappendr Úprocess_hook_executesÚpitÚ__name__Ú__file__r
ÚTAILÚsum)r r r r r r! r" r* ÚrewardsÚnum_games_per_playerÚnoop_numÚplayersÚplayerÚkwargsÚgameÚstateÚ_ÚstepÚmoveÚrewardÚdones r rA z MzArena.pit sß € àÔ Ô*ˆØ˜b�/�/ˆØ ğ :Ø#4Ğ Ğ à#4¸Ñ#9Ğ İ”> ! RÑ(Ô(ˆØ wĞ/Ğ/ˆØ ğ ñ ˆFØ);ÈvĞaeÔalØ ¨Uğ4ğ 4ˆFå Ğ 2Ñ3Ô3ğ
ñ
�ØÔ!×'Ò'Ñ)Ô)Ğ)à "Ô /× ?Ò ?ÀÔ@Q×@ZÒ@ZÑ@\Ô@\Ğ^dØKSğ @ñ Uô U‘ ��q˜!å" 5¨$Ô*<Ô*NĞPTÔPbÔPpÑqÔq�İ# E¨4Ô+=Ô+IÑJÔJ�İ ! $Ô"4Ô">Ñ ?Ô ?ğ ñ �DØÔ%×,Ò,Ñ.Ô.Ğ.Ø;˜7¥3 v¡;¤;Ô/Ô;¸EĞLĞLÀVĞLĞL�Dà*.Ô*;×*KÒ*KÈDĞRXÑ*YÔ*YÑ'�E˜6 4İ & u¨dÔ.@Ô.RĞTXÔTfÔTtÑ uÔ u�Eİ '¨¨tÔ/AÔ/MÑ NÔ N�Eğ Ø ¥ F¡ ¤ Ô,ÔDÔK×UÒUĞV[Õ]iĞjnØjnÔj{÷ kLò kLñ kNô kNñ^Oô ^OàV\ñ^ô ^ğ ^ğ ^øõ *ğ ğ ğ à ˜ğ øøøğ ˜F”O×*Ò*¨6Ñ2Ô2Ğ2Øğ Ø ˜ñ ùñ-
ğ2
Ô×/Ò/°°d´hÔ6GÍÕSYÔS^ĞahĞjpĞ`qÑrÔrĞrİ�7˜1”:‰Œ¥ G¨B¤KÑ 0Ô 0°!Ğ3Ğ3s Æ AG9Ç9
H ÈH c ór — | j | | j j t t
j ¦ « d S r )r r@ Úrun_on_training_endrB rC r
ÚALL)r s r rT zMzArena.run_on_training_endC s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr r )Fr )rB Ú
__module__Ú __qualname__r r Úthr r r r r ÚintÚboolrA rT © r r r r s¿ € € € € € à59ğ ğ Zğ À ğ ĞVXÔV_ğ Ø *Ğ 2¨dğ ğ ğ ğ ğ ;<ğ'4ğ '4˜4 œ<ğ '4°$°v´,ğ '4ĞSVğ '4Ğloğ '4Ø ğ'4Ø47ğ'4ğ '4ğ '4ğ '4ğRoğ oğ oğ oğ or r )r. Útypingr ÚtorchrX Ú%mu_alpha_zero.AlphaZero.Arena.playersr r r Úmu_alpha_zero.General.arenar Ú mu_alpha_zero.General.mz_gamer Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr
Úmu_alpha_zero.MuZero.utilsr r r
Úmu_alpha_zero.configr ÚwandbÚ$mu_alpha_zero.shared_storage_managerr r r[ r r ú<module>rg s ğØ
€
€
€
Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ à QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QØ 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø LĞ LĞ LĞ LĞ LĞ LĞ LĞ LĞ LĞ LØ -Ğ -Ğ -Ğ -Ğ -Ğ -Ø € € € à >Ğ >Ğ >Ğ >Ğ >Ğ >ğ2oğ 2oğ 2oğ 2oğ 2oˆlñ 2oô 2oğ 2oğ 2oğ 2or | 5,155 | Python | .py | 22 | 233.272727 | 971 | 0.387807 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,541 | networks.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/Network/networks.py | import time
from typing import Tuple, List, Any
import numpy as np
import torch as th
import torch.nn.functional as F
import wandb
from torch import nn, Tensor
from torch.nn.functional import mse_loss
from mu_alpha_zero.AlphaZero.Network.nnet import AlphaZeroNet as PredictionNet, OriginalAlphaZeroNetwork
from mu_alpha_zero.AlphaZero.checkpointer import CheckPointer
from mu_alpha_zero.AlphaZero.logger import Logger
from mu_alpha_zero.General.memory import GeneralMemoryBuffer
from mu_alpha_zero.General.mz_game import MuZeroGame
from mu_alpha_zero.General.network import GeneralMuZeroNetwork
from mu_alpha_zero.Hooks.hook_manager import HookManager
from mu_alpha_zero.Hooks.hook_point import HookAt
from mu_alpha_zero.MuZero.utils import match_action_with_obs_batch, scalar_to_support, support_to_scalar, \
scale_hidden_state
from mu_alpha_zero.config import MuZeroConfig
from mu_alpha_zero.shared_storage_manager import SharedStorage
class MuZeroNet(th.nn.Module, GeneralMuZeroNetwork):
def __init__(self, input_channels: int, dropout: float, action_size: int, num_channels: int, latent_size: list[int],
num_out_channels: int, linear_input_size: int or list[int], rep_input_channels: int,
use_original: bool, support_size: int, num_blocks: int,
state_linear_layers: int, pi_linear_layers: int, v_linear_layers: int, linear_head_hidden_size: int,
is_atari: bool,
hook_manager: HookManager or None = None, use_pooling: bool = True):
super(MuZeroNet, self).__init__()
self.input_channels = input_channels
self.rep_input_channels = rep_input_channels
self.dropout = dropout
self.use_original = use_original
self.use_pooling = use_pooling
self.device = th.device("cuda" if th.cuda.is_available() else "cpu")
self.action_size = action_size
self.num_channels = num_channels
self.latent_size = latent_size
self.optimizer = None
self.scheduler = None
self.num_out_channels = num_out_channels
self.linear_input_size = linear_input_size
self.support_size = support_size
self.is_atari = is_atari
self.num_blocks = num_blocks
self.state_linear_layers = state_linear_layers
self.pi_linear_layers = pi_linear_layers
self.v_linear_layers = v_linear_layers
self.linear_head_hidden_size = linear_head_hidden_size
self.hook_manager = hook_manager if hook_manager is not None else HookManager()
# self.action_embedding = th.nn.Embedding(action_size, 256)
if not is_atari:
self.representation_network = OriginalAlphaZeroNetwork(in_channels=rep_input_channels,
num_channels=num_out_channels,
dropout=dropout,
action_size=action_size,
linear_input_size=linear_input_size,
state_linear_layers=state_linear_layers,
pi_linear_layers=pi_linear_layers,
v_linear_layers=v_linear_layers,
linear_head_hidden_size=linear_head_hidden_size,
is_atari=is_atari,
support_size=support_size, latent_size=latent_size,
num_blocks=num_blocks, muzero=True,
is_dynamics=False,
is_representation=True)
else:
self.representation_network = RepresentationNet(rep_input_channels, use_pooling=use_pooling)
if use_original:
self.dynamics_network = OriginalAlphaZeroNetwork(in_channels=num_channels + 1,
num_channels=num_out_channels,
dropout=dropout,
action_size=action_size,
linear_input_size=linear_input_size,
state_linear_layers=state_linear_layers,
pi_linear_layers=pi_linear_layers,
v_linear_layers=v_linear_layers,
linear_head_hidden_size=linear_head_hidden_size,
is_atari=is_atari,
support_size=support_size, latent_size=latent_size,
num_blocks=num_blocks, muzero=True, is_dynamics=True)
self.prediction_network = OriginalAlphaZeroNetwork(in_channels=num_channels, num_channels=num_out_channels,
dropout=dropout,
action_size=action_size,
state_linear_layers=state_linear_layers,
pi_linear_layers=pi_linear_layers,
v_linear_layers=v_linear_layers,
linear_head_hidden_size=linear_head_hidden_size,
is_atari=is_atari,
linear_input_size=linear_input_size,
support_size=support_size, latent_size=latent_size,
num_blocks=num_blocks, muzero=True, is_dynamics=False)
else:
self.dynamics_network = DynamicsNet(in_channels=257, num_channels=num_channels, dropout=dropout,
latent_size=latent_size, out_channels=num_out_channels)
# prediction outputs 6x6 latent state
self.prediction_network = PredictionNet(in_channels=256, num_channels=num_channels, dropout=dropout,
action_size=action_size, linear_input_size=linear_input_size)
@classmethod
def make_from_config(cls, config: MuZeroConfig, hook_manager: HookManager or None = None):
return cls(config.num_net_in_channels, config.net_dropout, config.net_action_size, config.num_net_channels,
config.net_latent_size, config.num_net_out_channels, config.az_net_linear_input_size,
config.rep_input_channels, config.use_original, config.support_size, config.num_blocks,
config.state_linear_layers, config.pi_linear_layers, config.v_linear_layers,
config.linear_head_hidden_size, config.is_atari,
hook_manager=hook_manager, use_pooling=config.use_pooling)
def dynamics_forward(self, x: th.Tensor, predict: bool = False, return_support: bool = False,
convert_to_state: bool = True):
if predict:
state, r = self.dynamics_network.forward(x, muzero=True)
reward = r.detach().cpu().numpy()
else:
state, reward = self.dynamics_network(x, muzero=True, return_support=return_support)
if convert_to_state:
try:
state = state.view(self.num_out_channels, self.latent_size[0], self.latent_size[1])
except RuntimeError:
# The state is batched
state = state.view(-1, self.num_out_channels, self.latent_size[0], self.latent_size[1])
return state, reward
def prediction_forward(self, x: th.Tensor, predict: bool = False, return_support: bool = False):
if predict:
pi, v = self.prediction_network.predict(x, muzero=True)
return pi, v
pi, v = self.prediction_network(x, muzero=True, return_support=return_support)
return pi, v
def representation_forward(self, x: th.Tensor):
state = self.representation_network(x, muzero=True)
try:
state = state.view(self.num_out_channels, self.latent_size[0], self.latent_size[1])
except RuntimeError:
# The state is batched
state = state.view(-1, self.num_out_channels, self.latent_size[0], self.latent_size[1])
return state
def forward_recurrent(self, hidden_state_with_action: th.Tensor, all_predict: bool, return_support: bool = False):
next_state, reward = self.dynamics_forward(hidden_state_with_action, predict=all_predict,
return_support=return_support)
pi, v = self.prediction_forward(next_state, predict=all_predict, return_support=return_support)
return next_state, reward, pi, v
def make_fresh_instance(self):
return MuZeroNet(self.input_channels, self.dropout, self.action_size, self.num_channels, self.latent_size,
self.num_out_channels, self.linear_input_size, self.rep_input_channels,
hook_manager=self.hook_manager, use_original=self.use_original, support_size=self.support_size,
num_blocks=self.num_blocks, use_pooling=self.use_pooling,
state_linear_layers=self.state_linear_layers,
pi_linear_layers=self.pi_linear_layers, v_linear_layers=self.v_linear_layers,
linear_head_hidden_size=self.linear_head_hidden_size, is_atari=self.is_atari)
def train_net(self, memory_buffer: GeneralMemoryBuffer, muzero_config: MuZeroConfig) -> tuple[float, list[float]]:
if memory_buffer.train_length() <= 1:
return 0, []
device = th.device("cuda" if th.cuda.is_available() else "cpu")
if self.optimizer is None:
self.optimizer = th.optim.Adam(self.parameters(), lr=muzero_config.lr,
weight_decay=muzero_config.l2)
if self.scheduler is None and muzero_config.lr_scheduler is not None:
self.scheduler = muzero_config.lr_scheduler(self.optimizer, **muzero_config.lr_scheduler_kwargs)
losses = []
iteration = 0
loader = lambda: memory_buffer.batch_with_priorities(muzero_config.enable_per,
muzero_config.batch_size, muzero_config)
for epoch in range(muzero_config.epochs):
sampled_game_data, grad_scales, priorities, weights = loader()
if len(sampled_game_data) <= 1:
continue
loss, loss_v, loss_pi, loss_r = self.calculate_losses(sampled_game_data, grad_scales, weights, device,
muzero_config)
wandb.log({"combined_loss": loss.item(), "loss_v": loss_v.item(), "loss_pi": loss_pi.item(),
"loss_r": loss_r.item()})
losses.append(loss.item())
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
if self.scheduler is not None:
self.scheduler.step()
try:
wandb.log({"lr": self.scheduler.get_last_lr()[0]})
except:
pass
self.hook_manager.process_hook_executes(self, self.train_net.__name__, __file__, HookAt.MIDDLE, args=(
sampled_game_data, loss.item(), loss_v, loss_pi, loss_r,
iteration))
iteration += 1
self.hook_manager.process_hook_executes(self, self.train_net.__name__, __file__, HookAt.TAIL,
args=(memory_buffer, losses))
return sum(losses) / len(losses), losses
def eval_net(self, memory_buffer: GeneralMemoryBuffer, muzero_config: MuZeroConfig) -> None:
if memory_buffer.eval_length() == 0:
return
device = th.device("cuda" if th.cuda.is_available() else "cpu")
if self.optimizer is None:
self.optimizer = th.optim.Adam(self.parameters(), lr=muzero_config.lr,
weight_decay=muzero_config.l2)
loader = lambda: memory_buffer.batch_with_priorities(muzero_config.enable_per,
muzero_config.batch_size,
muzero_config,
is_eval=True)
for epoch in range(muzero_config.eval_epochs):
experience_batch, grad_scales, priorities, weights = loader()
if len(experience_batch) <= 1:
continue
loss, loss_v, loss_pi, loss_r = self.calculate_losses(experience_batch, grad_scales, weights, device,
muzero_config)
wandb.log({"eval_combined_loss": loss.item(), "eval_loss_v": loss_v.item(), "eval_loss_pi": loss_pi.item(),
"eval_loss_r": loss_r.item()})
def calculate_losses(self, experience_batch, grad_scales, weights, device, muzero_config):
init_states, rewards, scalar_values, moves, pis, masks = self.get_batch_for_unroll_index(0, experience_batch,
device)
loss_fn = muzero_config._value_reward_loss
# rewards = scalar_to_support(rewards, muzero_config.support_size)
if muzero_config.loss_gets_support:
values = scalar_to_support(scalar_values, muzero_config.support_size, muzero_config.is_atari)
else:
values = scalar_values
hidden_state = self.representation_forward(init_states)
if muzero_config.scale_hidden_state:
hidden_state = scale_hidden_state(hidden_state)
pred_pis, pred_vs = self.prediction_forward(hidden_state, return_support=muzero_config.loss_gets_support)
pi_loss, v_loss, r_loss = 0, 0, 0
pi_loss += self.muzero_loss(pred_pis, pis,masks=masks)
v_loss += loss_fn(pred_vs, values)
new_priorities = [[] for x in range(pred_pis.size(0))]
grad_scales = [[grad_scales[x][i] for x in range(len(grad_scales))] for i in range(len(grad_scales[0]))]
if muzero_config.enable_per:
self.populate_priorities((th.abs(support_to_scalar(pred_vs,
muzero_config.support_size,
muzero_config.is_atari) - scalar_values) ** muzero_config.alpha).reshape(
-1).tolist(), new_priorities)
for i in range(1, muzero_config.K + 1):
_, rewards, scalar_values, moves, pis, masks = self.get_batch_for_unroll_index(i, experience_batch,
device)
if muzero_config.loss_gets_support:
rewards = scalar_to_support(rewards, muzero_config.support_size, muzero_config.is_atari)
values = scalar_to_support(scalar_values, muzero_config.support_size, muzero_config.is_atari)
else:
values = scalar_values
hidden_state, pred_rs, pred_pis, pred_vs = self.forward_recurrent(
match_action_with_obs_batch(hidden_state, moves, muzero_config), False,
return_support=muzero_config.loss_gets_support)
if muzero_config.scale_hidden_state:
hidden_state = scale_hidden_state(hidden_state)
hidden_state.register_hook(lambda grad: grad * 0.5)
current_pi_loss = self.muzero_loss(pred_pis, pis,masks=masks)
current_v_loss = loss_fn(pred_vs, values)
current_r_loss = loss_fn(pred_rs, rewards)
current_r_loss.register_hook(
lambda grad: grad * (1 / th.tensor(grad_scales[i], device=device)).reshape(grad.shape))
current_v_loss.register_hook(
lambda grad: grad * (1 / th.tensor(grad_scales[i], device=device)).reshape(grad.shape))
current_pi_loss.register_hook(
lambda grad: grad * (1 / th.tensor(grad_scales[i], device=device)).reshape(grad.shape))
pi_loss += current_pi_loss
v_loss += current_v_loss
r_loss += current_r_loss
if muzero_config.enable_per:
self.populate_priorities((th.abs(support_to_scalar(pred_vs,
muzero_config.support_size,
muzero_config.is_atari) - scalar_values) ** muzero_config.alpha).reshape(
-1).tolist(), new_priorities)
# TODO: Multiply v by 0.25 when reanalyze implemented.
v_loss *= 0.25
loss = pi_loss + v_loss + r_loss
if muzero_config.enable_per:
loss *= th.tensor(weights, dtype=loss.dtype, device=loss.device)
loss = loss.mean()
if muzero_config.enable_per:
self.update_priorities(new_priorities, experience_batch)
return loss, v_loss.mean(), pi_loss.mean(), r_loss.mean()
def get_batch_for_unroll_index(self, index: int, experience_batch, device) -> tuple[
Tensor | None, Tensor, Tensor, list[Any], Tensor, Tensor]:
tensor_from_x = lambda x: th.tensor(x, dtype=th.float32, device=device)
init_states = None
if index == 0:
init_states = [np.array(x.datapoints[index].frame) for x in experience_batch]
init_states = tensor_from_x(np.array(init_states)).permute(0, 3, 1, 2)
move_index = max(0, index - 1)
rewards = np.array([x.datapoints[index].reward for x in experience_batch])
rewards = tensor_from_x(rewards)
values = np.array([x.datapoints[index].v for x in experience_batch])
values = tensor_from_x(values)
moves = [x.datapoints[move_index].move for x in experience_batch]
# moves = np.array([x.datapoints[index].move for x in experience_batch])
# moves = tensor_from_x(moves)
pis = np.array([x.datapoints[index].pi for x in experience_batch])
pis = tensor_from_x(pis)
masks = np.array([x.datapoints[index].action_mask for x in experience_batch])
masks = tensor_from_x(masks)
return init_states, rewards.unsqueeze(1), values.unsqueeze(1), moves, pis, masks
def populate_priorities(self, new_priorities: list, priorities_list: list):
for idx, priority in enumerate(new_priorities):
priorities_list[idx].append(priority)
def update_priorities(self, new_priorities: list, experience_batch: list):
for idx, game in enumerate(experience_batch):
for i in range(len(game.datapoints)):
game.datapoints[i].priority = new_priorities[idx][i]
def muzero_loss(self, y_hat, y,masks: th.Tensor or None = None) -> th.Tensor:
# if masks is not None:
# y_hat = y_hat * masks.reshape(y_hat.shape)
return -th.sum(y * y_hat, dim=1).unsqueeze(1)
def continuous_weight_update(self, shared_storage: SharedStorage, muzero_config: MuZeroConfig,
checkpointer: CheckPointer, logger: Logger):
wandb.init(project=muzero_config.wandbd_project_name, name="Continuous Weight Update")
self.train()
# losses = []
# loss_avgs = []
# num_epochs = muzero_config.epochs
muzero_config.epochs = 1
muzero_config.eval_epochs = 1
# while len(
# shared_storage.get_buffer()) < muzero_config.batch_size // 4: # await reasonable buffer size
# time.sleep(5)
logger.log("Finished waiting for target buffer size,starting training.")
for iter_ in range(muzero_config.num_worker_iters):
# if not shared_storage.get_was_pitted():
# time.sleep(5)
# continue
# self.load_state_dict(shared_storage.get_stable_network_params())
# for epoch in range(num_epochs):
avg, iter_losses = self.train_net(shared_storage, muzero_config)
shared_storage.set_experimental_network_params(self.state_dict())
# shared_storage.set_optimizer(self.optimizer.state_dict())
# loss_avgs.append(avg)
# losses.extend(iter_losses)
# shared_storage.set_was_pitted(False)
if iter_ % muzero_config.eval_interval == 0:
self.eval_net(shared_storage, muzero_config)
if iter_ % 250 == 0 and iter_ != 0:
logger.log(f"Saving checkpoint at iteration {iter_}.")
checkpointer.save_checkpoint(self, self, self.optimizer, muzero_config.lr, iter_, muzero_config)
# wandb.log({"loss_avg": avg})
def to_pickle(self, path: str):
th.save(self, path)
def run_on_training_end(self):
self.hook_manager.process_hook_executes(self, self.run_on_training_end.__name__, __file__, HookAt.ALL)
class RepresentationNet(th.nn.Module):
def __init__(self, rep_input_channels: int, use_pooling: bool = True):
super(RepresentationNet, self).__init__()
self.device = th.device("cuda" if th.cuda.is_available() else "cpu")
self.conv1 = th.nn.Conv2d(in_channels=rep_input_channels, out_channels=128, kernel_size=3, stride=2, padding=1)
self.residuals1 = th.nn.ModuleList([ResidualBlock(128) for _ in range(2)])
self.conv2 = th.nn.Conv2d(in_channels=128, out_channels=256, kernel_size=3, stride=2, padding=1)
self.residuals2 = th.nn.ModuleList([ResidualBlock(256) for _ in range(3)])
self.residuals3 = th.nn.ModuleList([ResidualBlock(256) for _ in range(3)])
if use_pooling:
self.pool1 = th.nn.AvgPool2d(kernel_size=3, stride=2, padding=1)
self.pool2 = th.nn.AvgPool2d(kernel_size=3, stride=2, padding=1)
else:
self.pool1 = th.nn.Identity()
self.pool2 = th.nn.Identity()
self.relu = th.nn.ReLU()
def forward(self, x: th.Tensor, muzero: bool = False):
# x.unsqueeze(0)
x = x.to(self.device)
x = self.relu(self.conv1(x))
for residual in self.residuals1:
x = residual(x)
x = self.relu(self.conv2(x))
for residual in self.residuals2:
x = residual(x)
x = self.pool1(x)
for residual in self.residuals3:
x = residual(x)
x = self.pool2(x)
return x
def trace(self):
data = th.rand((1, 128, 8, 8))
traced_script_module = th.jit.trace(self, data)
return traced_script_module
class DynamicsNet(nn.Module):
def __init__(self, in_channels, num_channels, dropout, latent_size, out_channels):
super(DynamicsNet, self).__init__()
self.out_channels = out_channels
self.latent_size = latent_size
# Convolutional layers
self.conv1 = nn.Conv2d(in_channels, num_channels, 3, padding=1)
self.bn1 = nn.BatchNorm2d(num_channels)
self.conv2 = nn.Conv2d(num_channels, num_channels, 3, padding=1)
self.bn2 = nn.BatchNorm2d(num_channels)
self.conv3 = nn.Conv2d(num_channels, num_channels, 3, padding=1)
self.bn3 = nn.BatchNorm2d(num_channels)
self.conv4 = nn.Conv2d(num_channels, num_channels, 3)
self.bn4 = nn.BatchNorm2d(num_channels)
# Fully connected layers
# 4608 (5x5) or 512 (3x3) or 32768 (10x10) or 18432 (8x8)
self.fc1 = nn.Linear(8192, 1024)
self.fc1_bn = nn.BatchNorm1d(1024)
self.fc2 = nn.Linear(1024, 512)
self.fc2_bn = nn.BatchNorm1d(512)
self.dropout = nn.Dropout(dropout)
# Output layers
self.state_head = nn.Linear(512, latent_size[0] * latent_size[1] * out_channels) # state head
self.reward_head = nn.Linear(512, 1) # reward head
def forward(self, x, muzero=False):
# x = x.unsqueeze(0)
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = F.relu(self.bn3(self.conv3(x)))
x = F.relu(self.bn4(self.conv4(x)))
x = x.view(x.size(0), -1) # Flatten
x = F.relu(self.fc1_bn(self.fc1(x)))
x = self.dropout(x)
x = F.relu(self.fc2_bn(self.fc2(x)))
x = self.dropout(x)
state = self.state_head(x)
r = self.reward_head(x)
return state, r
def trace(self) -> th.jit.ScriptFunction:
data = th.rand((1, 257, 6, 6)).to("cuda:0")
traced_script_module = th.jit.trace(self, data)
return traced_script_module
@th.no_grad()
def predict(self, x):
state, r = self.forward(x)
state = state.view(self.out_channels, self.latent_size[0], self.latent_size[1])
return state, r.detach().cpu().numpy()
class ResidualBlock(th.nn.Module):
def __init__(self, channels: int):
super(ResidualBlock, self).__init__()
self.convolution1 = th.nn.Conv2d(in_channels=channels, out_channels=channels, kernel_size=3, padding=1)
self.bnorm1 = th.nn.BatchNorm2d(channels)
self.convolution2 = th.nn.Conv2d(in_channels=channels, out_channels=channels, kernel_size=3, padding=1)
self.bnorm2 = th.nn.BatchNorm2d(channels)
self.relu = th.nn.ReLU()
def forward(self, x):
# x = x.unsqueeze(0)
x_res = x
convolved = self.convolution1(x)
x = self.relu(self.bnorm1(convolved))
x = self.bnorm2(self.convolution2(x))
x += x_res
x = self.relu(x)
return x
| 26,509 | Python | .py | 430 | 44.613953 | 140 | 0.567582 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,542 | networks.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/Network/__pycache__/networks.cpython-311.pyc | §
k¤føU ã ó˜ — d dl Z d dlZd dlZd dlmc mZ d dl Z d dlmZ d dlm
Z
d dl m Z
mZ d dlmZ d dlmZ d dlmZ d dlmZ d d lmZ d d
lmZ d d lm Z d d l m Z m Z m Z d d
l!m"Z" d dl#m$Z$ G d„ dej j% e¦ « Z& G d„ dej j% ¦ « Z' G d„ dej% ¦ « Z( G d„ dej j% ¦ « Z)dS )é N)Únn)Úmse_loss)Ú AlphaZeroNetÚOriginalAlphaZerNetwork)Ú CheckPointer)ÚLogger)ÚGeneralMemoryBuffer)Ú
MuZeroGame)ÚGeneralMuZeroNetwork)Ú HookManager)ÚHookAt)Úmatch_action_with_obs_batchÚscalar_to_supportÚsupport_to_scalar)Ú MuZeroConfig)Ú
SharedStoragec # ó ‡ — e Zd Z d8dededededee ded epee d
ed ed ed
edededededepddef"ˆ fd„
Ze d9de
depdfd„¦ « Z d:de j
dededefd „Zd;de j
dedefd „Zde j
fd „Zd<d e j
d edefd!„Zd"„ Zd#ed$e
d%eeee f fd&„Zd#ed$e
d%dfd'„Zd(„ Zd)ed%ee j
e j
e j
ee j
f fd*„Zd+ed,efd-„Zd+ed.efd/„Zd0„ Zd1e d$e
d2e d3e fd4„Z d5e fd6„Z!d7„ Z"ˆ xZ#S )=Ú MuZeroNetNTÚinput_channelsÚdropoutÚ action_sizeÚ num_channelsÚ latent_sizeÚnum_out_channelsÚlinear_input_sizeÚrep_input_channelsÚ use_originalÚ support_sizeÚ
num_blocksÚstate_linear_layersÚpi_linear_layersÚv_linear_layersÚlinear_head_hidden_sizeÚ hook_managerÚ use_poolingc ó •— t t | ¦ « ¦ « || _ || _ || _ | | _ || _ t j t j
¦ « rdnd¦ « | _ || _ || _
|| _ d | _ || _ || _ |
| _ | | _ | | _ |
| _ || _ || _ |�|n
t1 ¦ « | _ t5 ||¬¦ « | _ | rFt9 d||||| |
|||
|| dd¬¦ « | _ t9 d|||| |
||||
|| dd¬ ¦ « | _ d S t? d||||¬
¦ « | _ tA d||||¬ ¦ « | _ d S ) NÚcudaÚcpu)r% é T)Ú in_channelsr r r r r r! r" r# r r r ÚmuzeroÚ is_dynamicsé F)r* r r r r r! r" r# r r r r r+ r, )r* r r r Ú out_channels)r* r r r r )!Úsuperr Ú__init__r r r r r% ÚthÚdevicer' Ú is_availabler r r Ú optimizerr r r r r r! r" r# r r$ ÚRepresentationNetÚrepresentation_networkr Údynamics_networkÚprediction_networkÚ DynamicsNetÚ
PredictionNet)Úselfr r r r r r r r r r r r r! r" r# r$ r% Ú __class__s €úT/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/Network/networks.pyr0 zMuZeroNet.__init__ sê ø€ õ
�i˜Ñ Ô ×'Ò'Ñ)Ô)Ğ)Ø ,ˆÔØ"4ˆÔ Ø ˆŒ Ø (ˆÔØ&ˆÔİ”i"¬'×*>Ò*>Ñ*@Ô*@Ğ K ÀeÑLÔLˆŒ Ø&ˆÔØ (ˆÔØ&ˆÔØ ˆŒØ 0ˆÔ Ø!2ˆÔ Ø (ˆÔØ$ˆŒØ#6ˆÔ Ø 0ˆÔ Ø .ˆÔ Ø'>ˆÔ$Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔå&7Ğ8JĞXcĞ&dÑ&dÔ&dˆÔ#Ø ğ rİ$;ÈĞZjØDKØHSØN_ØPcØM]ØL[ØTkØIUĞcnØGQĞZ^Ğlpğ %rñ %rô %rˆDÔ !õ '>È#Ğ\lØFMØJUØReØO_ØN]ØVmØPaØKWĞepØISĞ\`Ğnsğ 'uñ 'uô 'uˆDÔ #Ğ #Ğ #õ %0¸CÈlĞdkØ<GĞVfğ%hñ %hô %hˆDÔ !õ '4ÀĞR^ĞhoØ@KĞ_pğ'rñ 'rô 'rˆDÔ #Ğ #Ğ #ó Úconfigc óÚ — | |j |j |j |j |j |j |j |j |j |j |j
|j |j |j
|j ||j ¬¦ « S )N)r$ r% )Únum_net_in_channelsÚ net_dropoutÚnet_action_sizeÚnum_net_channelsÚnet_latent_sizeÚnum_net_out_channelsÚaz_net_linear_input_sizer r r r r r! r" r# r% )Úclsr? r$ s r= Úmake_from_configzMuZeroNet.make_from_configO sx € àˆs�6Ô-¨vÔ/AÀ6ÔCYĞ[aÔ[rØÔ)¨6Ô+FÈÔHgØÔ,¨fÔ.AÀ6ÔCVĞX^ÔXiØÔ-¨vÔ/FÈÔH^ØÔ1Ø ,¸&Ô:Lğ Nñ Nô Nğ Nr> FÚxÚpredictÚreturn_supportÚconvert_to_statec óæ — |rX| j |d¬¦ « \ }}| ¦ « ¦ « ¦ « }n| |d|¬¦ « \ }}|rw | | j | j d | j d ¦ « }nC# t $ r6 | d| j | j d | j d ¦ « }Y nw xY w||fS )NT©r+ ©r+ rL r é éÿÿÿÿ) r7 ÚforwardÚdetachr( ÚnumpyÚviewr r Ú RuntimeError)r; rJ rK rL rM ÚstateÚrÚrewards r= Údynamics_forwardzMuZeroNet.dynamics_forwardX s € à ğ aØÔ,×4Ò4°Q¸tĞ4ÑDÔD‰HˆE�1Ø—X’X‘Z”Z—^’^Ñ%Ô%×+Ò+Ñ-Ô-ˆFˆFà × 1Ò 1°!¸DĞQ_Ğ 1Ñ `Ô `‰MˆE�6Ø ğ hğ
hØ Ÿ
š
4Ô#8¸$Ô:JÈ1Ô:MÈtÔO_Ğ`aÔObÑcÔc��øİ ğ
hğ
hğ
hà Ÿ
š
2 tÔ'<¸dÔ>NÈqÔ>QĞSWÔScĞdeÔSfÑgÔg���ğ
høøøğ �fˆ}Ğ s Á92B, Â,=C,Ã+C,c óŠ — |r#| j |d¬¦ « \ }}||fS | |d|¬¦ « \ }}||fS )NTrO rP )r8 rK )r; rJ rK rL ÚpiÚvs r= Úprediction_forwardz MuZeroNet.prediction_forwardg sY € Ø ğ ØÔ+×3Ò3°A¸dĞ3ÑCÔC‰EˆB�Ø�q�5ˆLØ×'Ò'¨°$À~Ğ'ÑVÔV‰ˆˆAØ�1ˆuˆ r> c ó0 — | |¦ « }|S ©N)r6 )r; rJ s r= Úrepresentation_forwardz MuZeroNet.representation_forwardn s € Ø × 'Ò '¨Ñ *Ô *ˆØˆr> Úhidden_state_with_actionÚ all_predictc óz — | |||¬¦ « \ }}| |||¬¦ « \ }}||||fS )N)rK rL )r[ r_ )r; rc rd rL Ú
next_staterZ r] r^ s r= Úforward_recurrentzMuZeroNet.forward_recurrentr sZ € Ø !× 2Ò 2Ğ3KĞU`ØBPğ 3ñ Rô Rш
�Fà×'Ò'¨
¸KĞXfĞ'ÑgÔg‰ˆˆAؘ6 2 qĞ(Ğ(r> c óì — t | j | j | j | j | j | j | j | j | j | j
| j | j | j
| j | j | j | j ¬¦ « S )N) r$ r r r r% r r! r" r# )r r r r r r r r r r$ r r r r% r r! r" r# ©r; s r= Úmake_fresh_instancez MuZeroNet.make_fresh_instancex sy € ݘÔ,¨d¬l¸DÔ<LÈdÔN_ĞaeÔaqØ Ô.°Ô0FÈÔH_Ø&*Ô&7ÀdÔFWĞfjÔfwØ$(¤OÀÔAQØ-1Ô-EØ*.Ô*?ĞQUÔQeØ15Ô1Mğ
Oñ Oô Oğ Or> Ú
memory_bufferÚ
muzero_configÚreturnc
󠇇— t j t j ¦ « rdnd¦ « }| j €Ct j | ¦ « ‰j ‰j ¬¦ « | _ g }d}ˆˆfd„}t ‰j ¦ « D �]e} |¦ « \ }} }
t |¦ « dk rŒ%|
||
|‰¦ « \ } } }
}t j | ¦ « | ¦ « |
¦ « | ¦ « dœ¦ « | | ¦ « ¦ « | j ¦ « | ¦ « | j ¦ « | j | | j j t2 t4 j || ¦ « | |
||f¬¦ « |dz
}�Œg| j | | j j t2 t4 j ‰|f¬¦ « t; |¦ « t |¦ « z |fS ) Nr' r( ©ÚlrÚ weight_decayr c óF •— ‰ ‰j ‰j ‰¦ « S ra ©Úbatch_with_prioritiesÚ
enable_perÚ
batch_size©rk rl s €€r= ú<lambda>z%MuZeroNet.train_net.<locals>.<lambda>ˆ s' ø€ ˜ ×<Ò<¸]Ô=UØ=JÔ=UĞWdñfô f€ r> rQ )Ú
combined_lossÚloss_vÚloss_piÚloss_r)Úargs) r1 r2 r' r3 r4 ÚoptimÚAdamÚ
parametersrp Úl2ÚrangeÚepochsÚlenÚcalculate_lossesÚwandbÚlogÚitemÚappendÚ zero_gradÚbackwardÚstepr$ Úprocess_hook_executesÚ train_netÚ__name__Ú__file__r
ÚMIDDLEÚTAILÚsum)r; rk rl r2 ÚlossesÚ iterationÚloaderÚepochÚsampled_game_dataÚ
prioritiesÚweightsÚlossrz r{ r| s `` r= r� zMuZeroNet.train_net� sG øø€ İ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆØ Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<À ÔAQØ8EÔ8Hğ +ñ Jô JˆDŒNàˆØˆ ğfğ fğ fğ fğ fˆå˜=Ô/Ñ0Ô0ğ ñ ˆEØ5;°V±X´XÑ 2Ğ ˜z¨7İĞ$Ñ%Ô%¨Ò*Ğ*ØØ,0×,AÒ,AĞBSĞU\Ğ^dĞfsÑ,tÔ,tÑ )ˆD�&˜' 6İ ŒI¨¯ ª © ¬ ¸v¿{º{¹}¼}ĞY`×YeÒYeÑYgÔYgØ!'§¢¡ ¤ ğ0ğ 0ñ
1ô
1ğ
1à �MŠM˜$Ÿ)š)™+œ+Ñ &Ô &Ğ &Ø ŒN× $Ò $Ñ &Ô &Ğ &Ø �MŠM‰OŒOˆOØ ŒN× Ò Ñ !Ô !Ğ !Ø Ô × 3Ò 3°D¸$¼.Ô:QÕS[Õ]cÔ]jØ! 4§9¢9¡;¤;°¸ÀØğrĞ 3ñ
ô
ğ
ğ
˜‰NˆI‰IØ Ô×/Ò/°°d´nÔ6MÍxÕY_ÔYdØ6CÀVĞ5Lğ 0ñ Nô Nğ Nå�6‰{Œ{�S ™[œ[Ñ(¨&Ğ0Ğ0r> c ó´ ‡‡— ‰ ¦ « dk rd S t j t j ¦ « rdnd¦ « }| j €Ct j | ¦ « ‰j ‰j
¬¦ « | _ ˆˆfd„}t ‰j ¦ « D ]¢} |¦ « \ }}}t |¦ « dk rŒ$| |||‰¦ « \ } }
} } t j | ¦ « |
¦ « | ¦ « | ¦ « dœ¦ « Œ£d S )Nr r' r( ro c óJ •— ‰ ‰j ‰j ‰d¬¦ « S )NT)Úis_evalrs rw s €€r= rx z$MuZeroNet.eval_net.<locals>.<lambda>¥ s/ ø€ ˜ ×<Ò<¸]Ô=UØ=JÔ=UØ=JØEIğ =ñ Kô K€ r> rQ )Úeval_combined_lossÚ eval_loss_vÚ eval_loss_piÚ eval_loss_r)Ú eval_lengthr1 r2 r' r3 r4 r~ r r€ rp r� r‚ Ú eval_epochsr„ r… r† r‡ rˆ )
r; rk rl r2 r– r— Úexperience_batchr™ rš r› rz r{ r| s
`` r= Úeval_netzMuZeroNet.eval_net� sj øø€ Ø × $Ò $Ñ &Ô &¨!Ò +Ğ +Ø ˆFİ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆØ Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<À ÔAQØ8EÔ8Hğ +ñ Jô JˆDŒNğKğ Kğ Kğ Kğ Kˆõ ˜=Ô4Ñ5Ô5ğ 6ğ 6ˆEØ4:°F±H´HÑ 1Ğ ˜j¨'İĞ#Ñ$Ô$¨Ò)Ğ)ØØ,0×,AÒ,AĞBRĞT[Ğ]cĞerÑ,sÔ,sÑ )ˆD�&˜' 6İ ŒI¨T¯YªY©[¬[ÈÏÊÉ Ì Ğho×htÒhtÑhvÔhvØ&,§k¢k¡m¤mğ5ğ 5ñ
6ô
6ğ
6ğ
6ğ 6ğ 6r> c ó ‡— | d||¦ « \ }}}}} t |‰j ¦ « }
| |¦ « } | | d¬¦ « \ } }
d\ }}}|| | | ¦ « z
}|| |
|
¦ « z
}d„ t
| d¦ « ¦ « D ¦ « }‰j rk| t j t |
‰j ¦ « |z
¦ « ‰j
z d¦ « ¦ « |¦ « t
d‰j dz ¦ « D �]–}| |||¦ « \ }}}}} t |‰j ¦ « }t |‰j ¦ « }
| t% | |¦ « dd¬¦ « \ } }} }
| d „ ¦ « | | | ¦ « }| |
|
¦ « }| ||¦ « }| ˆfd
„¦ « | ˆfd „¦ « | ˆfd „¦ « ||z
}||z
}||z
}‰j rk| t j t |
‰j ¦ « |z
¦ « ‰j
z d¦ « ¦ « |¦ « �Œ˜|d
z }||z |z }‰j r$|t j ||j |j ¬¦ « z }| ¦ « }‰j r| ||¦ « || ¦ « | ¦ « | ¦ « fS )Nr T)rL )r r r c ó — g | ]}g ‘ŒS © r© )Ú.0rJ s r= ú
<listcomp>z.MuZeroNet.calculate_losses.<locals>.<listcomp>º s € Ğ>Ğ>Ğ> ˜"Ğ>Ğ>Ğ>r> rR rQ Fc ó — | dz S )Ng à?r© )Úgrads r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Æ s
€ °D¸3±J€ r> c ó •— | d‰j z z S ©NrQ ©ÚK©r rl s €r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Ê ó ø€ °d¸aÀ-Ä/Ñ>QÑ6R€ r> c ó •— | d‰j z z S r¯ r° r² s €r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Ë r³ r> c ó •— | d‰j z z S r¯ r° r² s €r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Ì s ø€ °t¸qÀ=Ä?Ñ?RÑ7S€ r> g Ğ?©Údtyper2 )Úget_batch_for_unroll_indexr r rb r_ Ú muzero_lossr‚ Úsizeru Úpopulate_prioritiesr1 Úabsr ÚalphaÚreshapeÚtolistr± rg r Ú
register_hookÚtensorr· r2 r“ Úupdate_priorities)r; r¥ rš r2 rl Ú init_statesÚrewardsÚ
scalar_valuesÚmovesÚpisÚvaluesÚ hidden_stateÚpred_pisÚpred_vsÚpi_lossÚv_lossÚr_lossÚnew_prioritiesÚiÚ_Úpred_rsÚcurrent_pi_lossÚcurrent_v_lossÚcurrent_r_lossr› s ` r= r… zMuZeroNet.calculate_losses± s ø€ Ø:>×:YÒ:YĞZ[Ğ]mĞouÑ:vÔ:vÑ7ˆ �W˜m¨U°Cå" =°-Ô2LÑMÔMˆØ×2Ò2°;Ñ?Ô?ˆ Ø × 3Ò 3°LĞQUĞ 3Ñ VÔ Vш�'Ø")Ñ ˆ�˜Ø�4×#Ò# H¨cÑ2Ô2Ñ2ˆØ�$×"Ò" 7¨FÑ3Ô3Ñ3ˆØ>Ğ>¥e¨H¯MªM¸!Ñ,<Ô,<Ñ&=Ô&=Ğ>Ñ>Ô>ˆØ Ô #ğ .Ø × $Ò $¥b¤fÕ->¸wØ?LÔ?Yñ.[ô .[Ø]jñ.kñ 'lô 'lØo|ô pCñ'C÷ ELò ELØñEô Eß’F‘H”H˜nñ
.ô
.ğ
.õ �q˜-œ/¨AÑ-Ñ.Ô.ğ 2ñ 2ˆAØ48×4SÒ4SĞTUĞWgØTZñ5\ô 5\Ñ 1ˆAˆw˜
u¨cå'¨° Ô1KÑLÔLˆGİ& }°mÔ6PÑQÔQˆFØ7;×7MÒ7Mİ+¨L¸%Ñ@Ô@À%ĞX\ğ 8Nñ 8^ô 8^Ñ 4ˆL˜' 8¨Wà × &Ò &Ğ'>Ğ'>Ñ ?Ô ?Ğ ?Ø "× .Ò .¨x¸Ñ =Ô =ˆOØ !× -Ò -¨g°vÑ >Ô >ˆNØ !× -Ò -¨g°wÑ ?Ô ?ˆNØ × (Ò (Ğ)RĞ)RĞ)RĞ)RÑ SÔ SĞ SØ × (Ò (Ğ)RĞ)RĞ)RĞ)RÑ SÔ SĞ SØ × )Ò )Ğ*SĞ*SĞ*SĞ*SÑ TÔ TĞ TØ �Ñ &ˆGØ �nÑ $ˆFØ �nÑ $ˆFØ Ô'ğ
2Ø×(Ò("¬&Õ1BÀ7ØCPÔC]ñ2_ô 2_Øanñ2oñ +pô +pğ tAô tGñ+G÷ IPò IPØñIô Iß š™œ .ñ2ô 2ğ 2ùğ �$‰ˆØ˜Ñ &Ñ(ˆØ Ô #ğ MØ •B”I˜g¨T¬ZÀÄ ĞLÑLÔLÑ LˆDØ�xŠx‰zŒzˆØ Ô #ğ EØ × "Ò " >Ğ3CÑ DÔ DĞ DØ�V—Z’Z‘\”\ 7§;¢;¡=¤=°&·*²*±,´,Ğ>Ğ>r> Úindexc ó ‡‡— ˆfd„}d }‰dk rAˆfd„|D ¦ « } |t j |¦ « ¦ « dddd¦ « }t j ˆfd„|D ¦ « ¦ « } ||¦ « }t j ˆfd„|D ¦ « ¦ « } ||¦ « }ˆfd „|D ¦ « }t j ˆfd
„|D ¦ « ¦ « } || ¦ « } || d¦ « | d¦ « || fS ) Nc óF •— t j | t j ‰¬¦ « S )Nr¶ )r1 rÁ Úfloat32)rJ r2 s €r= rx z6MuZeroNet.get_batch_for_unroll_index.<locals>.<lambda>à s ø€ ¥"¤)¨AµR´ZÈĞ"OÑ"OÔ"O€ r> r c óX •— g | ]&}t j |j ‰ j ¦ « ‘Œ'S r© )ÚnpÚarrayÚ
datapointsÚframe©rª rJ rÖ s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>ã s- ø€ ĞYĞYĞYÀ1�2œ8 A¤L°Ô$7Ô$=Ñ>Ô>ĞYĞYĞYr> é rQ é c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ rZ rß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>å s# ø€ ĞQĞQĞQ¸1˜AœL¨Ô /Ô 6ĞQĞQĞQr> c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ r^ rß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>ç s# ø€ ĞKĞKĞK°Q˜1œ<¨Ô.Ô0ĞKĞKĞKr> c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ Úmoverß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>é s# ø€ ĞDĞDĞD¨a�” ˜eÔ$Ô)ĞDĞDĞDr> c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ r] rß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>ì s# ø€ ĞIĞIĞI°1˜œ UÔ+Ô.ĞIĞIĞIr> )rÛ rÜ ÚpermuteÚ unsqueeze)
r; rÖ r¥ r2 Ú
tensor_from_xrà rÄ rÈ rÆ rÇ s
` ` r= r¸ z$MuZeroNet.get_batch_for_unroll_indexŞ sM øø€ àOĞOĞOĞOˆ
؈ Ø �AŠ:ˆ:ØYĞYĞYĞYĞHXĞYÑYÔYˆKØ'˜-¬°Ñ(=Ô(=Ñ>Ô>×FÒFÀqÈ!ÈQĞPQÑRÔRˆKİ”(ĞQĞQĞQĞQĞ@PĞQÑQÔQÑRÔRˆØ �- Ñ(Ô(ˆİ”ĞKĞKĞKĞKĞ:JĞKÑKÔKÑLÔLˆØ � ˜vÑ&Ô&ˆØDĞDĞDĞDĞ3CĞDÑDÔDˆõ ŒhĞIĞIĞIĞIĞ8HĞIÑIÔIÑJÔJˆØˆm˜CÑ Ô ˆØ˜G× -Ò -¨aÑ 0Ô 0°&×2BÒ2BÀ1Ñ2EÔ2EÀuÈcĞQĞQr> rÏ Úpriorities_listc óf — t |¦ « D ] \ }}|| |¦ « Œ!d S ra )Ú enumerater‰ )r; rÏ rê ÚidxÚprioritys r= r» z MuZeroNet.populate_prioritiesğ sC € İ & ~Ñ 6Ô 6ğ 2ğ 2‰MˆC�Ø ˜CÔ × 'Ò '¨Ñ 1Ô 1Ğ 1Ğ 1ğ 2ğ 2r> r¥ c ó´ — t |¦ « D ]G\ }}t t |j ¦ « ¦ « D ] }|| | |j | _ Œ!ŒHd S ra )rì r‚ r„ rİ rî )r; rÏ r¥ rí ÚgamerĞ s r= r zMuZeroNet.update_prioritiesô sq € İ"Ğ#3Ñ4Ô4ğ Eğ E‰IˆC�İ�3˜tœÑ/Ô/Ñ0Ô0ğ
Eğ
E�Ø.<¸SÔ.AÀ!Ô.D�” Ô"Ô+Ğ+ğ
Eğ Eğ Er> c ó’ — t j ||z d¬¦ « d¦ « | ¦ « d z S )NrQ )Údimr )r1 r“ rè rº )r; Úy_hatÚys r= r¹ zMuZeroNet.muzero_lossù s= € İ”�q˜5‘y aĞ(Ñ(Ô(×2Ò2°1Ñ5Ô5Ğ5¸¿º¹¼À¼ ÑCĞCr> Úshared_storageÚ checkpointerÚloggerc óP — t j |j d¬¦ « | ¦ « d|_ d|_ t
| ¦ « ¦ « |j dz k rAt j
d¦ « t
| ¦ « ¦ « |j dz k °A| d¦ « t |j
¦ « D ]Ğ}| ||¦ « \ }}| | ¦ « ¦ « | | j ¦ « ¦ « | ||¦ « |dz dk rC|dk r=| d |› d
�¦ « | | | | j |j ||¦ « ŒÑd S ) NzContinuous Weight Update)ÚprojectÚnamerQ é é z:Finished waiting for target buffer size,starting training.iô r z Saving checkpoint at iteration ú.)r† ÚinitÚwandbd_project_nameÚtrainrƒ r¤ r„ Ú
get_bufferrv ÚtimeÚsleepr‡ r‚ Únum_worker_itersr� Ú set_experimental_network_paramsÚ
state_dictÚ
set_optimizerr4 r¦ Úsave_checkpointrp )r; rõ rl rö r÷ Úiter_ÚavgÚ iter_lossess r= Úcontinuous_weight_updatez"MuZeroNet.continuous_weight_updateü s· € å
Œ
˜=Ô<ĞC]Ğ^Ñ^Ô^Ğ^Ø �
Š
‰ Œ ˆ ğ !ˆ
Ô Ø$%ˆ
Ô!İØ ×)Ò)Ñ+Ô+ñ-ô -Ø/<Ô/GÈ1Ñ/LòMğ Må ŒJ�q‰MŒMˆMõ Ø ×)Ò)Ñ+Ô+ñ-ô -Ø/<Ô/GÈ1Ñ/LòMğ Mğ �
Š
ĞOÑPÔPĞPݘ=Ô9Ñ:Ô:ğ qğ qˆEğ $Ÿ~š~¨n¸mÑ LÔ LÑ ˆC�Ø × :Ò :¸4¿?º?Ñ;LÔ;LÑ MÔ MĞ MØ × (Ò (¨¬ ×)BÒ)BÑ)DÔ)DÑ EÔ EĞ Eğ
�MŠM˜.¨-Ñ 8Ô 8Ğ 8Ø�s‰{˜aÒ Ğ E¨Q¢J JØ—
’
ĞE¸UĞEĞEĞEÑFÔFĞFØ ×,Ò,¨T°4¸¼ È ÔIYĞ[`ĞboÑpÔpĞpøğ qğ qr> Úpathc ó0 — t j | |¦ « d S ra )r1 Úsave)r; r
s r= Ú to_picklezMuZeroNet.to_pickle s € İ
Œ��dÑÔĞĞĞr> c ór — | j | | j j t t
j ¦ « d S ra )r$ r� Úrun_on_training_endr� r� r
ÚALLri s r= r z MuZeroNet.run_on_training_end s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr> )NTra )FFT)FF©F)$r� Ú
__module__Ú __qualname__ÚintÚfloatÚlistÚboolr r0 Ú classmethodr rI r1 ÚTensorr[ r_ rb rg rj r Útupler� r¦ r… r¸ r» r r¹ r r r r Ústrr r Ú
__classcell__©r< s @r= r r s¨ ø€ € € € € ğ
PTğ 5rğ 5r sğ 5r°Uğ 5rÈğ 5rĞ\_ğ 5rĞnrĞsvÔnwğ 5rØ#&ğ5rØ;>Ğ;KÀ$ÀsÄ)ğ5rØadğ5rà #ğ5rà36ğ5ràDGğ5rğ '*ğ5rğ >Ağ5rğ TWğ5rğ ruğ5rğ +Ğ 2¨dğ 5rğ IMğ 5rğ 5rğ 5rğ 5rğ 5rğ 5rğn ğNğ N lğ NÀ+ĞBUĞQUğ Nğ Nğ Nñ „[ğNğ \aØ26ğ
ğ
"¤)ğ
°dğ
ĞTXğ
Ø+/ğ
ğ
ğ
ğ
ğ ğ B¤Iğ ¸ğ ĞVZğ ğ ğ ğ 𨬠ğ ğ ğ ğ ğ)ğ )¸"¼)ğ )ĞRVğ )Ğhlğ )ğ )ğ )ğ )ğ Oğ Oğ Oğ1Ğ':ğ 1È<ğ 1Ğ\aĞbgĞimĞnsÔitĞbtÔ\uğ 1ğ 1ğ 1ğ 1ğ86Ğ&9ğ 6È,ğ 6Ğ[_ğ 6ğ 6ğ 6ğ 6ğ(+?ğ +?ğ +?ğZR°ğ RĞRWØ
Œ �2”9˜bœi¨¨r¬yĞ8ôS:ğ Rğ Rğ Rğ Rğ$2°$ğ 2Èğ 2ğ 2ğ 2ğ 2ğE°ğ EÈğ Eğ Eğ Eğ Eğ
Dğ Dğ Dğ q°}ğ qĞUağ qØ/;ğ qØEKğ qğ qğ qğ qğ> ˜cğ ğ ğ ğ ğoğ oğ oğ oğ oğ oğ or> r c óF ‡ — e Zd Zddedefˆ fd„
Zdej fd„Zd„ Z ˆ xZ
S ) r5 Tr r% c óB •— t t | ¦ « ¦ « t j t j ¦ « rdnd¦ « | _ t j |dddd¬¦ « | _ t j
d„ t d¦ « D ¦ « ¦ « | _ t j dd ddd¬¦ « | _
t j
d
„ t d¦ « D ¦ « ¦ « | _ t j
d „ t d¦ « D ¦ « ¦ « | _ |rOt j ddd¬ ¦ « | _ t j ddd¬ ¦ « | _ nFt j ¦ « | _ t j ¦ « | _ t j ¦ « | _ d S )
Nr' r( é€ rà rá rQ )r* r. Ú kernel_sizeÚstrideÚpaddingc ó, — g | ]}t d ¦ « ‘ŒS )r# ©Ú
ResidualBlock©rª rÑ s r= r« z.RepresentationNet.__init__.<locals>.<listcomp>' ó € Ğ+QĞ+QĞ+QÀ1M¸#Ñ,>Ô,>Ğ+QĞ+QĞ+Qr> r- c ó, — g | ]}t d ¦ « ‘ŒS ©r- r( r* s r= r« z.RepresentationNet.__init__.<locals>.<listcomp>) r+ r> c ó, — g | ]}t d ¦ « ‘ŒS r- r( r* s r= r« z.RepresentationNet.__init__.<locals>.<listcomp>+ r+ r> )r$ r% r& )r/ r5 r0 r1 r2 r' r3 r ÚConv2dÚconv1Ú
ModuleListr‚ Ú
residuals1Úconv2Ú
residuals2Ú
residuals3Ú AvgPool2dÚpool1Úpool2ÚIdentityÚReLUÚrelu)r; r r% r< s €r= r0 zRepresentationNet.__init__# sŒ ø€ İ
Õ Ñ&Ô&×/Ò/Ñ1Ô1Ğ1İ”i"¬'×*>Ò*>Ñ*@Ô*@Ğ K ÀeÑLÔLˆŒ İ”U—\’\Ğ.@ÈsĞ`aĞjkĞuv�\ÑwÔwˆŒ
İ œ%×*Ò*Ğ+QĞ+QÍÈaÉÌĞ+QÑ+QÔ+QÑRÔRˆŒİ”U—\’\¨cÀĞQRĞ[\Ğfg�\ÑhÔhˆŒ
İ œ%×*Ò*Ğ+QĞ+QÍÈaÉÌĞ+QÑ+QÔ+QÑRÔRˆŒå œ%×*Ò*Ğ+QĞ+QÍÈaÉÌĞ+QÑ+QÔ+QÑRÔRˆŒØ ğ *İœŸ š °Q¸qÈ!˜ ÑLÔLˆDŒJİœŸ š °Q¸qÈ!˜ ÑLÔLˆDŒJˆJ土 š Ñ)Ô)ˆDŒJİœŸ š Ñ)Ô)ˆDŒJİ”E—J’J‘L”LˆŒ ˆ ˆ r> rJ c ó¬ — | | j ¦ « }| | |¦ « ¦ « }| j D ]
} ||¦ « }Œ| | |¦ « ¦ « }| j D ]
} ||¦ « }Œ| |¦ « }| j D ]
} ||¦ « }Œ| |¦ « }|S ra )
Útor2 r; r0 r2 r3 r4 r7 r5 r8 )r; rJ Úresiduals r= rS zRepresentationNet.forward4 sÎ € à
�DŠD�”Ñ Ô ˆØ �IŠI�d—j’j ‘m”mÑ $Ô $ˆØ œğ ğ ˆHØ�˜‘ ” ˆAˆAØ �IŠI�d—j’j ‘m”mÑ $Ô $ˆØ œğ ğ ˆHØ�˜‘ ” ˆAˆAØ �JŠJ�q‰MŒMˆØ œğ ğ ˆHØ�˜‘ ” ˆAˆAØ �JŠJ�q‰MŒMˆØˆr> c ón — t j d¦ « }t j | |¦ « }|S )N)rQ r# é r@ )r1 ÚrandÚjitÚtrace©r; ÚdataÚtraced_script_modules r= rC zRepresentationNet.traceC s- € İŒw�~Ñ&Ô&ˆİ !œvŸ|š|¨D°$Ñ 7Ô 7Ğ Ø#Ğ#r> )T) r� r r r r r0 r1 r rS rC r r s @r= r5 r5 " s| ø€ € € € € ğ!ğ !¨3ğ !¸Tğ !ğ !ğ !ğ !ğ !ğ !ğ"
˜œğ
ğ
ğ
ğ
ğ $ğ $ğ $ğ $ğ $ğ $ğ $r> r5 c óv ‡ — e Zd Zˆ fd„Zdd„Zdej j fd„Z ej ¦ « d„ ¦ « Z
ˆ xZ S )r9 c ó¨ •— t t | ¦ « ¦ « || _ || _ t j ||dd¬¦ « | _ t j |¦ « | _ t j ||dd¬¦ « | _
t j |¦ « | _ t j ||dd¬¦ « | _ t j |¦ « | _
t j ||d¦ « | _ t j |¦ « | _ t j dd¦ « | _ t j d¦ « | _ t j dd¦ « | _ t j d¦ « | _ t j |¦ « | _ t j d|d |d z |z ¦ « | _ t j dd¦ « | _ d S )Nrà rQ )r& i i i r )r/ r9 r0 r. r r r/ r0 Ú BatchNorm2dÚbn1r3 Úbn2Úconv3Úbn3Úconv4Úbn4ÚLinearÚfc1Ú BatchNorm1dÚfc1_bnÚfc2Úfc2_bnÚDropoutr Ú
state_headÚ reward_head)r; r* r r r r. r< s €r= r0 zDynamicsNet.__init__J st ø€ İ
�k˜4Ñ Ô ×)Ò)Ñ+Ô+Ğ+Ø (ˆÔØ&ˆÔõ ”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1Ñ=Ô=ˆŒ
İ”> ,Ñ/Ô/ˆŒõ ”9˜T 4Ñ(Ô(ˆŒİ”n TÑ*Ô*ˆŒ İ”9˜T 3Ñ'Ô'ˆŒİ”n SÑ)Ô)ˆŒ å”z 'Ñ*Ô*ˆŒ õ œ) C¨°Q¬ ¸+Àa¼.Ñ)HÈ<Ñ)WÑXÔXˆŒİ œ9 S¨!Ñ,Ô,ˆÔĞĞr> Fc ó¼ — t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }|
| d¦ « d¦ « }t j | |
|¦ « ¦ « ¦ « }| |¦ « }t j | | |¦ « ¦ « ¦ « }| |¦ « }| |¦ « }| |¦ « }||fS )Nr rR )ÚFr; rJ r0 rK r3 rM rL rO rN rV rº rS rQ r rU rT rW rX )r; rJ r+ rX rY s r= rS zDynamicsNet.forwardf sH € å
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆà
�FŠF�1—6’6˜!‘9”9˜bÑ !Ô !ˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆà—’ Ñ"Ô"ˆØ × Ò ˜QÑ Ô ˆà�aˆxˆr> rm c ó” — t j d¦ « d¦ « }t j | |¦ « }|S )N)rQ r) é r\ zcuda:0)r1 rA r= rB rC rD s r= rC zDynamicsNet.tracez s; € İŒw�~Ñ&Ô&×)Ò)¨(Ñ3Ô3ˆİ !œvŸ|š|¨D°$Ñ 7Ô 7Ğ Ø#Ğ#r> c ó
— | |¦ « \ }}| | j | j d | j d ¦ « }|| ¦ « ¦ « ¦ « fS )Nr rQ )rS rV r. r rT r( rU )r; rJ rX rY s r= rK zDynamicsNet.predict sg € à—<’< ‘?”?‰ˆˆqØ—
’
˜4Ô,¨dÔ.>¸qÔ.AÀ4ÔCSĞTUÔCVÑWÔWˆØ�a—h’h‘j”j—n’nÑ&Ô&×,Ò,Ñ.Ô.Ğ.Ğ.r> r ) r� r r r0 rS r1 rB ÚScriptFunctionrC Úno_gradrK r r s @r= r9 r9 I s� ø€ € € € € ğ-ğ -ğ -ğ -ğ -ğ8ğ ğ ğ ğ($�r”vÔ,ğ $ğ $ğ $ğ $ğ
€R„Z�\„\ğ/ğ /ñ „\ğ/ğ /ğ /ğ /ğ /r> r9 c ó* ‡ — e Zd Zdefˆ fd„ Zd„ Zˆ xZS )r) Úchannelsc óÌ •— t t | ¦ « ¦ « t j ||dd¬¦ « | _ t j |¦ « | _ t j ||dd¬¦ « | _ t j |¦ « | _
t j ¦ « | _ d S )Nrà rQ )r* r. r$ r& )
r/ r) r0 r1 r r/ Ú convolution1rI Úbnorm1Ú convolution2Úbnorm2r: r; )r; ra r< s €r= r0 zResidualBlock.__init__‡ s¢ ø€ İ
�m˜TÑ"Ô"×+Ò+Ñ-Ô-Ğ-İ œEŸLšL°XÈHĞbcĞmn˜LÑ oÔ oˆÔİ”e×'Ò'¨Ñ1Ô1ˆŒ İ œEŸLšL°XÈHĞbcĞmn˜LÑ oÔ oˆÔİ”e×'Ò'¨Ñ1Ô1ˆŒ İ”E—J’J‘L”LˆŒ ˆ ˆ r> c ó — |}| |¦ « }| | |¦ « ¦ « }| | |¦ « ¦ « }||z
}| |¦ « }|S ra )rc r; rd rf re )r; rJ Úx_resÚ convolveds r= rS zResidualBlock.forward� ss € àˆØ×%Ò% aÑ(Ô(ˆ Ø �IŠI�d—k’k )Ñ,Ô,Ñ -Ô -ˆØ �KŠK˜×)Ò)¨!Ñ,Ô,Ñ -Ô -ˆØ ˆU‰
ˆØ �IŠI�a‰LŒLˆØˆr> )r� r r r r0 rS r r s @r= r) r) † sS ø€ € € € € ğ! ğ !ğ !ğ !ğ !ğ !ğ !ğğ ğ ğ ğ ğ ğ r> r) )*r rU rÛ Útorchr1 Útorch.nn.functionalr Ú
functionalrZ r† r Ú$mu_alpha_zero.AlphaZero.Network.nnetr r: r Ú$mu_alpha_zero.AlphaZero.checkpointerr Ú mu_alpha_zero.AlphaZero.loggerr Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.mz_gamer
Ú mu_alpha_zero.General.networkr Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr
Úmu_alpha_zero.MuZero.utilsr r r Úmu_alpha_zero.configr Ú$mu_alpha_zero.shared_storage_managerr ÚModuler r5 r9 r) r© r> r= ú<module>ry s ğØ € € € à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ø € € € Ø Ğ Ğ Ğ Ğ Ğ Ø (Ğ (Ğ (Ğ (Ğ (Ğ (à gĞ gĞ gĞ gĞ gĞ gĞ gĞ gØ =Ğ =Ğ =Ğ =Ğ =Ğ =Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø >Ğ >Ğ >Ğ >Ğ >Ğ >Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø hĞ hĞ hĞ hĞ hĞ hĞ hĞ hĞ hĞ hØ -Ğ -Ğ -Ğ -Ğ -Ğ -Ø >Ğ >Ğ >Ğ >Ğ >Ğ >ğHoğ Hoğ Hoğ Hoğ Ho�”” Ğ 2ñ Hoô Hoğ HoğV$$ğ $$ğ $$ğ $$ğ $$˜œœ ñ $$ô $$ğ $$ğN:/ğ :/ğ :/ğ :/ğ :/�"”)ñ :/ô :/ğ :/ğzğ ğ ğ ğ �B”E”Lñ ô ğ ğ ğ r> | 33,954 | Python | .py | 94 | 360.12766 | 2,500 | 0.306695 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,543 | hook_callables.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/hook_callables.py | from abc import abstractmethod, ABC
from typing import Callable
class HookCallable(ABC):
@abstractmethod
def execute(self, cls: object, *args):
"""
Execute the hook
:return: The return of the hook
"""
pass
class HookMethodCallable(HookCallable):
def __init__(self, method: Callable, args: tuple):
super().__init__()
self.__method = method
self.__args = args
def execute(self, cls: object, *args):
return self.__method(cls, *self.__args, *args)
| 541 | Python | .py | 17 | 25.117647 | 54 | 0.623301 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,544 | hook_manager.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/hook_manager.py | from mu_alpha_zero.Hooks.hook_callables import HookCallable
from mu_alpha_zero.Hooks.hook_point import HookPoint, HookAt
class HookManager:
def __init__(self):
self.hooks = {}
def register_method_hook(self, where: HookPoint, hook: HookCallable):
self.hooks[where] = hook
def process_hook_executes(self, cls: object, fn_name: str, file: str, at: HookAt, args: tuple = ()):
file_name = file.replace("\\", "/").split("/")[-1]
for hook_point, hook_callable in self.hooks.items():
if hook_point.here(file_name, fn_name, at):
hook_callable.execute(cls, *args)
return
| 653 | Python | .py | 13 | 42.153846 | 104 | 0.638365 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,545 | hook_point.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/hook_point.py | import enum
class HookAt(enum.Enum):
HEAD = "head"
TAIL = "tail"
MIDDLE = "middle"
ALL = "all"
class HookPoint:
def __init__(self, at: HookAt, file: str, fn_name: str):
self.__at = at
self.__file = file
self.__function_name = fn_name
@property
def at(self) -> HookAt:
return self.__at
@property
def file(self) -> str:
return self.__file
@property
def function_name(self) -> str:
return self.__function_name
def here(self, file: str, function_name: str, at: HookAt):
return file == self.file and function_name == self.function_name and (at == self.at or at == HookAt.ALL)
| 683 | Python | .py | 22 | 24.954545 | 112 | 0.598775 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,546 | hook_manager.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/__pycache__/hook_manager.cpython-311.pyc | §
føf� ã ó< — d dl mZ d dlmZmZ G d„ d¦ « ZdS )é )Ú HookCallable)Ú HookPointÚHookAtc óB — e Zd Zd„ Zdedefd„Zd
dededed e d
e
f
d „Z d S )Ú HookManagerc ó — i | _ d S ©N©Úhooks)Úselfs úO/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Hooks/hook_manager.pyÚ__init__zHookManager.__init__ s
€ ؈Œ
ˆ
ˆ
ó ÚwhereÚhookc ó — || j |< d S r r
)r r r s r
Úregister_method_hookz HookManager.register_method_hook s € Ø ˆŒ
�5ÑĞĞr © ÚclsÚfn_nameÚfileÚatÚargsc óò — | dd¦ « d¦ « d }| j ¦ « D ]-\ }}| |||¦ « r |j |g|¢R � d S Œ.d S )Nú\ú/éÿÿÿÿ)ÚreplaceÚsplitr ÚitemsÚhereÚexecute) r r r r r r Ú file_nameÚ
hook_pointÚ
hook_callables r
Úprocess_hook_executesz!HookManager.process_hook_executes s“ € Ø—L’L sÑ+Ô+×1Ò1°#Ñ6Ô6°rÔ:ˆ Ø)-¬×)9Ò)9Ñ);Ô);ğ ğ Ñ %ˆJ˜
Ø�Š˜y¨'°2Ñ6Ô6ğ
Ø%�
Ô% cĞ1¨DĞ1Ğ1Ğ1Ğ1Ø��ğ
ğ ğ r N)r ) Ú__name__Ú
__module__Ú __qualname__r r r r ÚobjectÚstrr Útupler& r r r
r r s‚ € € € € € ğğ ğ ğ!¨)ğ !¸<ğ !ğ !ğ !ğ !ğğ ¨ğ ¸#ğ ÀSğ Èfğ Ğ\ağ ğ ğ ğ ğ ğ r r N)Ú"mu_alpha_zero.Hooks.hook_callablesr Ú mu_alpha_zero.Hooks.hook_pointr r r r r r
ú<module>r/ sa ğØ ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ğ <Ğ <ğ ğ ğ ğ ğ ñ ô ğ ğ ğ r | 1,829 | Python | .py | 12 | 151.333333 | 480 | 0.388889 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,547 | hook_callables.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/__pycache__/hook_callables.cpython-311.pyc | §
føf ã óZ — d dl mZmZ d dlmZ G d„ de¦ « Z G d„ de¦ « ZdS )é )ÚabstractmethodÚABC)ÚCallablec ó* — e Zd Zedefd„¦ « ZdS )Ú HookCallableÚclsc ó — dS )zJ
Execute the hook
:return: The return of the hook
N© ©Úselfr Úargss úQ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Hooks/hook_callables.pyÚexecutezHookCallable.execute s € ğ
ˆó N)Ú__name__Ú
__module__Ú __qualname__r Úobjectr r
r r r r s9 € € € € € àğ
˜6ğ
ğ
ğ
ñ „^ğ
ğ
ğ
r r c ó4 ‡ — e Zd Zdedefˆ fd„ Zdefd„Zˆ xZS )ÚHookMethodCallableÚmethodr
c ód •— t ¦ « ¦ « || _ || _ d S ©N)ÚsuperÚ__init__Ú_HookMethodCallable__methodÚ_HookMethodCallable__args)r r r
Ú __class__s €r r zHookMethodCallable.__init__ s+ ø€ İ
‰Œ×ÒÑÔĞØ ˆŒ
؈Œ ˆ ˆ r r c ó, — | j |g| j ¢|¢R � S r )r r r s r r zHookMethodCallable.execute s# € Ø ˆtŒ}˜SĞ6 4¤;Ğ6°Ğ6Ğ6Ğ6Ğ6r ) r r r r Útupler r r Ú
__classcell__)r s @r r r sf ø€ € € € € ğ˜xğ ¨uğ ğ ğ ğ ğ ğ ğ
7˜6ğ 7ğ 7ğ 7ğ 7ğ 7ğ 7ğ 7ğ 7r r N)Úabcr r Útypingr r r r
r r ú<module>r$ s� ğØ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ø Ğ Ğ Ğ Ğ Ğ ğ
ğ
ğ
ğ
ğ
�3ñ
ô
ğ
ğ7ğ 7ğ 7ğ 7ğ 7˜ ñ 7ô 7ğ 7ğ 7ğ 7r | 1,774 | Python | .py | 9 | 193.333333 | 893 | 0.378256 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,548 | hook_point.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/__pycache__/hook_point.cpython-311.pyc | §
føf« ã óN — d dl Z G d„ de j ¦ « Z G d„ d¦ « ZdS )é Nc ó — e Zd ZdZdZdZdZdS )ÚHookAtÚheadÚtailÚmiddleÚallN)Ú__name__Ú
__module__Ú __qualname__ÚHEADÚTAILÚMIDDLEÚALL© ó úM/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Hooks/hook_point.pyr r s" € € € € € Ø €DØ €DØ
€FØ
€C€C€Cr r c óŠ — e Zd Zdededefd„Zedefd„¦ « Zedefd„¦ « Zedefd„¦ « Z ded edefd
„Z
d S ) Ú HookPointÚatÚfileÚfn_namec ó0 — || _ || _ || _ d S ©N)Ú_HookPoint__atÚ_HookPoint__fileÚ_HookPoint__function_name)Úselfr r r s r Ú__init__zHookPoint.__init__ s € ؈Œ ؈Œ Ø &ˆÔ Ğ Ğ r Úreturnc ó — | j S r )r ©r s r r z HookPoint.at s
€ àŒyĞr c ó — | j S r )r r! s r r zHookPoint.file s
€ àŒ{Ğr c ó — | j S r )r r! s r Ú
function_namezHookPoint.function_name s € àÔ#Ğ#r r$ c ód — || j k o%|| j k o|| j k p|t j k S r )r r$ r r r )r r r$ r s r ÚherezHookPoint.here s8 € Ø�t”yÒ Ğp ]°dÔ6HÒ%HĞpÈbĞTXÔT[ÊmĞNoĞ_aÕekÔeoÒ_oĞpr N) r r
r r Ústrr Úpropertyr r r$ r& r r r r r sä € € € € € ğ'˜6ğ '¨ğ '°sğ 'ğ 'ğ 'ğ 'ğ
ğ�Fğ ğ ğ ñ „Xğğ ğ�cğ ğ ğ ñ „Xğğ ğ$˜sğ $ğ $ğ $ñ „Xğ$ğq˜ğ q¨Sğ q°fğ qğ qğ qğ qğ qğ qr r )ÚenumÚEnumr r r r r ú<module>r+ sv ğØ € € € ğğ ğ ğ ğ ˆTŒYñ ô ğ ğqğ qğ qğ qğ qñ qô qğ qğ qğ qr | 2,064 | Python | .py | 10 | 205.3 | 453 | 0.345985 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,549 | donkey_kong.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/examples/__pycache__/donkey_kong.cpython-311.pyc | §
à6Ûeî ã óJ — d dl Zd dlZd dlmZ d dlmZ G d„ de¦ « ZdS )é N)Úmake)Ú
MuZeroGamec óæ — e Zd Zd„ Zdedepddej pej ee ffd„Z
dej pej fd„Z defd„Z defd „Z
depdde pdfd
„Zd „ Zd „ Zddedepddedej pej ee ffd„ZdS )ÚDonkeyKongGamec ó@ — t dd¬¦ « | _ d| _ d S )NzALE/DonkeyKong-v5Úrgb)Úobs_typeF)r Údonkey_kong_envÚis_done©Úselfs úC/home/skyr/PycharmProjects/MuAlphaZeroBuild/examples/donkey_kong.pyÚ__init__zDonkeyKongGame.__init__ s# € İ #Ğ$7À%Ğ HÑ HÔ HˆÔ Ø ˆŒ ˆ ˆ ó ÚactionÚplayerNÚreturnc óL — | j |¦ « \ }}}}}|||fS ©N)r
Ústep)r
r r ÚobsÚrewÚdoneÚ_s r Úget_next_statez DonkeyKongGame.get_next_state s/ € à #Ô 3× 8Ò 8¸Ñ @Ô @Ñ ˆˆS�$˜˜1Ø�C˜ˆ~Ğ r c ó> — | j ¦ « \ }}|S r )r
Úreset)r
r r s r r zDonkeyKongGame.reset s € ØÔ%×+Ò+Ñ-Ô-‰ˆˆQ؈
r c ó — dS )Nr © r s r Úget_noopzDonkeyKongGame.get_noop s € ؈qr c ó$ — | j j j S r )r
Ú action_spaceÚnr s r Úget_num_actionsz DonkeyKongGame.get_num_actions s € ØÔ#Ô0Ô2Ğ2r c ó — | j S r )r )r
r s r Ú game_resultzDonkeyKongGame.game_result s
€ ØŒ|Ğr c ó — t ¦ « S r )r r s r Úmake_fresh_instancez"DonkeyKongGame.make_fresh_instance s € İ Ñ Ô Ğ r c ó — d S r r r s r ÚrenderzDonkeyKongGame.render! s € Ø ˆr é Ú
frame_skipc ó� — | ||¦ « \ }}}t |dz
¦ « D ] }| ||¦ « \ }}}Œ |||fS )Né )r Úrange)r
r r r, r r r Úis r Úframe_skip_stepz DonkeyKongGame.frame_skip_step$ sc € à ×,Ò,¨V°VÑ<Ô<‰ˆˆS�$İ�z A‘~Ñ&Ô&ğ Ağ AˆAØ !× 0Ò 0°¸Ñ @Ô @‰NˆC��d�dØ�C˜ˆ~Ğ r )r+ )Ú__name__Ú
__module__Ú __qualname__r ÚintÚnpÚndarrayÚthÚTensorÚboolr r r r$ r& r( r* r1 r r r r r sS € € € € € ğ ğ ğ ğ Sğ °#°+¸ğ Ø ŒJĞ #˜"œ) S¨$ğC0ğ ğ ğ ğ ğ
�r”zĞ. R¤Yğ ğ ğ ğ ğ˜#ğ ğ ğ ğ ğ3 ğ 3ğ 3ğ 3ğ 3ğ # +¨ğ °$°,¸$ğ ğ ğ ğ ğ ğ ğ ğ
ğ
ğ
ğ ğ cğ °3°;¸$ğ ÈCğ Ø ŒJĞ #˜"œ) S¨$ğY0ğ ğ ğ ğ ğ ğ r r ) Únumpyr6 Útorchr8 Ú gymnasiumr Ú
mu_alpha_zeror r r r r ú<module>r? su ğØ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ø $Ğ $Ğ $Ğ $Ğ $Ğ $ğ" ğ " ğ " ğ " ğ " �Zñ " ô " ğ " ğ " ğ " r | 3,166 | Python | .pyt | 14 | 225.071429 | 561 | 0.355217 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,550 | mem_buffer.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/__pycache__/mem_buffer.cpython-311.pyc | §
aG¡f«@ ã ó� — d dl Z d dlZd dlZd dlmZ d dl mZ d dlZd dlZd dl Z d dl
Z d dl m
Z
d dlmZmZ d dlmZ d dlmZ d dlmZ d d lmZ d d
lmZ d d lm Z d d l m Z d d
l m Z G d„ de¦ « Z! G d„ de¦ « Z" G d„ d¦ « Z# G d„ d¦ « Z$ G d„ d¦ « Z% G d„ de¦ « Z& G d„ de¦ « Z'dS ) é N)Údeque)Úchain)ÚDeque)ÚDatasetÚ
DataLoader)ÚGeneralMemoryBuffer)Úfind_project_root)Ú HookManager)ÚHookAt)Ú LazyArray)Ú DataPickler)Ú scale_action)Ú MuZeroConfigc ó — e Zd Zd„ Zd„ Zd„ ZdS )Ú
MemDatasetc ó. — t |¦ « | _ d S ©N)ÚlistÚ
mem_buffer)Úselfr s úG/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/mem_buffer.pyÚ__init__zMemDataset.__init__ s € İ ˜zÑ*Ô*ˆŒˆˆó c ó* — t | j ¦ « S r )Úlenr ©r s r Ú__len__zMemDataset.__len__ s € İ�4”?Ñ#Ô#Ğ#r c ó — | j | S r )r )r Úidxs r Ú __getitem__zMemDataset.__getitem__ s € ØŒ˜sÔ#Ğ#r N)Ú__name__Ú
__module__Ú __qualname__r r r © r r r r sA € € € € € ğ+ğ +ğ +ğ$ğ $ğ $ğ$ğ $ğ $ğ $ğ $r r c ó — e Zd Z d&dedededepdfd„Zd'defd „Zdepdfd
„Zd „ Z d(d
e
fd„Z d„ Z d'defd„Z
d'de
defd„Zd'dedefd„Zd„ Zd„ Zd'dede
dedefd„Zd„ Zd'ded edefd „Zd ede
d e
defd „Zd!„ Zd"„ Z d#„ Z d$„ Z d%„ Z dS ))Ú MemBufferFNÚdiskÚ full_diskÚdir_pathÚ hook_managerc óğ — || _ || _ || _ || _ |�|n
t ¦ « | _ | |¦ « | _ | |¦ « | _ d| _ || _
d S ©Nr ) Úmax_sizer' r( r) r
r* Ú init_bufferÚbufferÚ eval_bufferÚlast_buffer_sizeÚis_disk)r r- r' r( r) r* s r r zMemBuffer.__init__$ ss € à ˆŒ
؈Œ Ø"ˆŒØ ˆŒ
Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ×&Ò& xÑ0Ô0ˆŒ Ø ×+Ò+¨HÑ5Ô5ˆÔØ !ˆÔ 𠈌 ˆ ˆ r Úis_evalc ó¾ — t |t ¦ « st d¦ « ‚|s | j |¦ « d S | j |¦ « d S )Nz'Experience must be a DataPoint instance)Ú
isinstanceÚSingleGameDataÚ
ValueErrorr/ Úappendr0 )r Ú
experiencer3 s r Úaddz
MemBuffer.add2 sg € ݘ*¥nÑ5Ô5ğ Hİ Ğ FÑGÔGĞ Gğ
ğ 0Ø ŒK× Ò ˜zÑ *Ô *Ğ *Ğ *Ğ *à Ô × #Ò # JÑ /Ô /Ğ /Ğ /Ğ /r c óš — | j r0| j r)|€t ¦ « › d�}t | j |¬¦ « S t | j ¬¦ « S )Nz
/Pickles/Data)ÚmaxlenÚ directory©r< )r' r( r r r- r )r r) s r r. zMemBuffer.init_buffer> sW € Ø Œ9ğ /˜œ ğ /ØĞ İ /Ñ 1Ô 1Ğ@Ğ@Ğ@�İ ¤
¸ĞBÑBÔBĞ Bå ¤
Ğ.Ñ.Ô.Ğ .r c ó — | j S r )r) r s r Ú get_dir_pathzMemBuffer.get_dir_pathF s
€ ØŒ}Ğ r é
Ú percent_evalc óè — t | j ¦ « dk r:t j ¦ « }|dk r!|D ] }| j |¦ « Œ d S |D ] }| j |¦ « Œ d S )NrA gš™™™™™¹?)r r/ Úrandomr0 r8 )r Úexperience_listrB ÚprobÚitems r Úadd_listzMemBuffer.add_listI s‹ € İ ˆtŒ{Ñ Ô ˜rÒ !Ğ !İ”=‘?”?ˆDØ�cŠzˆzà +ğ 2ğ 2�DØÔ$×+Ò+¨DÑ1Ô1Ğ1Ğ1Ø�Ø#ğ %ğ %ˆDØ ŒK× Ò ˜tÑ $Ô $Ğ $Ğ $ğ %ğ %r c ó6 — t j | j |¦ « S r )rD Úsampler/ ©r Ú
batch_sizes r rJ zMemBuffer.sampleT s € İŒ}˜Tœ[¨*Ñ5Ô5Ğ5r c ór — |rt j | j ¦ « d S t j | j ¦ « d S r )rD Úshuffler0 r/ )r r3 s r rN zMemBuffer.shuffleW s; € Ø ğ (İ ŒN˜4Ô+Ñ ,Ô ,Ğ ,Ğ ,Ğ ,å ŒN˜4œ;Ñ 'Ô 'Ğ 'Ğ 'Ğ 'r rL c óæ — g }|s| j n| j }t |¦ « }t d||¦ « D ]=}| t |¦ « |t
||z |¦ « … ¦ « Œ>|S r, )r/ r0 r Úranger8 r Úmin)r rL r3 Úbatched_bufferÚbufÚ
buffer_lenr s r ÚbatchzMemBuffer.batch] s~ € ؈Ø!(Ğ>ˆdŒkˆk¨dÔ.>ˆİ˜‘X”Xˆ
ݘ˜J¨
Ñ3Ô3ğ Tğ TˆCØ × !Ò !¥$ s¡)¤)¨Cµ°C¸*Ñ4DÀjÑ0QÔ0QĞ,QÔ"RÑ SÔ SĞ SĞ Sà Ğ r Úreturnc ó0 — | ||¬¦ « S )N)r3 )rU )r rL r3 s r Ú__call__zMemBuffer.__call__f s € Ø�zŠz˜*¨gˆzÑ6Ô6Ğ6r c ó* — t | j ¦ « S r )r r/ r s r r zMemBuffer.__len__i s € İ�4”;Ñ Ô Ğ r c ó* — t | j ¦ « S r )r r0 r s r Ú eval_lengthzMemBuffer.eval_lengthl s € İ�4Ô#Ñ$Ô$Ğ$r Ú
enable_perÚconfigc
ó ‡‡‡‡ — |s| j n| j Š|r t j d„ ‰D ¦ « ¦ « Šn"t j t ‰¦ « f¦ « Š‰‰ ¦ « z Št j t j t ‰¦ « ¦ « dt |t ‰¦ « ¦ « ‰¬¦ « ¦ « }ˆˆfd„|D ¦ « Š ‰j rYˆˆ fd„t t ‰ ¦ « ¦ « D ¦ « }t j |¦ « }d||j d z ‰j z z }n"t j t ‰ ¦ « f¦ « }‰ ‰| dd¦ « fS ) Nc ó — g | ] }|j ‘Œ
S r$ )Ú
game_priority©Ú.0Úxs r ú
<listcomp>z3MemBuffer.batch_with_priorities.<locals>.<listcomp>w s € Ğ'EĞ'EĞ'E¸A¨¬Ğ'EĞ'EĞ'Er F)ÚreplaceÚsizeÚpc ól •— g | ]0}‰| ‰¦ « ‰¦ « ‘Œ1S r$ )Ú normalizeÚsample_position_with_unroll)rb ÚindexrS r] s €€r rd z3MemBuffer.batch_with_priorities.<locals>.<listcomp>~ s: ø€ ĞsĞsĞsĞZ_�S˜”Z×)Ò)¨&Ñ1Ô1×MÒMÈfÑUÔUĞsĞsĞsr c óR •— g | ]#}‰| j d j ‰| z ‘Œ$S ©r )Ú
datapointsÚpriority)rb ÚiÚgame_prioritiesÚ positionss €€r rd z3MemBuffer.batch_with_priorities.<locals>.<listcomp>€ s3 ø€ ĞoĞoĞoĞTU�y ”|Ô.¨qÔ1Ô:¸_ÈQÔ=OÑOĞoĞoĞor é r éÿÿÿÿ)r/ r0 ÚnpÚarrayÚonesr ÚsumrD ÚchoiceÚarangerQ Útolistr\ rP ÚshapeÚbetaÚreshape)
r r\ rL r] r3 Úsampled_indexesÚweightsrS rq rr s
` @@@r Úbatch_with_prioritiesz MemBuffer.batch_with_prioritieso s øøøø€ Ø!(Ğ>ˆdŒkˆk¨dÔ.>ˆğ ğ 3İ œhĞ'EĞ'EÀĞ'EÑ'EÔ'EÑ FÔ FˆOˆOå œg¥s¨3¡x¤x kÑ 2Ô 2ˆOà˜?×.Ò.Ñ0Ô0Ñ0ˆİ œ)×*Ò*2¬9µS¸±X´XÑ+>Ô+>ÈÕTWĞXbÕdgĞhkÑdlÔdlÑTmÔTmØ-<ğ +ñ >ô >ß>Dºf¹h¼hğ àsĞsĞsĞsĞsĞcrĞsÑsÔsˆ Ø Ô ğ 1ØoĞoĞoĞoĞoÕY^Õ_bĞclÑ_mÔ_mÑYnÔYnĞoÑoÔoˆGİ”h˜wÑ'Ô'ˆGؘ7 W¤]°1Ô%5Ñ5¸&¼+ÑEÑEˆGˆGå”g�s 9™~œ~Ğ /Ñ0Ô0ˆGؘ/¨7¯?ª?¸2¸aÑ+@Ô+@Ğ@Ğ@r c ó — d | _ d S r ©Ú
prioritiesr s r Úreset_prioritieszMemBuffer.reset_priorities‡ s
€ Ø ˆŒˆˆr Úuser_perÚalphac ó² ‡— |s| j n| j Š|s2t j t ‰¦ « f¦ « t ‰¦ « z S ˆfd„t t ‰¦ « ¦ « D ¦ « }t j |¦ « }||z }|| ¦ « z }| j | | j
j t t j |f¬¦ « |S )Nc óp •— g | ]2}t ‰| d ‰| d d z
¦ « ‘Œ3S )rs é ©Úabs)rb rp rS s €r rd z2MemBuffer.calculate_priorities.<locals>.<listcomp>� s: ø€ Ğ
EĞ
EĞ
E°�c�#�a”&˜”)˜c !œf Qœi¨œlÑ*Ñ+Ô+Ğ
EĞ
EĞ
Er )Úargs)r/ r0 ru rw r rP rv rx r* Úprocess_hook_executesÚcalculate_prioritiesr! Ú__file__r ÚALL)r r† r‡ r3 ÚpsrS s @r r� z MemBuffer.calculate_prioritiesŠ sÊ ø€ Ø!(Ğ>ˆdŒkˆk¨dÔ.>ˆØğ 3İ”7�C ™HœH˜;Ñ'Ô'#¨c©(¬(Ñ2Ğ 2Ø
EĞ
EĞ
EĞ
EµU½3¸s¹8¼8±_´_Ğ
EÑ
EÔ
Eˆİ
ŒX�b‰\Œ\ˆØ
�5‰[ˆØ
�"—&’&‘(”(‰]ˆØ Ô×/Ò/°°dÔ6OÔ6XÕZbÕdjÔdnØ68°Uğ 0ñ <ô <ğ <àˆ r rS ÚKc
ó ‡ ‡‡— t j t j t ‰¦ « ¦ « t t ‰ j ¦ « ‰z t |‰z d¦ « ¦ « d‰ j ¬¦ « ¦ « }ˆˆfd„|D ¦ « }ˆˆ fd„|D ¦ « }t t j |¦ « ¦ « t j
t t j |¦ « ¦ « t j ¬¦ « fS )Nrs F©rf re rg c
ó\ •— g | ](}t t j ‰||‰z ¦ « ¦ « ‘Œ)S r$ )r Ú itertoolsÚislice)rb rp r“ rS s €€r rd z+MemBuffer.simple_sample.<locals>.<listcomp>š s4 ø€ ĞOĞOĞO¸1••iÔ& s¨A¨q°1©uÑ5Ô5Ñ6Ô6ĞOĞOĞOr c ó4 •— g | ]}‰j ||‰z … ‘ŒS r$ rƒ )rb rp r“ r s €€r rd z+MemBuffer.simple_sample.<locals>.<listcomp>› s( ø€ ĞAĞAĞA¨Q�” ! a¡% Ô(ĞAĞAĞAr ©Údtype)ru rD ry rz r rQ r„ Úmaxr{ r r Ú
from_iterableÚthÚtensorÚfloat32)r rS rL r“ Úrandom_indexesrU Úpriss `` ` r Ú
simple_samplezMemBuffer.simple_sample– sö øøø€ İœ×)Ò)"¬)µC¸±H´HÑ*=Ô*=İ/2µ3°t´Ñ3GÔ3GÈ1Ñ3LÍcĞR\Ğ`aÑRaĞcdÑNeÔNeÑ/fÔ/fØ27¸4¼?ğ *ñ Lô LçLRÊFÉHÌHğ ğ PĞOĞOĞOĞOÀĞOÑOÔOˆØAĞAĞAĞAĞA°.ĞAÑAÔAˆİ•EÔ'¨Ñ.Ô.Ñ/Ô/µ´½4ÅÔ@SĞTXÑ@YÔ@YÑ;ZÔ;ZÕbdÔblĞ1mÑ1mÔ1mĞmĞmr c ó\ — t | j | j | j | j | j ¬¦ « S )N)r* )r& r- r' r( r) r* r s r Úmake_fresh_instancez MemBuffer.make_fresh_instance� s' € ݘœ ¨¬ °4´>À4Ä=Ğ_cÔ_pĞqÑqÔqĞqr c óN — t t | j ¦ « |dd„ ¬¦ « S )NTc ó — | S r r$ )rc s r ú<lambda>z)MemBuffer.to_dataloader.<locals>.<lambda>¢ s € Ğmn€ r )rL rN Ú
collate_fn)r r r/ rK s r Ú
to_dataloaderzMemBuffer.to_dataloader¡ s( € İ�* T¤[Ñ1Ô1¸jĞRVĞcnĞcnĞoÑoÔoĞor c ór — | j | | j j t t
j ¦ « d S r )r* r� Úrun_on_training_endr! r� r r‘ r s r r¬ z MemBuffer.run_on_training_end¤ s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr c óº — | j sIt | j ¦ « }t j |t ¦ « › d�¦ « t ¦ « › d�S | j j S )Nz/Pickles/memory_buffer.pt)r' r r/ r� Úsaver r= )r r/ s r r® zMemBuffer.save§ s] € ØŒyğ )ݘ$œ+Ñ&Ô&ˆFİ ŒG�FÕ 0Ñ 2Ô 2Ğ MĞ MĞ MÑ NÔ NĞ Nİ'Ñ)Ô)ĞDĞDĞDĞ Dà”;Ô(Ğ (r c ó — | j S r )r/ r s r Ú
get_bufferzMemBuffer.get_buffer¯ s
€ ØŒ{Ğr )FFNN)F)rA ) r! r" r# ÚboolÚstrr
r r: r. r@ ÚintrH rJ rN rU r rX r r[ r r� r… Úfloatr� r Útupler£ r¥ rª r¬ r® r° r$ r r r& r& # sg € € € € € Ø^bØ59ğ ğ tğ Àğ ĞX[ğ Ø *Ğ 2¨dğ ğ ğ ğ ğ
0ğ
0 tğ
0ğ
0ğ
0ğ
0ğ/ C K¨4ğ /ğ /ğ /ğ /ğ ğ ğ ğ %ğ %°cğ %ğ %ğ %ğ %ğ6ğ 6ğ 6ğ(ğ (˜tğ (ğ (ğ (ğ (ğ ğ ğ ¨dğ ğ ğ ğ ğ7ğ 7¨Dğ 7¸Tğ 7ğ 7ğ 7ğ 7ğ ğ ğ ğ%ğ %ğ %ğAğ A°ğ AÀ#ğ AÈ|ğ AĞfjğ Ağ Ağ Ağ Ağ0 ğ ğ ğ
ğ
¨Tğ
¸%ğ
È$ğ
ğ
ğ
ğ
ğn ğ n°Cğ n¸Cğ nÀEğ nğ nğ nğ nğrğ rğ rğpğ pğ pğoğ oğ oğ)ğ )ğ )ğğ ğ ğ ğ r r& c óJ — e Zd Zd„ Zd„ Zdefd„Zdefd„Zdefd„Zdefd„Z dS ) r6 c ó< — g | _ t d¦ « | _ d S )Nz-inf)rn r´ r` r s r r zSingleGameData.__init__´ s € Ø ˆŒİ$)¨&¡M¤MˆÔĞĞr c ó: — | j |¦ « d S r )rn r8 )r Úexperience_points r Úadd_data_pointz SingleGameData.add_data_point¸ s € Ø Œ× Ò Ğ /Ñ0Ô0Ğ0Ğ0Ğ0r r] c óª — |j sd S t | j ¦ « D �]3\ }}||j z }|t | j ¦ « k rS| j | j |j k r| j | j n| j | j }||j |j z z }nd}t | j ||dz … ¦ « D ]2\ }}||j |j k r|j n|j |j |z z z
}Œ3t | j | j |z
¦ « |j
z | j | _ t | j
| j | j ¦ « | _
�Œ5d S )Nr rs )r\ Ú enumeratern Ú num_td_stepsr ÚplayerÚvÚgammaÚrewardrŒ r‡ ro rœ r` )r r] rk Ú
data_pointÚ future_indexÚvalr Úd_points r Úcompute_initial_prioritiesz)SingleGameData.compute_initial_priorities» s… € ØÔ ğ Ø ˆFİ!*¨4¬?Ñ!;Ô!;ğ Zñ ZÑ ˆE�:Ø 6Ô#6Ñ6ˆLØ�c $¤/Ñ 2Ô 2Ò2Ğ2Ø9=¼ Ø=Iô:KÜKQĞU_ÔUfò:gğ :g�d”o lÔ3Ô5Ğ5à”O LÔ1Ô3ğm4ğ ğ ˜FœL¨FÔ,?Ñ ?Ñ?��à�İ )¨$¬/¸%À ĞPQÑAQĞ:QÔ*RÑ SÔ Sğ
}ğ
}‘ ��WØØ-4¬^¸zÔ?PÒ-PĞ-P˜7œ>˜>ĞW^ÔWeĞVeĞioÔiuĞy|Ñi|ñ}ñ }��å.1°$´/À%Ô2HÔ2JÈSÑ2PÑ.QÔ.QĞU[ÔUaÑ.aˆDŒO˜EÔ "Ô +å!$ TÔ%7¸¼ ÈÔ9OÔ9XÑ!YÔ!YˆDÔ Ñ ğ Zğ Zr c óx — |j s| S d}| j D ] }||j z
}Œ
| j D ]}|xj |z c_ Œ| S r, )r\ rn ro )r r] Úsum_Ú datapoints r ri zSingleGameData.normalizeÎ sd € ØÔ ğ ؈KØˆØ œ ğ 'ğ 'ˆIØ �IÔ&Ñ &ˆDˆDØ œ ğ 'ğ 'ˆIØ Ğ Ô $Ñ &Ğ Ô Ğ Øˆ r c ób — |j rpd„ | j D ¦ « }t j t j t | j ¦ « ¦ « d|¬¦ « ¦ « d }n*t j dt | j ¦ « dz
¦ « }| j | |fS )Nc ó — g | ] }|j ‘Œ
S r$ ©ro ra s r rd z2SingleGameData.sample_position.<locals>.<listcomp>Ú s € Ğ>Ğ>Ğ>¨˜!œ*Ğ>Ğ>Ğ>r ©rs )rf rg r rs )
r\ rn ÚnumpyrD ry ru rz r r{ Úrandint)r r] r„ rk s r Úsample_positionz SingleGameData.sample_positionØ s¡ € Ø Ô ğ @Ø>Ğ>¨d¬oĞ>Ñ>Ô>ˆJİ”L×'Ò'¬ µ#°d´oÑ2FÔ2FÑ(GÔ(GÈdØ*4ğ (ñ 6ô 6ß6<²f±h´h¸qôBˆEˆEõ ”N 1¥c¨$¬/Ñ&:Ô&:¸QÑ&>Ñ?Ô?ˆEàŒ˜uÔ% uĞ,Ğ,r c
ó — | |¦ « \ }}t ¦ « }t |||j z dz ¦ « D ]¿}|t | j ¦ « k r!| | j | ¦ « Œ;| t t j t j
|j f¦ « |j ¦ « ¦ « ddt j d|j dz
¦ « dd ¦ « ¦ « ŒÀ|S )Nrs r )rĞ r6 rP r“ r rn rº Ú DataPointru Údividerw Únet_action_sizer{ rD rÏ )r r] Úpositionrk Ú game_dataÚ unroll_indexs r rj z*SingleGameData.sample_position_with_unrollâ sş € Ø ×.Ò.¨vÑ6Ô6‰ˆ�%İ"Ñ$Ô$ˆ İ ! %¨°´Ñ)9¸AÑ)=Ñ >Ô >ğ ğ ˆLØ�c $¤/Ñ 2Ô 2Ò2Ğ2Ø×(Ò(¨¬ ¸ Ô)FÑGÔGĞGĞGà×(Ò(İ”I�bœg vÔ'=Ğ&?Ñ @Ô @À&ÔBXÑYÔY×`Ò`ÑbÔbØØİ”N 1 fÔ&<¸qÑ&@ÑAÔAØØñ
*ô *ñ ô ğ ğ ğ Ğr N)
r! r" r# r rº r rÆ ri rĞ rj r$ r r r6 r6 ³ s£ € € € € € ğ2ğ 2ğ 2ğ1ğ 1ğ 1ğZ° ğ Zğ Zğ Zğ Zğ&˜|ğ ğ ğ ğ ğ- \ğ -ğ -ğ -ğ -ğ°,ğ ğ ğ ğ ğ ğ r r6 c ó< — e Zd Zdedededededej pdf d„ZdS ) rÒ Úpir¿ rÁ Úmover¾ ÚframeNc óÎ — t |t ¦ « r d„ | ¦ « D ¦ « n|| _ || _ || _ || _ || _ || _ d | _ d S )Nc ó — g | ]}|‘ŒS r$ r$ ra s r rd z&DataPoint.__init__.<locals>.<listcomp>ö s € Ğ*Ğ*Ğ*˜�1Ğ*Ğ*Ğ*r )
r5 ÚdictÚvaluesrÙ r¿ rÁ rÚ r¾ rÛ ro )r rÙ r¿ rÁ rÚ r¾ rÛ s r r zDataPoint.__init__õ sb € İ.8¸½DÑ.AÔ.AĞIĞ*Ğ*˜bŸiši™kœkĞ*Ñ*Ô*Ğ*ÀrˆŒØˆŒØ ˆŒ ؈Œ Ø ˆŒ ؈Œ
Ø ˆŒ
ˆ
ˆ
r ) r! r" r# rŞ r´ r³ ru Úndarrayr r$ r r rÒ rÒ ô s\ € € € € € ğ ˜4ğ Eğ °5ğ Àğ ÈSğ ĞY[ÔYcĞYkĞgkğ ğ ğ ğ ğ ğ r rÒ c ó6 — e Zd Zdedefd„Zd„ Zd„ Zd„ Zd„ ZdS ) ÚMuZeroFrameBufferÚ noop_actionÚaction_space_sizec ó| — || _ || _ || _ t |¬¦ « t |¬¦ « dœ| _ d S )Nr> )rs rt )r- rã rä r Úbuffers)r Úframe_buffer_sizerã rä s r r zMuZeroFrameBuffer.__init__ sD € Ø)ˆŒ
Ø&ˆÔØ!2ˆÔ İ Ğ(9Ğ:Ñ:Ô:ÅĞM^Ğ@_Ñ@_Ô@_Ğ`Ğ`ˆŒ ˆ ˆ r c óJ — | j | ||f¦ « d S r )ræ r8 )r rÛ Úactionr¾ s r Ú add_framezMuZeroFrameBuffer.add_frame s' € Ø Œ �VÔ ×#Ò# U¨F OÑ4Ô4Ğ4Ğ4Ğ4r c ó\ — d„ | j | D ¦ « }t j |d¬¦ « S )Nc
óğ — g | ]s\ }}t j t j |t j ¬ ¦ « t j |j d |j d df|t j ¬ ¦ « fd¬¦ « ‘ŒtS )rš r rs rŠ ©Údim)r� ÚcatrŸ r Úfullr| )rb rÛ ré s r rd z3MuZeroFrameBuffer.concat_frames.<locals>.<listcomp>
s� € ğ Jğ Jğ Já#0 5¨&õ "œv¥r¤y°½b¼jĞ'IÑ'IÔ'Iİ')¤w°´ ¸A´ÀÄ ÈAÄĞPQĞ/RĞTZÕbdÔblĞ'mÑ'mÔ'mğ'oØtuğ wñ wô wğ Jğ Jğ Jr rŠ rí )ræ r� rï )r r¾ Úframes_with_actionss r Ú
concat_framesz MuZeroFrameBuffer.concat_frames sE € ğ Jğ Jà48´LÀÔ4Hğ Jñ Jô JĞõ ŒvĞ)¨qĞ1Ñ1Ô1Ğ1r c ó¨ — t | j ¦ « D ]<}| j | |t | j | j ¦ « f¦ « Œ=d S r )rP r- ræ r8 r rã rä )r Ú
init_stater¾ Ú_s r r. z MuZeroFrameBuffer.init_buffer s] € İ�t”}Ñ%Ô%ğ nğ nˆAØ ŒL˜Ô × 'Ò '¨µ\À$ÔBRĞTXÔTjÑ5kÔ5kĞ(lÑ mÔ mĞ mĞ mğ nğ nr c ó6 — t | j | ¦ « S r )r ræ )r r¾ s r r zMuZeroFrameBuffer.__len__ s € İ�4”< Ô'Ñ(Ô(Ğ(r N) r! r" r# r³ r rê rò r. r r$ r r râ râ ÿ s{ € € € € € ğa°sğ aÈsğ ağ ağ ağ ağ 5ğ 5ğ 5ğ2ğ 2ğ 2ğnğ nğ nğ)ğ )ğ )ğ )ğ )r râ c óR — e Zd Zd„ Zd„ Zd„ Zd„ Zd„ Zd„ Zdd„Z d „ Z
d
„ Z d „ Z d „ Z
d
S )ÚMongoDBMemBufferc ón — t j dd¦ « j | _ d| _ d| _ d| _ d S )NÚ localhosti‰i r F)ÚpymongoÚ MongoClientÚmuzeroÚdbÚcalculated_buffer_sizer2 r( r s r r zMongoDBMemBuffer.__init__ s3 € İÔ% k°5Ñ9Ô9Ô@ˆŒØ&'ˆÔ#Ø ˆŒ Ø ˆŒˆˆr c óŒ — t |t ¦ « st d¦ « ‚| j j |¦ « d S )NzExperience must be a dict)r5 rŞ r7 rş rÖ Úinsert©r r9 s r r: zMongoDBMemBuffer.add sB € ݘ*¥dÑ+Ô+ğ :İ Ğ 8Ñ9Ô9Ğ 9Ø ŒÔ× Ò Ñ,Ô,Ğ,Ğ,Ğ,r c óD — | j j |¦ « d S r )rş rÖ Ú insert_many©r rE s r rH zMongoDBMemBuffer.add_list$ s! € Ø ŒÔ×%Ò% oÑ6Ô6Ğ6Ğ6Ğ6r c ó — t j d| j j i ¦ « |z
¦ « }t | j j i ¦ « |¦ « |¦ « ¦ « S r, ) rD rÏ rş rÖ Úcount_documentsr ÚfindÚskipÚlimit)r rL Ú
random_idxs r rU zMongoDBMemBuffer.batch' sh € İ”^ A t¤wÔ'8×'HÒ'HÈÑ'LÔ'LÈzÑ'YÑZÔZˆ
İ�D”GÔ%×*Ò*¨2Ñ.Ô.×3Ò3°JÑ?Ô?×EÒEÀjÑQÔQÑRÔRĞRr c ól ‡— | j j i ¦ « | _ | j j i ddddœ¦ « }ˆfd„|D ¦ « }| j j i ddi¦ « }t ||¦ « D ])\ }}| j j |dd|ii¦ « Œ*d S )Nr rs )Ú_idÚ pred_rewardÚt_rewardc óR •— g | ]#}t |d |d z
¦ « ‰z ‘Œ$S )r r r‹ )rb rc r‡ s €r rd z9MongoDBMemBuffer.calculate_priorities.<locals>.<listcomp>. s4 ø€ Ğ
MĞ
MĞ
MÀ�c�!�MÔ" Q z¤]Ñ2Ñ3Ô3°uÑ<Ğ
MĞ
MĞ
Mr r
z$setro )rş rÖ r rÿ r ÚzipÚ
update_one) r rL r‡ r“ Úfieldsr’ Ú document_idsÚdoc_idrg s ` r r� z%MongoDBMemBuffer.calculate_priorities+ sÌ ø€ Ø&*¤gÔ&7×&GÒ&GÈÑ&KÔ&KˆÔ#Ø”Ô"×'Ò'¨°AÀaĞUVĞ,WĞ,WÑXÔXˆØ
MĞ
MĞ
MĞ
MÀfĞ
MÑ
MÔ
Mˆà”wÔ(×-Ò-¨b°5¸!°*Ñ=Ô=ˆ İ ˜\¨2Ñ.Ô.ğ Lğ L‰IˆF�AØ ŒGÔ × (Ò (¨°&¸:Àq¸/Ğ1JÑ KÔ KĞ KĞ Kğ Lğ Lr c ó’ — | j | j j i ¦ « k r | | j ||¦ « d S d S r )rÿ rş rÖ r r� )r r‡ r“ s r Úupdate_priorities_if_neededz,MongoDBMemBuffer.update_priorities_if_needed4 sM € Ø Ô &¨¬Ô):×)JÒ)JÈ2Ñ)NÔ)NÒ NĞ NØ × %Ò % dÔ&AÀ5È!Ñ LÔ LĞ LĞ LĞ Lğ OĞ Nr rs c # óL ‡ ‡‡
K — t |¦ « D �]�}‰ |‰¦ « t ‰ j j i dddœ¦ « d¦ « ¦ « }d„ ‰ j j i dddœ¦ « D ¦ « }t |¦ « Š
ˆ
fd„|D ¦ « }t j
t j ‰ j j i ¦ « ¦ « t ‰ j |‰z ¦ « d|¬¦ « ¦ « }ˆˆ fd „|D ¦ « } t t! j | ¦ « ¦ « } t% d
„ | D ¦ « ¦ « } | t' j |t& j ¬ ¦ « fV — �Œ�d S ) Nrs r )ro r
é c ó — g | ]
}|d ‘Œ S rÌ r$ ra s r rd z:MongoDBMemBuffer.batch_with_priorities.<locals>.<listcomp>= s € ĞgĞgĞg¨A˜!˜Jœ-ĞgĞgĞgr c ó •— g | ]}|‰z ‘ŒS r$ r$ ©rb rg Úsum_ps €r rd z:MongoDBMemBuffer.batch_with_priorities.<locals>.<listcomp>? s ø€ Ğ8Ğ8Ğ8¨˜!˜e™)Ğ8Ğ8Ğ8r Fr• c ó´ •— g | ]T}t ‰j j i ¦ « |¦ « ‰¦ « ¦ « ‘ŒUS r$ )r rş rÖ r r r
)rb rc r“ r s €€r rd z:MongoDBMemBuffer.batch_with_priorities.<locals>.<listcomp>C sP ø€ ĞXĞXĞXÈ1•T˜$œ'Ô+×0Ò0°Ñ4Ô4×9Ò9¸!Ñ<Ô<×BÒBÀ1ÑEÔEÑFÔFĞXĞXĞXr c óh — g | ]/}|d |d |d |d |d f|d f‘Œ0S )Ú
probabilitiesÚvsr Ú game_mover Ú
game_stater$ ra s r rd z:MongoDBMemBuffer.batch_with_priorities.<locals>.<listcomp>F sS € ğ ğ ğ Ğwx�!�OÔ$ a¨¤g°°*´
¸qÀ¼~ÈqĞQ^ÔO_Ğ/`ĞbcĞdpÔbqĞrğ ğ ğ r rš )rP r r rş rÖ r r
rx ru rD ry rz r rQ rÿ r{ r r� rµ r� rŸ r ) r ÚepochsrL r“ r‡ rõ Útest_pr„ ÚindexesÚitemsr s ` ` @r r� z&MongoDBMemBuffer.batch_with_priorities8 s¾ øøøè è € İ�v‘ ” ğ Añ AˆAØ × ,Ò ,¨U°AÑ 6Ô 6Ğ 6ݘ$œ'Ô+×0Ò0°À!ÈAĞ5NĞ5NÑOÔO×UÒUĞVWÑXÔXÑYÔYˆFàgĞg°´Ô1B×1GÒ1GÈĞYZĞcdĞLeĞLeÑ1fÔ1fĞgÑgÔgˆJݘ
‘O”OˆEØ8Ğ8Ğ8Ğ8¨ZĞ8Ñ8Ô8ˆJİ”i×&Ò&¥r¤y°´Ô1B×1RÒ1RĞSUÑ1VÔ1VÑ'WÔ'Wİ,/°Ô0KÈZĞ[\É_Ñ,]Ô,]ĞglØ)3ğ 'ñ 5ô 5ç5;²V±X´Xğ
ğ YĞXĞXĞXĞXĞPWĞXÑXÔXˆEİ�Ô,¨UÑ3Ô3Ñ4Ô4ˆEİğğ Øğñ ô ñ ô ˆEğ �œ :µR´ZĞ@Ñ@Ô@Ğ@Ğ @Ğ @Ğ @Ñ @ğ Ağ Ar c óh — | j j dt j fg¬¦ « d S )Nr
)Úsort)rş rÖ Úfind_onerû Ú
DESCENDINGr s r Úget_last_greatest_idz%MongoDBMemBuffer.get_last_greatest_idJ s. € ØŒwÔ ×)Ò)°½Ô8JĞ0KĞ/LĞ)ÑMÔMÈeÔTĞTr c ó@ — | j j i ¦ « S r )rş rÖ r r s r r zMongoDBMemBuffer.__len__M s € ØŒwÔ ×0Ò0°Ñ4Ô4Ğ4r c óB — | j j ¦ « d S r )rş rÖ Údropr s r Údrop_game_dataz MongoDBMemBuffer.drop_game_dataP s € Ø ŒÔ× Ò Ñ Ô Ğ Ğ Ğ r c ó — t ¦ « S r )rø r s r r¥ z$MongoDBMemBuffer.make_fresh_instanceS s € İ Ñ!Ô!Ğ!r NrÍ )r! r" r# r r: rH rU r� r r� r, r r0 r¥ r$ r r rø rø sÎ € € € € € ğ ğ ğ ğ -ğ -ğ -ğ
7ğ 7ğ 7ğSğ Sğ SğLğ Lğ LğMğ Mğ MğAğ Ağ Ağ Ağ$Uğ Uğ Uğ5ğ 5ğ 5ğ!ğ !ğ !ğ"ğ "ğ "ğ "ğ "r rø c óL — e Zd Zdefd„Zd„ Zd„ Zd„ Zd„ Zd
d„Z d „ Z
d
„ Z d „ Z d S )ÚPickleMemBufferÚ
pickle_dirc óX — || _ d| _ d| _ t |¦ « | _ d S )NT)r4 r2 r( r
Úpickler)r r4 s r r zPickleMemBuffer.__init__Y s* € Ø$ˆŒØˆŒ Ø ˆŒİ" :Ñ.Ô.ˆŒ ˆ ˆ r c ó — t d¦ « ‚)NzqSingle experience addition shouldn't be performed in PickleMemBuffer, please add entire bach with add_list method©ÚNotImplementedErrorr s r r: zPickleMemBuffer.add_ s € İ!ğ #Eñ Fô Fğ Fr c ó: — | j |¦ « d S r )r6 Ú
pickle_bufferr s r rH zPickleMemBuffer.add_listc s € Ø Œ ×"Ò" ?Ñ3Ô3Ğ3Ğ3Ğ3r c ó — t d¦ « ‚)NzRBatch method not implemented for PickleMemBuffer, please use batch_with_prioritiesr8 rK s r rU zPickleMemBuffer.batchf s € İ!Ğ"vÑwÔwĞwr c ó> ‡‡‡‡— | j d¦ « Š| j d¦ « Šˆˆˆfd„t t ‰¦ « ¦ « D ¦ « d | … }t d„ |D ¦ « ¦ « Šˆfd„|D ¦ « }d„ |D ¦ « S )Nrs rŠ c ób •— g | ]+}t ‰| ‰| d z
¦ « ‰z |f‘Œ,S )rŠ r‹ )rb rp r‡ Úrewsr! s €€€r rd z8PickleMemBuffer.calculate_priorities.<locals>.<listcomp>l s< ø€ Ğ
LĞ
LĞ
L¸�s�2�a”5˜4 œ7 1œ:Ñ%Ñ&Ô&¨%Ñ/°Ğ3Ğ
LĞ
LĞ
Lr c ó — g | ]
}|d ‘Œ S rm r$ ©rb rg s r rd z8PickleMemBuffer.calculate_priorities.<locals>.<listcomp>m s € Ğ&Ğ&Ğ&˜a�Q�q”TĞ&Ğ&Ğ&r c ó6 •— g | ]}|d ‰z |d f‘ŒS )r rs r$ r s €r rd z8PickleMemBuffer.calculate_priorities.<locals>.<listcomp>n s* ø€ Ğ
/Ğ
/Ğ
/ qˆq�Œt�e‰|˜Q˜qœTĞ"Ğ
/Ğ
/Ğ
/r c ó, — i | ]}|d |d “ŒS )rs r r$ rA s r ú
<dictcomp>z8PickleMemBuffer.calculate_priorities.<locals>.<dictcomp>o s" € Ğ'Ğ'Ğ'˜q��!”�a˜”dĞ'Ğ'Ğ'r )r6 Ú
load_indexrP r rx )r rL r‡ r“ r’ r? r r! s ` @@@r r� z$PickleMemBuffer.calculate_prioritiesi s¯ øøøø€ Ø
Œ\×
$Ò
$ QÑ
'Ô
'ˆØŒ|×&Ò& qÑ)Ô)ˆØ
LĞ
LĞ
LĞ
LĞ
LĞ
L½UÅ3ÀrÁ7Ä7¹^¼^Ğ
LÑ
LÔ
LÈSÈqÈbÈSÔ
QˆİĞ&Ğ& 2Ğ&Ñ&Ô&Ñ'Ô'ˆØ
/Ğ
/Ğ
/Ğ
/¨BĞ
/Ñ
/Ô
/ˆØ'Ğ' BĞ'Ñ'Ô'Ğ'r rs c # óÊ ‡‡ K — | ||‰¦ « Š t j t ‰ ¦ « ¦ « ¦ « }t |¦ « D �]}t j t j t ‰ ¦ « ¦ « t t ‰ ¦ « ‰z t |‰z d¦ « ¦ « dt |¦ « ¬¦ « ¦ « }| j
||‰¦ « }ˆˆ fd„|D ¦ « } t t j | ¦ « ¦ « }
|t# j |
t" j ¬¦ « fV — �Œd S )Nrs Fr• c óh •— g | ].}t ‰ ¦ « ¦ « ||‰z … ‘Œ/S r$ )r rß )rb rp r“ r„ s €€r rd z9PickleMemBuffer.batch_with_priorities.<locals>.<listcomp>y s: ø€ ĞOĞOĞO¸1•D˜×*Ò*Ñ,Ô,Ñ-Ô-¨a°°A±¨gÔ6ĞOĞOĞOr rš )r� ru rv r rß rP rD ry rz r rQ rœ r{ r6 Úload_allr r� r� rŸ r ) r r$ rL r“ r‡ Úps_probsrõ r¡ rU r¢ Útmpr„ s ` @r r� z%PickleMemBuffer.batch_with_prioritiesq sG øøè è € Ø×.Ò.¨z¸5À!ÑDÔDˆ
İ”8�D ×!2Ò!2Ñ!4Ô!4Ñ 5Ô 5Ñ6Ô6ˆİ�v‘ ” ğ :ñ :ˆAİ œY× -Ò -b¬i½¸J¹¼Ñ.HÔ.Hİ36µs¸:±´È!Ñ7KÍSĞQ[Ğ_`ÑQ`ĞbcÑMdÔMdÑ3eÔ3eĞotİ04°X±´ğ .ñ @ô @ç@FÂÁÄğ
ğ ”L×)Ò)¨*°nÀaÑHÔHˆEØOĞOĞOĞOĞOÀĞOÑOÔOˆDİ•uÔ*¨4Ñ0Ô0Ñ1Ô1ˆCØ�œ 3b¬jĞ9Ñ9Ô9Ğ9Ğ 9Ğ 9Ğ 9Ñ 9ğ :ğ :r c ó. — t d¦ « | j S )Nz/Load using datapickler.load_all(float(inf),...))Úprintr4 r s r r® zPickleMemBuffer.save} s € İ
Ğ?Ñ@Ô@Ğ@،Рr c ó* — t | j ¦ « S r )r3 r4 r s r r¥ z#PickleMemBuffer.make_fresh_instance� s € İ ˜tœÑ/Ô/Ğ/r c ó — t d¦ « ‚)Nz+Length not implemented for PickleMemBuffer.r8 r s r r zPickleMemBuffer.__len__„ s € İ!Ğ"OÑPÔPĞPr NrÍ )
r! r" r# r² r r: rH rU r� r� r® r¥ r r$ r r r3 r3 W s· € € € € € ğ/ 3ğ /ğ /ğ /ğ /ğ Fğ Fğ Fğ4ğ 4ğ 4ğxğ xğ xğ(ğ (ğ (ğ
:ğ
:ğ
:ğ
:ğ ğ ğ ğ0ğ 0ğ 0ğQğ Qğ Qğ Qğ Qr r3 )(r— rD ÚuuidÚ collectionsr r rÎ ru Ú numpy.randomrû Útorchr� Ú diskcacher Útorch.utils.datar r Ú mu_alpha_zero.General.memoryr Úmu_alpha_zero.General.utilsr Ú mu_alpha_zero.Hooks.hook_managerr
Ú mu_alpha_zero.Hooks.hook_pointr Ú mu_alpha_zero.MuZero.lazy_arraysr Ú mu_alpha_zero.MuZero.picklerr
Úmu_alpha_zero.MuZero.utilsr Úmu_alpha_zero.configr r r& r6 rÒ râ rø r3 r$ r r ú<module>r] sp ğØ Ğ Ğ Ğ Ø
€
€
€
Ø € € € Ø Ğ Ğ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø €€€Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ø 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0à <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 9Ğ 9Ğ 9Ğ 9Ğ 9Ğ 9Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø 6Ğ 6Ğ 6Ğ 6Ğ 6Ğ 6Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 3Ğ 3Ğ 3Ğ 3Ğ 3Ğ 3Ø -Ğ -Ğ -Ğ -Ğ -Ğ -ğ$ğ $ğ $ğ $ğ $�ñ $ô $ğ $ğMğ Mğ Mğ Mğ MĞ#ñ Mô Mğ Mğ`>ğ >ğ >ğ >ğ >ñ >ô >ğ >ğB ğ ğ ğ ğ ñ ô ğ ğ)ğ )ğ )ğ )ğ )ñ )ô )ğ )ğ2<"ğ <"ğ <"ğ <"ğ <"Ğ*ñ <"ô <"ğ <"ğ~.Qğ .Qğ .Qğ .Qğ .QĞ)ñ .Qô .Qğ .Qğ .Qğ .Qr | 34,763 | Python | .pyt | 105 | 330.047619 | 1,519 | 0.323523 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,551 | __init__.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/__pycache__/__init__.cpython-311.pyc | §
µ%îet ã óÂ — d dl mZ d dlmZ d dlmZ d dlmZ d dlm Z d dl
m Z d dl m
Z
d dlmZ d d lmZmZ d d
lmZmZ d d lmZ d d lmZ d d
lmZ d dl m Z d dl m Z dS )é )ÚMuZero)Ú AlphaZero)ÚGeneralMemoryBuffer)ÚGeneralNetwork)Ú GeneralArena)Ú
AlphaZeroGame)Ú
MuZeroGame)Ú
SearchTree)ÚSAMPLE_AZ_ARGSÚSAMPLE_MZ_ARGS)Úfind_project_rootÚclear_disk_data)ÚTicTacToeGameManager)Ú Asteroids)Ú MuZeroNet)Ú AlphaZeroNet)Ú
resize_obsN) Úmu_alpha_zero.MuZero.muzeror Ú"mu_alpha_zero.AlphaZero.alpha_zeror Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr Úmu_alpha_zero.General.arenar Ú mu_alpha_zero.General.az_gamer Ú mu_alpha_zero.General.mz_gamer Ú!mu_alpha_zero.General.search_treer
Ú!mu_alpha_zero.AlphaZero.constantsr r Úmu_alpha_zero.General.utilsr
r Ú!mu_alpha_zero.Game.tictactoe_gamer Ú mu_alpha_zero.Game.asteroidsr Ú%mu_alpha_zero.MuZero.Network.networksr Ú$mu_alpha_zero.AlphaZero.Network.nnetr Úmu_alpha_zero.MuZero.utilsr © ó úE/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/__init__.pyú<module>r& s% ðØ .Ð .Ð .Ð .Ð .Ð .Ø 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø <Ð <Ð <Ð <Ð <Ð <Ø 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø 4Ð 4Ð 4Ð 4Ð 4Ð 4Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7Ø 4Ð 4Ð 4Ð 4Ð 4Ð 4Ø 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø LÐ LÐ LÐ LÐ LÐ LÐ LÐ LØ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JØ BÐ BÐ BÐ BÐ BÐ BØ 2Ð 2Ð 2Ð 2Ð 2Ð 2Ø ;Ð ;Ð ;Ð ;Ð ;Ð ;Ø =Ð =Ð =Ð =Ð =Ð =Ø 1Ð 1Ð 1Ð 1Ð 1Ð 1Ð 1Ð 1r$ | 1,494 | Python | .pyt | 8 | 185.625 | 682 | 0.65232 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,552 | config.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/__pycache__/config.cpython-311.pyc | §
G¤f* ã ó˜ — d dl mZmZ e G d„ d¦ « ¦ « Ze G d„ de¦ « ¦ « Ze G d„ de¦ « ¦ « ZdS ) é )Ú dataclassÚfieldc óˆ — e Zd ZU dZeed< dZeed< dZeed< dZ eed< d Z
d Z d Z d
Z
eed < d Zeed
< dZeed< dZeed< dZeed< dZeed< dZeed< dZeed< dZeed< d Zeed < dZeed < d Zeed < dZeed!< d"Zd#Ze ed$< d%Z eed&< d'Z eed(< d)Z eed*< d'Z eed+< d,Z!epe"e ed-< d.Z#e ed/< d#Z$e ed0< d1Z%eed2< d3Z&eed4< dZ'eed5< e(d6„ ¬7¦ « Z)e"e ed8< d9Z*eed:< d;Z+e,ed<< d=Z-e,ed>< d=Z.e,ed?< d=Z/e,ed@< dAZ0eedB< d;Z1e,edC< d=Z2e,edD< dEZ3e edF< dG„ Z4e5dHe6fdI„¦ « Z7d#S )JÚConfigé Únum_net_channelsé Únum_net_in_channelsç333333Ó?Ú net_dropouté Únet_action_sizeé
é Úlinear_head_hidden_sizeé Únum_simulationséd Úself_play_gamesé2 Ú num_itersé@ ÚepochsçÜç…‰Õƒé>Úlrép Úmax_buffer_sizeé( Ú
num_pit_gamesé Úrandom_pit_freqéÿ Ú
batch_sizeÚtauç{®Gáz„?Ú arena_tauÚcgš™™™™™É?NÚcheckpoint_dirç333333ã?Úupdate_thresholdé Ú num_workersç•Ö&è .>Ú log_epsilonÚzero_tau_afteré Úaz_net_linear_input_sizeÚLogsÚlog_dirÚpushbullet_tokené Ú
num_blocksg-C ëâ6?Úl2Ú eval_epochsc ó
— ddgS )Né © r; ó úC/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/config.pyú<lambda>zConfig.<lambda>- s
€ ¸qÀ¸e€ r< )Údefault_factoryÚnet_latent_sizeé, Ú support_sizeTÚunravelFÚrequires_player_to_resetÚarena_running_muzeroÚ use_wandbé † Únum_worker_itersÚenable_frame_bufferÚrecalculate_p_on_every_callÚmzÚwandbd_project_namec ó — | j S )N)Ú__dict__)Úselfs r= Úto_dictzConfig.to_dict8 s
€ ØŒ}Ğ r< Úargsc óB — g }t ¦ « }| ¦ « D ]S\ }}|| ¦ « ¦ « vr| |¦ « t |||¦ « ŒTt
|¦ « dk rt d|› d�¦ « |S )Nr zHThe following keys were missing from the default config and were added: ú.)Ú MuZeroConfigÚitemsrP ÚkeysÚappendÚsetattrÚlenÚprint)rQ ÚmissingÚconfigÚkeyÚvals r= Ú from_argszConfig.from_args; s¤ € àˆİ ‘ ” ˆØ Ÿ
š
™ œ ğ &ğ &‰HˆC�ؘ&Ÿ.š.Ñ*Ô*×/Ò/Ñ1Ô1Ğ1Ğ1Ø—’˜sÑ#Ô#Ğ#İ �F˜C Ñ %Ô %Ğ %Ğ %å ˆw‰<Œ<˜!Ò Ğ İ ĞgĞ]dĞgĞgĞgÑ hÔ hĞ h؈
r< )8Ú__name__Ú
__module__Ú __qualname__r ÚintÚ__annotations__r
r Úfloatr Ústate_linear_layersÚpi_linear_layersÚv_linear_layersr r r r r r r r r! r# r$ r& r' Údirichlet_alphar( Ústrr* r, r. r/ r1 Úlistr3 r4 r6 r7 r8 r r@ rB rC ÚboolrD rE rF rH rI rJ rL rP Ú staticmethodÚdictr_ r; r< r= r r sä € € € € € € à Ğ�cĞ Ğ Ñ Ø Ğ˜Ğ Ğ Ñ Ø €K�Ğ Ğ Ñ Ø €O�SĞ Ğ Ñ Ø ĞØĞØ€OØ#&ИSĞ&Ğ&Ñ&Ø €O�SĞ Ğ Ñ Ø €O�SĞ Ğ Ñ Ø€IˆsĞĞÑØ€FˆCĞĞÑØ&€BˆĞ&Ğ&Ñ&Ø!€O�SĞ!Ğ!Ñ!Ø€M�3ĞĞÑØ €O�SĞ Ğ Ñ Ø€J�ĞĞÑØ€Cˆ€N€N�NØ€IˆuĞĞÑØ€A€u€L€L�LØ€OØ €N�CĞ Ğ Ñ Ø !Ğ�eĞ!Ğ!Ñ!Ø€K�ĞĞÑØ €K�Ğ Ğ Ñ Ø€N�CĞĞÑØ15Ğ ˜cĞ . T¨#¤YĞ5Ğ5Ñ5Ø€GˆSĞĞÑØ Ğ�cĞ Ğ Ñ Ø€J�ĞĞÑØ€BˆĞĞÑØ€K�ĞĞÑØ!& °|°|Ğ!DÑ!DÔ!D€O�T˜#”YĞDĞDÑDØ€L�#ĞĞÑØ€GˆTĞĞÑØ%*Ğ ˜dĞ*Ğ*Ñ*Ø!&И$Ğ&Ğ&Ñ&Ø€IˆtĞĞÑØ #Ğ�cĞ#Ğ#Ñ#Ø $ИĞ$Ğ$Ñ$Ø(-Ğ Ğ-Ğ-Ñ-Ø #ИĞ#Ğ#Ñ#ğ ğ ğ ğ ğ
˜ğ
ğ
ğ
ñ „\ğ
ğ
ğ
r< r c ó´ — e Zd ZU dZeed< dZeed< dZeed< dZeed< d Z e
ed
< d Z eed < d
Z eed< dZ
eed< dZeed< dZe
ed< dZeed< dZeed< dZeed< dZeed < d Zeed < d Ze
ed < d!Zeed"< d#Zeed$< d%Zeed&< d'Zeed(< dZe
ed)< d*Ze
ed+< dZ e
ed,< d-Z eed.< d/Z e
ed0< d1Z e ed2< d3Z!e
ed4< dZ"eed5< d6Z#e
ed7< dZ$eed8< dZ%eed9< d:Z&e ed;< d<Z'e(eef ed=< d>Z)epe*e ed?< d1Z+e ed@< d1Z,e edA< dBZ-e.edC< dDZ/e.edE< dDZ0e.edF< dDZ1e.edG< dDZ2e.edH< dBZ3e.edI< dDZ4e.edJ< dZ5e
edK< dDZ6e.edL< dMZ7eedN< d1S )OrT r r r Únum_net_out_channelsr r
é€ Úrep_input_channelsr r r
r r r r r r+ ÚKg�•C‹lçï?Úgammaé Úframe_buffer_sizeé Ú
frame_skipi� Ú num_stepsr r r r r r r r r r r r! r" r# r$ r% r& r' iÄL Úc2gš™™™™™é?ÚalphaNr( r) r* r, r- r. r/ Úbetaz Pickles/DataÚ
pickle_dir)é` r~ Útarget_resolutionr0 r1 r3 r4 FÚ show_tqdmTÚ
resize_imagesÚmuzeroÚ use_originalÚ use_poolingÚmultiple_playersÚ scale_stateÚ balance_termÚ
enable_perr Ú num_td_steps)8r` ra rb r rc rd rp r
rr r re r r r rs rt rv rx ry r r r r r r! r# r$ r& r' rz r{ r( rj r* r, r. r/ r| r} r Útupler1 rk r3 r4 r€ rl r� r‚ rƒ r„ r… r† r‡ rˆ r‰ r; r< r= rT rT I sş € € € € € € à Ğ�cĞ Ğ Ñ Ø #И#Ğ#Ğ#Ñ#Ø Ğ˜Ğ Ğ Ñ Ø !ИĞ!Ğ!Ñ!Ø €K�Ğ Ğ Ñ Ø €O�SĞ Ğ Ñ Ø €O�SĞ Ğ Ñ Ø €O�SĞ Ğ Ñ Ø
€A€s€J€J�JØ€Eˆ5ĞĞÑØ Ğ�sĞ Ğ Ñ Ø€J�ĞĞÑØ€IˆsĞĞÑØ€IˆsĞĞÑØ€FˆCĞĞÑØ&€BˆĞ&Ğ&Ñ&Ø!€O�SĞ!Ğ!Ñ!Ø€M�3ĞĞÑØ €O�SĞ Ğ Ñ Ø€J�ĞĞÑØ€Cˆ€N€N�NØ€IˆuĞĞÑØ€A€u€L€L�LØ€Bˆ€O€O�OØ€Eˆ5ĞĞÑØ €N�CĞ Ğ Ñ Ø !Ğ�eĞ!Ğ!Ñ!Ø€K�ĞĞÑØ €K�Ğ Ğ Ñ Ø€N�CĞĞÑØ€Dˆ#€M€M�MØ$€J�Ğ$Ğ$Ñ$Ø)1Ğ�u˜S #˜X”Ğ1Ğ1Ñ1Ø15Ğ ˜cĞ . T¨#¤YĞ5Ğ5Ñ5Ø€GˆSĞĞÑØ Ğ�cĞ Ğ Ñ Ø€IˆtĞĞÑØ €M�4Ğ Ğ Ñ Ø€FˆDĞĞÑØ €L�$Ğ Ğ Ñ Ø €K�Ğ Ğ Ñ Ø "Ğ�dĞ"Ğ"Ñ"Ø €K�Ğ Ğ Ñ Ø€L�%ĞĞÑØ€J�ĞĞÑØ€L�#ĞĞÑĞĞr< rT c ó¬ — e Zd ZU dZeed< dZeed< dZeed< dZe ed< ed z Z
eed
< d Z eed < d
Z eed< dZ
eed< dZeed< dZe ed< dZeed< dZeed< dZeed< dZeed < dZe ed < d Ze ed < dZe ed < d!Zeed"< d#Ze ed$< d%Zeed&< d'Ze ed(< d)Z eed*< d)Z eed+< d,Z e ed-< d)Z eed.< d/Z!eed0< d!Z"eed1< d!Z#eed2< d3Z$e ed4< d!S )5ÚAlphaZeroConfigr5 Ú
board_sizer r r r
r r r r i% r rA r r r iô r g{äL?´œj?r rG r r r é r! r r# r$ gëö‹ÓG1¥?r& r' Nr( r) r* rw Ú
minimax_depthTr€ r+ r, Ú
num_to_winr- r. r/ i H r1 r3 r4 Fr‚ )%r` ra rb r� rc rd r r
r re r r r r r r r r r! r# r$ r& r' r( rj r* r� r€ rl r, r� r. r/ r1 r3 r4 r‚ r; r< r= rŒ rŒ ~ så € € € € € € à€J�ĞĞÑØ Ğ�cĞ Ğ Ñ Ø Ğ˜Ğ Ğ Ñ Ø €K�Ğ Ğ Ñ Ø%¨™?€O�SĞ*Ğ*Ñ*Ø €O�SĞ Ğ Ñ Ø €O�SĞ Ğ Ñ Ø€IˆsĞĞÑØ€FˆCĞĞÑØ%€BˆĞ%Ğ%Ñ%Ø"€O�SĞ"Ğ"Ñ"Ø€M�3ĞĞÑØ €O�SĞ Ğ Ñ Ø€J�ĞĞÑØ€Cˆ€N€N�NØ*€IˆuĞ*Ğ*Ñ*Ø€A€u€L€L�LØ €N�CĞ Ğ Ñ Ø !Ğ�eĞ!Ğ!Ñ!Ø€M�3ĞĞÑØ€IˆtĞĞÑØ€K�ĞĞÑØ€J�ĞĞÑØ €K�Ğ Ğ Ñ Ø€N�CĞĞÑØ$)Ğ ˜cĞ)Ğ)Ñ)Ø€GˆSĞĞÑØ Ğ�cĞ Ğ Ñ Ø€FˆDĞĞÑĞĞr< rŒ N)Ú dataclassesr r r rT rŒ r; r< r= ú<module>r’ sË ğğ (Ğ 'Ğ 'Ğ 'Ğ 'Ğ 'Ğ 'Ğ 'ğ ğ:ğ :ğ :ğ :ğ :ñ :ô :ñ „ğ:ğz ğ.ğ .ğ .ğ .ğ .�6ñ .ô .ñ „ğ.ğh ğ ğ ğ ğ ğ �fñ ô ñ „ğ ğ ğ r< | 8,034 | Python | .pyt | 43 | 185.813953 | 1,158 | 0.413163 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,553 | memory.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/General/__pycache__/memory.cpython-311.pyc | §
føf¢ ã ó2 — d dl mZmZ G d„ de¦ « ZdS )é )ÚABCÚabstractmethodc ó| — e Zd Zed„ ¦ « Zed„ ¦ « Zed„ ¦ « Zed„ ¦ « Zed„ ¦ « ZdS )ÚGeneralMemoryBufferc ó — dS )z8
Add a single experience to the buffer.
N© )ÚselfÚ
experiences úK/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/General/memory.pyÚaddzGeneralMemoryBuffer.add ó € ð
ˆó c ó — dS )z:
Add a list of experiences to the buffer.
Nr )r Úexperience_lists r Úadd_listz GeneralMemoryBuffer.add_list r
r c ó — dS )z0
Return a batch of experiences.
Nr )r Ú
batch_sizes r ÚbatchzGeneralMemoryBuffer.batch r
r c ó — dS )z2
Return the length of the buffer.
Nr ©r s r Ú__len__zGeneralMemoryBuffer.__len__ r
r c ó — d S )Nr r s r Úmake_fresh_instancez'GeneralMemoryBuffer.make_fresh_instance! s € à ˆr N) Ú__name__Ú
__module__Ú __qualname__r r r r r r r r r r r s™ € € € € € Øð
ð
ñ „^ð
ð ð
ð
ñ „^ð
ð ð
ð
ñ „^ð
ð ð
ð
ñ „^ð
ð ð
ð
ñ „^ð
ð
ð
r r N)Úabcr r r r r r ú<module>r sQ ðØ #Ð #Ð #Ð #Ð #Ð #Ð #Ð #ð
ð
ð
ð
ð
˜#ñ
ô
ð
ð
ð
r | 1,621 | Python | .pyt | 14 | 110.142857 | 399 | 0.402363 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,554 | az_game.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/General/__pycache__/az_game.cpython-311.pyc | §
’Ô†f" ã óF — d dl mZmZ d dlmZ d dlZ G d„ de¦ « ZdS )é )ÚABCÚabstractmethod)Úadjust_probabilitiesNc ó2 — e Zd ZdZdedz fd„Zededefd„¦ « Ze d„ ¦ « Z
e d e d
e j
defd „¦ « Zee d
e j
d e de j
fd „¦ « ¦ « Ze d
e j
d
e ped e de j
fd„¦ « Ze d
e j
defd„¦ « Ze d
e j
d e pdde j
fd„¦ « Ze de j
fd„¦ « Ze d„ ¦ « Zd
e j
d e fd„Ze d„ ¦ « Ze d
e j
fd„¦ « Ze de j
d e fd„¦ « ZdS )Ú
AlphaZeroGamezP
Make your custom game extend this class and implement all the methods.
ÚreturnNc óœ — | ||¬¦ « rdS | | |¬¦ « rdS | |¬¦ « rdS dS )a.
Returns the result of the game from the perspective of the supplied player.
:param player: The player to check for (1 or -1).
:param board: The board to check on. If None, the current board is used.
:return: The game result. None when the game is not over yet.
)Úboardg ğ?g ğ¿g-C ëâ6?N)Ú check_winÚ
is_board_full©ÚselfÚplayerr
s úL/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/General/az_game.pyÚ game_resultzAlphaZeroGame.game_result sc € ğ �>Š>˜&¨ˆ>Ñ .Ô .ğ Ø�3Ø �>Š>˜6˜'¨ˆ>Ñ /Ô /ğ Ø�4Ø × Ò EĞ Ñ *Ô *ğ Ø�4؈tó Ú action_probsÚtauc ó¢ — t | |¬¦ « } t | ¦ « � \ }}t j ||¬¦ « S )N)r )Úp)r ÚzipÚitemsÚnpÚrandomÚchoice)r r ÚmovesÚprobss r Ú select_movezAlphaZeroGame.select_move sI € å+¨L¸cĞBÑBÔBˆ ݘL× .Ò .Ñ 0Ô 0Ğ1‰ ˆˆuİŒy× Ò ¨Ğ Ñ/Ô/Ğ/r c ó — dS )zD
Returns a fresh instance of the game (not a copy).
N© ©r s r Úmake_fresh_instancez!AlphaZeroGame.make_fresh_instance ó € ğ
ˆr r r
c ó — dS )zk
Checks if the current player 'player' won on the 'board'. If so returns true, else false.
Nr r
s r r zAlphaZeroGame.check_win' r# r c ó — dS )z¾
Returns the canonical form of the board.
In this case, that is a board where the current player is always player 1.
You probably want: return board * player
Nr ©r r
r s r Úget_canonical_formz AlphaZeroGame.get_canonical_form. s € ğ
ˆr Úactionc ó — dS )z‡
This method plays move given by 'board_index' as 'player' on the given 'board'
and returns the updated board.
Nr )r r
r( r s r Úget_next_statez AlphaZeroGame.get_next_state8 ó € ğ
ˆr c ó — dS )zj
Checks if the given board is completely full. Returns true if so, false other otherwise,
Nr )r r
s r r zAlphaZeroGame.is_board_full@ r# r c ó — dS )zŸ
Returns a list of valid moves for the given player on the given board.
If valid moves are independent of player, use None for player.
Nr r& s r Úget_valid_movesz AlphaZeroGame.get_valid_movesG r+ r c ó — dS )z6
Returns a copy of the current board.
Nr r! s r Ú get_boardzAlphaZeroGame.get_boardO r# r c ó — dS )z@
Resets the game and returns the initial board.
Nr r! s r ÚresetzAlphaZeroGame.resetV r# r c ó — dS )z§
Optional. Evaluates the board from the perspective of the given player.
This is used for minimax, override this method if you want to use it.
Nr r& s r Ú
eval_boardzAlphaZeroGame.eval_board] r# r c ó — dS )z-
Render the GUI of the game.
Nr r! s r ÚrenderzAlphaZeroGame.renderd r# r c ó — dS )zZ
Returns a random valid action for the given board and optionally player.
Nr )r r
Úkwargss r Úget_random_valid_actionz%AlphaZeroGame.get_random_valid_actionk r# r Ústatec ó — dS )zI
Returns a list of invalid actions for the given player.
Nr )r r: r s r Úget_invalid_actionsz!AlphaZeroGame.get_invalid_actionsr r# r )Ú__name__Ú
__module__Ú __qualname__Ú__doc__Úfloatr Ú staticmethodÚdictr r r" Úintr ÚndarrayÚboolr r' Útupler* r r. r0 r2 r4 r6 r9 r< r r r r r s� € € € € € ğğ ğ
¨E°D©Lğ
ğ
ğ
ğ
ğ ğ0 $ğ 0¨Uğ 0ğ 0ğ 0ñ „\ğ0ğ
ğ
ğ
ñ „^ğ
ğ ğ
ğ
¨B¬Jğ
¸4ğ
ğ
ğ
ñ „^ğ
ğ Øğ
¨¬
ğ
¸Cğ
ÀBÄJğ
ğ
ğ
ñ „^ñ „\ğ
ğ ğ
B¤Jğ
¸¸ ¸uğ
Ècğ
ĞVXÔV`ğ
ğ
ğ
ñ „^ğ
ğ ğ
2¤:ğ
°$ğ
ğ
ğ
ñ „^ğ
ğ ğ
R¤Zğ
¸¸Àğ
ÈÌğ
ğ
ğ
ñ „^ğ
ğ ğ
˜2œ:ğ
ğ
ğ
ñ „^ğ
ğ ğ
ğ
ñ „^ğ
ğ
¤
ğ
°Cğ
ğ
ğ
ğ
ğ ğ
ğ
ñ „^ğ
ğ ğ
¨R¬Zğ
ğ
ğ
ñ „^ğ
ğ ğ
¨¬ğ
¸Sğ
ğ
ğ
ñ „^ğ
ğ
ğ
r r )Úabcr r Úmu_alpha_zero.General.utilsr Únumpyr r r r r ú<module>rK sy ğØ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ø Ğ Ğ Ğ ğq
ğ q
ğ q
ğ q
ğ q
�Cñ q
ô q
ğ q
ğ q
ğ q
r | 5,992 | Python | .pyt | 58 | 97.465517 | 706 | 0.436731 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,555 | utils.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/General/__pycache__/utils.cpython-311.pyc | §
’Ô†f; ã ó˜ — d dl Z d dlZd dlZdefd„Zd„ Zdefd„Zdefd„Z G d„ d¦ « Zd „ Z d
„ Z
de e fd „Z dd
e
de
fd„ZdS )é NÚreturnc óò — t j t j t ¦ « ¦ « } t j | ¦ « dt j | ¦ « vrzt j ¦ « dk st j ¦ « dk rt d¦ « ‚t j d¦ « t j ¦ « } dt j | ¦ « v°zt j ¦ « S )Nz root.rootú/zC:\z Could not find project root.z..) ÚosÚpathÚdirnameÚabspathÚ__file__ÚchdirÚlistdirÚgetcwdÚFileNotFoundError)Údir_s úJ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/General/utils.pyÚfind_project_rootr s² € İ
Œ7�?Š?�2œ7Ÿ?š?8Ñ4Ô4Ñ 5Ô 5€Dİ„HˆT�N„N€NØ
�RœZ¨Ñ -Ô -Ğ
-Ğ
-İ
Œ9‰;Œ;˜#Ò Ğ ¥¤¡¤°Ò!6Ğ!6İ#Ğ$BÑCÔCĞ Cİ
Œ�‰ŒˆİŒy‰{Œ{ˆğ �RœZ¨Ñ -Ô -Ğ
-Ğ
-õ
Œ9‰;Œ;Ğó c ó — | dk r| ndS )Nr é © )Úxs r Únot_zeror s € Ø�Q’�ˆ1ˆ1˜AĞ r c ó� — t j t j t j ¦ « ¦ « S ©N)r r r ÚsysÚ
executabler r r Úget_python_homer s& € İ
Œ7�?Š?�2œ7Ÿ?š?3¬>Ñ:Ô:Ñ ;Ô ;Ğ;r c ó¶ — t j d¦ « } | d¦ « d d¦ « d ¦ « S )Nzpip show torchz Location:r ÚRequiresr )Ú
subprocessÚ getoutputÚsplitÚstrip)Úoutputs r Úget_site_packages_pathr$ sI € İ
Ô
!Ğ"2Ñ
3Ô
3€FØ �<Š<˜ Ñ $Ô $ QÔ '× -Ò -¨jÑ 9Ô 9¸!Ô <× BÒ BÑ DÔ DĞDr c ó6 — e Zd Zg fdefd„Zd„ Zd„ Zd„ Zd„ ZdS )Ú
OnlyUniqueÚiterablec ó> — g | _ | |¦ « d S r )ÚuniqueÚextend)Úselfr' s r Ú__init__zOnlyUnique.__init__ s! € ؈Œ Ø � Š �HÑ Ô Ğ Ğ Ğ r c óP — || j vr | j |¦ « d S d S r )r) Úappend)r+ Úitems r ÚaddzOnlyUnique.add# s3 € Ø �t”{Ğ "Ğ "Ø ŒK× Ò ˜tÑ $Ô $Ğ $Ğ $Ğ $ğ #Ğ "r c ó: — |D ]}| |¦ « Œd S r )r0 )r+ r' r/ s r r* zOnlyUnique.extend' s, € Ø ğ ğ ˆDØ �HŠH�T‰NŒNˆNˆNğ ğ r c ó* — t | j ¦ « S r )Úlenr) ©r+ s r Ú__len__zOnlyUnique.__len__+ s € İ�4”;Ñ Ô Ğ r c ó — | j S r )r) r4 s r ÚgetzOnlyUnique.get. s
€ ØŒ{Ğr N) Ú__name__Ú
__module__Ú __qualname__Úlistr, r0 r* r5 r7 r r r r& r& ss € € € € € Ø(*ğ ğ ğ ğ ğ ğ ğ%ğ %ğ %ğğ ğ ğ ğ ğ ğğ ğ ğ ğ r r& c óˆ — t ¦ « › d�} t j | ¦ « D ]}t j | › d|› �¦ « Œ d S )Nz
/Pickles/Datar )r r r Úremove)r Úfiles r Úclear_disk_datar? 2 sZ € İ Ñ!Ô!Ğ 0Ğ 0Ğ 0€Dİ”
˜4Ñ Ô ğ $ğ $ˆİ
Œ �TĞ"Ğ"˜DĞ"Ğ"Ñ#Ô#Ğ#Ğ#ğ$ğ $r c ó — | €
J d¦ « ‚d S )NzNetwork is None, can't train/predict/pit. Make sure you initialize the network with eitherload_checkpoint or create_new method.r )Únets r Ú net_not_nonerB 8 s € Ø ˆ?ˆ?ğ 0ñ Œ?ˆ?ˆ?ˆ?r c óş — t ¦ « dd¦ « d¦ « d } d„ t t j | › d� j ¦ « ¦ « D ¦ « S )Nú\r éÿÿÿÿc ó< — g | ]}| d ¦ « ¯|‘ŒS )ÚPlayer)Úendswith©Ú.0r s r ú
<listcomp>z get_players.<locals>.<listcomp>@ s8 € ğ "ğ "ğ "�!Ø
�JŠJ�xÑ Ô ğ "ˆAğ "ğ "ğ "r z.AlphaZero.Arena.players)r Úreplacer! r; r ÚmodulesÚ__dict__Úkeys)Ú path_prefixs r Ú get_playersrQ > sw € İ#Ñ%Ô%×-Ò-¨d°CÑ8Ô8×>Ò>¸sÑCÔCÀBÔG€Kğ "ğ "•t�CœK¨;Ğ(PĞ(PĞ(PÔ QÔ Z× _Ò _Ñ aÔ aÑbÔbğ "ñ "ô "ğ "r ç ğ?Úaction_probabilitiesc ó ‡‡ — ‰dk ršd„ | ¦ « D ¦ « }| t |¦ « ¦ « }d„ t t |¦ « ¦ « D ¦ « }d||< t t
| ¦ « |¦ « ¦ « S t
| ¦ « � \ }}ˆfd„|D ¦ « }t |¦ « Š ˆ fd„|D ¦ « }t t
||¦ « ¦ « S )aı
Selects a move from the action probabilities using either greedy or stochastic policy.
The stochastic policy uses the tau parameter to adjust the probabilities. This is based on the
temperature parameter in DeepMind's AlphaZero paper.
:param action_probabilities: A dictionary containing the action probabilities in the form of {action_index: probability}.
:param tau: The temperature parameter. 0 for greedy, >0 for stochastic.
:return: The selected move as an integer (index).
r c ó — g | ]}|‘ŒS r r rI s r rK z(adjust_probabilities.<locals>.<listcomp>O s € Ğ9Ğ9Ğ9�a�Ğ9Ğ9Ğ9r c ó — g | ]}d ‘ŒS )r r )rJ Ú_s r rK z(adjust_probabilities.<locals>.<listcomp>Q s € Ğ-Ğ-Ğ-�q�Ğ-Ğ-Ğ-r r c ó •— g | ]
}|d ‰z z ‘Œ S )r r )rJ ÚprobÚtaus €r rK z(adjust_probabilities.<locals>.<listcomp>V s" ø€ ĞBĞBĞB¨D�d˜q 3™wÑ'ĞBĞBĞBr c ó •— g | ]}|‰z ‘ŒS r r )rJ rY Úadjusted_probs_sums €r rK z(adjust_probabilities.<locals>.<listcomp>X s ø€ ĞMĞMĞM°d˜Ğ 1Ñ1ĞMĞMĞMr )
ÚvaluesÚindexÚmaxÚranger3 ÚdictÚziprO ÚitemsÚsum)
rS rZ ÚvalsÚmax_idxÚprobsÚmovesÚ
probabilitiesÚadjusted_probsÚnormalized_probsr\ s
` @r Úadjust_probabilitiesrl D s øø€ ğ ˆa‚x€xØ9Ğ9Ğ/×6Ò6Ñ8Ô8Ğ9Ñ9Ô9ˆØ—*’*�S ™YœYÑ'Ô'ˆØ-Ğ-�E¥# d¡)¤)Ñ ,Ô ,Ğ-Ñ-Ô-ˆØˆˆg‰İ•CĞ,×1Ò1Ñ3Ô3°UÑ;Ô;Ñ<Ô<Ğ<å Ğ 4× :Ò :Ñ <Ô <Ğ=Ñ€Eˆ=ØBĞBĞBĞB°MĞBÑBÔB€Nİ ˜^Ñ,Ô,ĞØMĞMĞMĞM¸nĞMÑMÔMĞİ •�EĞ+Ñ,Ô,Ñ -Ô -Ğ-r )rR )r r r Ústrr r r r$ r& r? rB r; rQ ra rl r r r ú<module>rn s ğØ € € € Ø Ğ Ğ Ğ Ø
€
€
€
ğ˜3ğ ğ ğ ğ ğ ğ ğ ğ<˜ğ <ğ <ğ <ğ <ğE ğ Eğ Eğ Eğ Eğ
ğ ğ ğ ğ ñ ô ğ ğ($ğ $ğ $ğ 1ğ 1ğ 1ğ "�T˜#”Yğ "ğ "ğ "ğ "ğ .ğ .¨tğ .Àğ .ğ .ğ .ğ .ğ .ğ .r | 7,465 | Python | .pyt | 36 | 205.527778 | 966 | 0.344192 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,556 | search_tree.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/General/__pycache__/search_tree.cpython-311.pyc | §
føfÙ ã óZ — d dl mZmZ d dlZd dlZd dlmZ d dl m
Z
G d„ de¦ « Z dS )é )ÚABCÚabstractmethodN)ÚGeneralMemoryBuffer)ÚGeneralNetworkc ó — e Zd Zeddej depddeee e e f fd„¦ « Z
e dde j de pddej de
pdfd „¦ « Zed
„ ¦ « Zed e pdfd „¦ « Zed
edej de dedee e e f f
d„¦ « ZdS )Ú
SearchTreeNÚdeviceÚdir_pathÚreturnc ó — dS )z4
Performs one game of the algorithm
N© )ÚselfÚnetworkr r
s úP/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/General/search_tree.pyÚ
play_one_gamezSearchTree.play_one_game ó € ğ
ˆó ÚstateÚcurrent_playerÚtauc ó — dS )zG
Performs MCTS search for given number of simulations.
Nr
)r r r r r r s r ÚsearchzSearchTree.search s € ğ
ˆr c ó — dS )z4
Return new instance of this class.
Nr
)r s r Úmake_fresh_instancez SearchTree.make_fresh_instance r r Úactionc ó — dS )z:
Steps the root node to the given action.
Nr
)r r s r Ú step_rootzSearchTree.step_root# r r ÚnetÚ num_gamesÚmemoryc ó — dS )z?
Performs self play for given number of games.
Nr
)r r r r r s r Ú self_playzSearchTree.self_play* s € ğ
ˆr )N)Ú__name__Ú
__module__Ú __qualname__r Úthr ÚstrÚtupleÚlistÚintr ÚnpÚndarrayÚfloatr r r r r r" r
r r r r
sT € € € € € àğ
ğ
¨R¬Yğ
ÀÀÈğ
ĞX]Ğ^bĞdgĞilĞnqĞ^qÔXrğ
ğ
ğ
ñ „^ğ
ğ à$(ğ
ğ
R¤Zğ
ÀÀÈğ
ĞVXÔV_ğ
Ø�M˜Tğ
ğ
ğ
ñ „^ğ
ğ ğ
ğ
ñ „^ğ
ğ ğ
tğ
ğ
ğ
ñ „^ğ
ğ ğ
˜^ğ
°R´Yğ
È3ğ
ĞXkğ
ĞpuØ ˆS�#ˆ
ôqğ
ğ
ğ
ñ „^ğ
ğ
ğ
r r ) Úabcr r Únumpyr+ Útorchr& Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr r r
r r ú<module>r3 s� ğØ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #à Ğ Ğ Ğ Ø Ğ Ğ Ğ à <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8ğ&
ğ &
ğ &
ğ &
ğ &
�ñ &
ô &
ğ &
ğ &
ğ &
r | 2,471 | Python | .pyt | 22 | 107.636364 | 660 | 0.433061 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,557 | network.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/General/__pycache__/network.cpython-311.pyc | §
"¨žf¢
ã ó– — d dl mZmZ d dlmZ d dlmZ d dlmZ d dl Z
G d„ de¦ « Z G d„ d e ¦ « Z G d
„ d e ¦ « Z
dS ) é )ÚABCÚabstractmethod)Ú
MuZeroGame)Ú HookManager)ÚConfigNc ó¶ — e Zd Zed„ ¦ « Zeed
dedepdfd„¦ « ¦ « Zedede e
e e
f fd„¦ « Z ededdfd „¦ « Z
dS ) ÚGeneralNetworkc ó — dS )z9
Returns a fresh instance of the network
N© )Úselfs úL/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/General/network.pyÚmake_fresh_instancez"GeneralNetwork.make_fresh_instance ó € ð
ˆó NÚconfigÚ hook_managerc ó — dS )zC
Builds the network from the given arguments dict.
Nr )Úclsr r s r
Úmake_from_configz GeneralNetwork.make_from_config s € ð
ˆr Úmuzero_alphazero_configÚreturnc ó — dS )z?
Trains the network for given number of epochs
Nr ©r Ú
memory_bufferr s r
Ú train_netzGeneralNetwork.train_net r r c ó — dS )a/
Evaluates the network against the evaluation dataset and reports directly to wandb.
:param memory_buffer: The memory buffer where the datasets should be pulled from.
:param muzero_alphazero_config: The config for this algorithm.
:return: None, reports to wandb.
Nr r s r
Úeval_netzGeneralNetwork.eval_net! ó € ð
ˆr )N)Ú__name__Ú
__module__Ú __qualname__r r Ú classmethodr r r ÚtupleÚfloatÚlistr r r r r
r r
sÝ € € € € € Øð
ð
ñ „^ð
ð Øð
ð
fð
¸KÐ<OÈ4ð
ð
ð
ñ „^ñ „[ð
ð ð
Àð
È5ÐQVÐX\Ð]bÔXcÐQcÔKdð
ð
ð
ñ „^ð
ð ð
¸vð
È$ð
ð
ð
ñ „^ð
ð
ð
r r c ó — e Zd Zed dej dedej ej ffd„¦ « Zed dej dedej ej ffd„¦ « Zedej dej fd„¦ « Z edej d ej dej fd
„¦ « Z
d S )
ÚGeneralMuZeroNetworkFÚxÚpredictr c ó — dS )aa
Forward pass for the dynamics network. Should operate only on the hidden state, not the actual game state.
Call representation_forward first.
:param x: The input hidden state.
:param predict: Whether to predict or do pure forward pass.
:return: A tuple of the next hidden state and the immediate reward.
Nr ©r r( r) s r
Údynamics_forwardz%GeneralMuZeroNetwork.dynamics_forward. s € ð
ˆr c ó — dS )a�
Forward pass for the prediction network. As dynamics forward, this should operate only on the hidden state,
not the actual game state.
Call representation_forward first.
:param x: The input hidden state.
:param predict: Whether to predict or do pure forward pass.
:return: A tuple of the action probability distribution and the value of the current state.
Nr r+ s r
Úprediction_forwardz'GeneralMuZeroNetwork.prediction_forward9 s € ð
ˆr c ó — dS )z·
Forward pass for the representation network. This should operate on the actual game state.
:param x: The input game state.
:return: The hidden state.
Nr )r r( s r
Úrepresentation_forwardz+GeneralMuZeroNetwork.representation_forwardE s € ð
ˆr Úy_hatÚyc ó — dS )zò
Calculate the loss for the action probability distribution.
:param y_hat: The predicted action probability distribution.
:param y: The MCTS improved action probability distribution.
:return: The loss.
Nr )r r1 r2 s r
Ú muzero_lossz GeneralMuZeroNetwork.muzero_lossN r r N)F) r r r! r ÚthÚTensorÚboolr, r. r0 r4 r r r
r' r' , s
€ € € € € àð
ð
"¤)ð
°dð
ÈÌ ÐSUÔS\ÐG]ð
ð
ð
ñ „^ð
ð ð
ð
B¤Ið
¸ð
È"Ì)ÐUWÔU^ÐI_ð
ð
ð
ñ „^ð
ð ð
¨¬ ð
°b´ið
ð
ð
ñ „^ð
ð ð
¤ð
¨r¬yð
¸R¼Yð
ð
ð
ñ „^ð
ð
ð
r r' c óV — e Zd Zeddej dedej ej ffd„¦ « ZdS )ÚGeneralAlphZeroNetworkTr( Úmuzeror c ó — dS )a0
Predict the action probability distribution and the value of the current state.
:param x: The input game state.
:param muzero: Whether to predict for MuZero or AlphaZero.
:return: A tuple of the action probability distribution and the value of the current state.
Nr )r r( r: s r
r) z GeneralAlphZeroNetwork.predictZ r r N)T)r r r! r r5 r6 r7 r) r r r
r9 r9 Y sW € € € € € Øð
ð
˜œð
¨Dð
¸R¼YÈÌ Ð<Rð
ð
ð
ñ „^ð
ð
ð
r r9 )Úabcr r Ú mu_alpha_zero.General.mz_gamer Ú mu_alpha_zero.Hooks.hook_managerr Úmu_alpha_zero.configr Útorchr5 r r' r9 r r r
ú<module>rA sé ðØ #Ð #Ð #Ð #Ð #Ð #Ð #Ð #à 4Ð 4Ð 4Ð 4Ð 4Ð 4à 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø 'Ð 'Ð 'Ð 'Ð 'Ð 'Ø Ð Ð Ð ð
ð
ð
ð
ð
�Sñ
ô
ð
ðD*
ð *
ð *
ð *
ð *
˜>ñ *
ô *
ð *
ðZ
ð
ð
ð
ð
˜^ñ
ô
ð
ð
ð
r | 5,721 | Python | .pyt | 51 | 105.196078 | 695 | 0.492506 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,558 | mz_game.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/General/__pycache__/mz_game.cpython-311.pyc | §
¯a�fc
ã óN — d dl mZmZ d dlZd dlZd dlmZ G d„ de¦ « Z dS )é )ÚABCÚabstractmethodN)Úadjust_probabilitiesc óú — e Zd Zededepddej pej ee ffd„¦ « Z
edej pej fd„¦ « Z edefd„¦ « Z edefd„¦ « Z
edepdde pdfd „¦ « Zed
„ ¦ « Zed „ ¦ « Zeddedepdd
edej pej ee ffd„¦ « Zededefd„¦ « Zedej defd„¦ « Zedej fd„¦ « Zedej defd„¦ « ZdS )Ú
MuZeroGameÚactionÚplayerNÚreturnc ó — dS )a„
Given a game state and an action return the next state. Currently this implementation only supports one player, so supply None.
:param state: The current state of the game.
:param action: The action to be taken.
:param player: The player taking the action.
:return: The next state, the reward and a boolean indicating if the game is done.
N© )Úselfr r s úL/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/General/mz_game.pyÚget_next_statezMuZeroGame.get_next_state
s € ğ
ˆó c ó — dS )z@
Resets the game and returns the initial board.
Nr ©r
s r ÚresetzMuZeroGame.reset ó € ğ
ˆr c ó — dS )z*
Returns the noop action.
Nr r s r Úget_noopzMuZeroGame.get_noop r r c ó — dS )zT
Returns the number of actions possible in the current environment.
Nr r s r Úget_num_actionszMuZeroGame.get_num_actions$ r r c ó — dS )zã
Returns true if this environment is done, false otherwise. If this environment doesn't have a clear terminal
state, return None. Currently this implementation only supports one player, so supply None.
Nr )r
r s r Ú game_resultzMuZeroGame.game_result+ ó € ğ
ˆr c ó — dS )z=
Return fresh instance of this game manager.
Nr r s r Úmake_fresh_instancez MuZeroGame.make_fresh_instance3 r r c ó — dS )z-
Render the GUI of the game.
Nr r s r ÚrenderzMuZeroGame.render: r r é Ú
frame_skipc ó — dS )zj
Wrapper for the frame skip method. This method should be used instead of get_next_state.
Nr )r
r r r! s r Úframe_skip_stepzMuZeroGame.frame_skip_stepA r r Ú action_probsÚtauc ó¢ — t | |¬¦ « } t | ¦ « � \ }}t j ||¬¦ « S )z?
Samples a move from the action probabilities.
)r% )Úp)r ÚzipÚitemsÚnpÚrandomÚchoice)r$ r% ÚmovesÚprobss r Ú select_movezMuZeroGame.select_moveI sK € õ
,¨L¸cĞBÑBÔBˆ ݘL× .Ò .Ñ 0Ô 0Ğ1‰ ˆˆuİŒy× Ò ¨Ğ Ñ/Ô/Ğ/r Ústatec ó — dS )zÿ
Calculates and returns the invalid actions in the given state.
:return: A numpy array containing the invalid
actions in the current state. In this array actions marked as valid will be one, while invalid actions will
be 0.Nr ©r
r0 r s r Úget_invalid_actionsz MuZeroGame.get_invalid_actionsR ó € ğ
ˆr c ó — dS )z]
Returns a random valid action in the given state and optionally the player.
Nr )r
r0 Úkwargss r Úget_random_valid_actionz"MuZeroGame.get_random_valid_action[ r r c ó — dS )aV
Returns a state with information specific for the provided player. This should only be done when a part of
the actual observation is player specific (for example the player's cards in a card game). In other cases
this method should return the provided state as is (for example in chess or other board games).
Nr r2 s r Ú get_state_for_passive_playerz'MuZeroGame.get_state_for_passive_playerb r4 r )r )Ú__name__Ú
__module__Ú __qualname__r Úintr* ÚndarrayÚthÚTensorÚboolr r r r r r r r# Ú staticmethodÚdictÚfloatr/ r3 r7 r9 r r r r r sY € € € € € àğ
Sğ
°#°+¸ğ
Ø ŒJĞ #˜"œ) S¨$ğC0ğ
ğ
ğ
ñ „^ğ
ğ ğ
�r”zĞ. R¤Yğ
ğ
ğ
ñ „^ğ
ğ ğ
˜#ğ
ğ
ğ
ñ „^ğ
ğ ğ
ğ
ğ
ğ
ñ „^ğ
ğ ğ
# +¨ğ
°$°,¸$ğ
ğ
ğ
ñ „^ğ
ğ ğ
ğ
ñ „^ğ
ğ ğ
ğ
ñ „^ğ
ğ ğ
ğ
cğ
°3°;¸$ğ
ÈCğ
Ø ŒJĞ #˜"œ) S¨$ğY0ğ
ğ
ğ
ñ „^ğ
ğ ğ0 $ğ 0¨Uğ 0ğ 0ğ 0ñ „\ğ0ğ ğ
¨¬ğ
¸Sğ
ğ
ğ
ñ „^ğ
ğ ğ
¨R¬Zğ
ğ
ğ
ñ „^ğ
ğ ğ
°"´*ğ
Àcğ
ğ
ğ
ñ „^ğ
ğ
ğ
r r )
Úabcr r Únumpyr* Útorchr? Úmu_alpha_zero.General.utilsr r r r r ú<module>rI s… ğØ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø <Ğ <Ğ <Ğ <Ğ <Ğ <ğa
ğ a
ğ a
ğ a
ğ a
�ñ a
ô a
ğ a
ğ a
ğ a
r | 5,517 | Python | .pyt | 44 | 118.363636 | 799 | 0.47004 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,559 | arena.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/General/__pycache__/arena.cpython-311.pyc | §
Ái¡f
ã óz — d dl Z d dlmZmZ d dlmZ d dlZd dlmZm Z m
Z
d dl m Z d dl
mZ G d„ de¦ « ZdS ) é N)ÚabstractmethodÚABC)ÚType)ÚPlayerÚ NetPlayerÚ RandomPlayer)Ú CheckPointer)Ú
SharedStoragec óŒ — e Zd Ze ddee dee dedededef d „¦ « Z dde de d
e
deded e d e dedefd
„Z
dS )Ú GeneralArenaFé Úplayer1Úplayer2Únum_games_to_playÚnum_mc_simulationsÚ
one_playerÚ start_playerc ó — d S )N© )Úselfr r r r r r s úJ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/General/arena.pyÚpitzGeneralArena.pit
s € ğ
ˆó Ú
player_2_2Úshared_storageÚ checkpointerc
óø — t j dd¬¦ « t | d¦ « r| j n| j }
|j ¦ « |j ¦ « d} t |
j ¦ « D �]{} | ¦ « }
|
�|
¦ « rt j d¦ « ŒB|j
|
¦ « |j
| ¦ « ¦ « | |||||| ¬¦ « \ }}}t j ||dœ¦ « d „ }| |||z ¦ « z |
j k rU| |
¦ « | d
z
} | |j |j | ¦ « |
j | |
¦ « | d ¦ « | |||||| ¬¦ « \ }}}t j ||d œ¦ « �Œ}d S )
NÚMZz Arena Pit)ÚprojectÚnameÚ
muzero_configr é )r r )Ú
wins_p1_vs_p2Ú
wins_p2_vs_p1c ó — | dk r| ndS )Nr r
r )Úxs r ú<lambda>z-GeneralArena.continuous_pit.<locals>.<lambda>' s € a¨1¢f f °!€ r r
T)Úwins_p1_vs_randomÚwins_random_vs_p1)ÚwandbÚinitÚhasattrr! Úalpha_zero_configÚnetworkÚevalÚrangeÚnum_worker_itersÚ get_experimental_network_paramsÚget_was_pittedÚtimeÚsleepÚload_state_dictÚget_stable_network_paramsr ÚlogÚupdate_thresholdÚset_stable_network_paramsÚsave_checkpointÚ
get_optimizerÚlrÚset_was_pitted)r r r r r r r r r r ÚconfÚ
accept_numÚiter_Ú
tested_paramsÚ
results_p1Ú
results_p2Ú_Únot_zeros r Úcontinuous_pitzGeneralArena.continuous_pit s& € õ
Œ
˜4 kĞ2Ñ2Ô2Ğ2İ%,¨T°?Ñ%CÔ%CĞ_ˆtÔ!Ğ!ÈÔI_ˆØŒ× Ò Ñ Ô Ğ ØŒ× Ò Ñ Ô Ğ Øˆ
ݘ4Ô0Ñ1Ô1ğ Zñ ZˆEØ *× JÒ JÑ LÔ LˆMØ Ğ$¨×(EÒ(EÑ(GÔ(GĞ$İ”
˜1‘
”
�
ØØ ŒO× +Ò +¨MÑ :Ô :Ğ :Ø ŒO× +Ò +¨N×,TÒ,TÑ,VÔ,VÑ WÔ WĞ WØ(,¯ª°¸'ĞCTĞVhØ<FĞUağ )1ñ )cô )cÑ %ˆJ˜
Aå ŒI¨
ÀZĞPĞPÑ QÔ QĞ QØ3Ğ3ˆHؘH˜H Z°*Ñ%<Ñ =Ô =Ñ=ÀÔAVÒVĞVØ ×8Ò8¸ ÑGÔGĞGؘa‘�
Ø ×,Ò,¨W¬_¸g¼oÈ~×OkÒOkÑOmÔOmĞosÔovØ-7¸ñ?ô ?ğ ?à × )Ò )¨$Ñ /Ô /Ğ /à(,¯ª°¸*ĞFWĞYkØ<FĞUağ )1ñ )cô )cÑ %ˆJ˜
Aå ŒI¨JÈZĞXĞXÑ YÔ YĞ YÑ Yğ) Zğ Zr N)Fr
)Ú__name__Ú
__module__Ú __qualname__r r r ÚintÚboolr r r r
r rG r r r r r sñ € € € € € ØàPUØ !ğ
ğ
˜4 œ<ğ
°$°v´,ğ
Ø "ğ
Ø8;ğ
ØIMğ
à ğ
ğ
ğ
ñ „^ğ
ğ FGğ Zğ Z iğ Z¸)ğ ZĞQ]ğ ZĞruğ ZØ+.ğ Zà'4ğ Zğ &2ğ Zğ $(ğ Zğ @Cğ Zğ Zğ Zğ Zğ Zğ Zr r )r4 Úabcr r Útypingr r* Ú%mu_alpha_zero.AlphaZero.Arena.playersr r r Ú$mu_alpha_zero.AlphaZero.checkpointerr Ú$mu_alpha_zero.shared_storage_managerr
r r r r ú<module>rR sÁ ğØ € € € Ø #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ø Ğ Ğ Ğ Ğ Ğ à € € € à QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QØ =Ğ =Ğ =Ğ =Ğ =Ğ =Ø >Ğ >Ğ >Ğ >Ğ >Ğ >ğ%Zğ %Zğ %Zğ %Zğ %Z�3ñ %Zô %Zğ %Zğ %Zğ %Zr | 4,033 | Python | .pyt | 31 | 128.967742 | 855 | 0.394954 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,560 | checkpointer.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/checkpointer.cpython-311.pyc | §
-*¢fê ã ól — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dlm Z d dl
m Z G d„ d¦ « Z dS )é N)ÚDotDict)Úfind_project_root)ÚConfigc óˆ — e Zd Zd$dedz deddfd„Zd%d„Z d&dej j d ej j d
ej
d e d e d
e
deddfd„Zdej j deddfd„Zdej j deddfd„Zdedefd„Zde defd„Zd%d„Zd%d„Zd%d„Zde fd„Zde fd„Zdefd„Zdefd „Zdefd „Zd „ Z d eddfd „Z d!„ Z d"e e fd#„Z dS )'Ú CheckPointerTÚcheckpoint_dirNÚverboseÚreturnc ó¼ — || _ | ¦ « | ¦ « | _ d| _ || _ t
j | j ¦ « d S )NÚ
improved_net_) Ú _CheckPointer__checkpoint_dirÚmake_dirÚinitialize_checkpoint_numÚ _CheckPointer__checkpoint_numÚ_CheckPointer__name_prefixr ÚatexitÚregisterÚcleanup)Úselfr r s úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/checkpointer.pyÚ__init__zCheckPointer.__init__ sR € Ø .ˆÔ Ø �
Š
‰ŒˆØ $× >Ò >Ñ @Ô @ˆÔ Ø ,ˆÔØ ˆŒ İŒ˜œ Ñ%Ô%Ğ%Ğ%Ğ%ó c ó® — | j � t j | j d¬¦ « d S t ¦ « }|› d�}|| _ t j |d¬¦ « d S )NT©Úexist_okz/Checkpoints/NetVersions)r
ÚosÚmakedirsr )r Úroot_dirr s r r zCheckPointer.make_dir sd € Ø Ô Ğ ,İ ŒK˜Ô-¸Ğ =Ñ =Ô =Ğ =Ø ˆFİ$Ñ&Ô&ˆØ $Ğ>Ğ>Ğ>ˆØ .ˆÔ İ
Œ �N¨TĞ2Ñ2Ô2Ğ2Ğ2Ğ2r ÚnetÚopponentÚ optimizerÚlrÚ iterationÚmu_alpha_zero_configÚnamec ó¾ — |€ | j t | j ¦ « z }| j › d|› d�}t j | ¦ « t |t j j ¦ « r| ¦ « n||||
¦ « | ¦ « dœ|¦ « | d|› d|› d�¦ « | xj dz
c_ d S )Nú/ú.pth)r r! r" r# ÚargsÚopponent_state_dictzSaved checkpoint to z at iteration ú.é ) r Ústrr r
ÚthÚsaveÚ
state_dictÚ
isinstanceÚoptimÚ OptimizerÚto_dictÚ
print_verbose) r r r r! r" r# r$ r% Úcheckpoint_paths r Úsave_checkpointz CheckPointer.save_checkpoint sü € ğ ˆ<ØÔ%¨DÔ,AÑ(BÔ(BÑBˆDà !Ô 2Ğ?Ğ?°TĞ?Ğ?Ğ?ˆİ
ŒØ—>’>Ñ#Ô#İ3=¸iÍÌÔI[Ñ3\Ô3\Ğk˜×-Ò-Ñ/Ô/Ğ/ĞbkØØ"Ø(×0Ò0Ñ2Ô2Ø#+×#6Ò#6Ñ#8Ô#8ğ
ğ
ğ ñ ô ğ ğ
×ÒĞ]°/Ğ]Ğ]ĞQZĞ]Ğ]Ğ]Ñ^Ô^Ğ^Ø Ğ Ô Ñ"Ğ Ô Ğ Ğ r c ó˜ — | j › d|› d�}t j | ¦ « |¦ « | d¦ « d S )Nr' r( z Saved state dict checkpoint.)r
r. r/ r0 r5 ©r r r% r6 s r Úsave_state_dict_checkpointz'CheckPointer.save_state_dict_checkpoint1 sR € Ø !Ô 2Ğ?Ğ?°TĞ?Ğ?Ğ?ˆİ
Œ�—’Ñ Ô /Ñ2Ô2Ğ2Ø ×ÒĞ:Ñ;Ô;Ğ;Ğ;Ğ;r c ó˜ — | j › d|› d�}| t j |¦ « ¦ « | d¦ « d S )Nr' r( z Loaded state dict checkpoint.)r
Úload_state_dictr. Úloadr5 r9 s r Úload_state_dict_checkpointz'CheckPointer.load_state_dict_checkpoint6 sT € Ø !Ô 2Ğ?Ğ?°TĞ?Ğ?Ğ?ˆØ ×Ò�BœG OÑ 4Ô 4Ñ5Ô5Ğ5Ø ×ÒĞ;Ñ<Ô<Ğ<Ğ<Ğ<r r6 c ó0 — t j d t j d< t j |¦ « }| d|› d|d › d�¦ « d|v r |d }nd }|d |d ||d
t |d ¦ « |d fS )
Nzmu_alpha_zero.mem_bufferÚ
mem_bufferzRestoring checkpoint z made at iteration r# r+ Úmemoryr r! r" r) r* )ÚsysÚmodulesr. r= r5 r )r r6 Ú
checkpointrA s r Úload_checkpoint_from_pathz&CheckPointer.load_checkpoint_from_path; s¬ € İ$'¤KĞ0JÔ$K�Œ �LÑ!İ”W˜_Ñ-Ô-ˆ
Ø ×ÒĞq°?ĞqĞqĞWaĞbmÔWnĞqĞqĞqÑrÔrĞrØ �zĞ !Ğ !Ø Ô)ˆFˆFàˆFؘ%Ô *¨[Ô"9¸6À:ÈdÔCSİ �J˜vÔ&Ñ 'Ô '¨Ğ4IÔ)JğKğ Kr Úcheckpoint_numc óT — | j › d| j › |› d�}| |¦ « S )Nr' r( )r
r rE )r rF r6 s r Úload_checkpoint_from_numz%CheckPointer.load_checkpoint_from_numF s7 € Ø !Ô 2Ğ]Ğ]°TÔ5GĞ]È Ğ]Ğ]Ğ]ˆØ×-Ò-¨oÑ>Ô>Ğ>r c óR — t d¦ « t d¦ « }|dk rt d¦ « d S t j | j ¦ « D ] }t j | j › d|› �¦ « Œ!t dt
t j | j ¦ « ¦ « › d�¦ « d S )NzClearing all checkpoints.zAre you sure?? (y/n): ÚyzAborted.r' zCleared z saved checkpoints (all).)ÚprintÚinputr Úlistdirr
ÚremoveÚlen)r ÚanswerÚ file_names r Úclear_checkpointsz CheckPointer.clear_checkpointsJ s² € õ Ğ)Ñ*Ô*Ğ*İĞ/Ñ0Ô0ˆØ �SŠ=ˆ=İ �*Ñ Ô Ğ Ø ˆFİœ DÔ$9Ñ:Ô:ğ >ğ >ˆIİ ŒI˜Ô.Ğ<Ğ<°Ğ<Ğ<Ñ =Ô =Ğ =Ğ =İ
ĞZ��RœZ¨Ô(=Ñ >Ô >Ñ?Ô?ĞZĞZĞZÑ[Ô[Ğ[Ğ[Ğ[r c óĞ — t j ¦ « }t j | j › d�d¬¦ « | j › d|› d�}t j | ¦ « |¦ « d S )Nz/TempTr ú/Temp/temp_net_r( )r Úgetpidr r
r. r/ r0 ©r r Ú process_pidr6 s r Úsave_temp_net_checkpointz%CheckPointer.save_temp_net_checkpointV si € İ”i‘k”kˆ İ
Œ �tÔ,Ğ3Ğ3Ğ3¸dĞCÑCÔCĞCØ !Ô 2ĞTĞTÀ;ĞTĞTĞTˆİ
Œ�—’Ñ Ô /Ñ2Ô2Ğ2Ğ2Ğ2r c ó” — t j ¦ « }| j › d|› d�}| t j |¦ « ¦ « d S )NrT r( )r rU r
r< r. r= rV s r Úload_temp_net_checkpointz%CheckPointer.load_temp_net_checkpoint\ sJ € İ”i‘k”kˆ Ø !Ô 2ĞTĞTÀ;ĞTĞTĞTˆØ ×Ò�BœG OÑ 4Ô 4Ñ5Ô5Ğ5Ğ5Ğ5r c ób — t d„ t j | j ¦ « D ¦ « ¦ « S )Nc ó< — g | ]}| d ¦ « ¯|‘ŒS )r( )Úendswith)Ú.0Úxs r ú
<listcomp>z:CheckPointer.initialize_checkpoint_num.<locals>.<listcomp>b s) € ĞWĞWĞW˜!ÀAÇJÂJÈvÑDVÔDVĞW�AĞWĞWĞWr )rO r rM r
©r s r r z&CheckPointer.initialize_checkpoint_numa s, € İĞWĞW�rœz¨$Ô*?Ñ @Ô @ĞWÑWÔWÑXÔXĞXr c ób — t d„ t j | j ¦ « D ¦ « ¦ « S )Nc ó� — g | ]C}t | d ¦ « d d¦ « d ¦ « ‘ŒDS )Ú_é r+ r )ÚintÚsplit)r^ rQ s r r` z;CheckPointer.get_highest_checkpoint_num.<locals>.<listcomp>e sD € ĞrĞrĞrÀ9•C˜ Ÿš¨Ñ,Ô,¨QÔ/×5Ò5°cÑ:Ô:¸1Ô=Ñ>Ô>ĞrĞrĞrr )Úmaxr rM r
ra s r Úget_highest_checkpoint_numz'CheckPointer.get_highest_checkpoint_numd s/ € İĞrĞrÕPRÔPZĞ[_Ô[pÑPqÔPqĞrÑrÔrÑsÔsĞsr c ó — | j › d�S )Nz/Temp/temp_net.pth©r
ra s r Ú
get_temp_pathzCheckPointer.get_temp_pathg s € ØÔ'Ğ;Ğ;Ğ;Ğ;r c ó — | j S ©Nrk ra s r Úget_checkpoint_dirz CheckPointer.get_checkpoint_dirj s € ØÔ$Ğ$r c ó� ‡ ‡— ˆˆ fd„t j ‰ j ¦ « D ¦ « }| d„ ¬¦ « |d S )Nc ób •— g | ]+}‰|v ¯t j ‰j |¦ « ‘Œ,S © )r ÚpathÚjoinr
)r^ r_ r% r s €€r r` z6CheckPointer.get_latest_name_match.<locals>.<listcomp>n s8 ø€ ĞwĞwĞwÀ1ĞmqĞuvĞmvĞmv�œŸ š TÔ%:¸AÑ>Ô>ĞmvĞmvĞmvr c ó@ — t j | ¦ « S rn )r rs Úgetctime)r_ s r ú<lambda>z4CheckPointer.get_latest_name_match.<locals>.<lambda>o s € ¬×(8Ò(8¸Ñ(;Ô(;€ r )Úkeyéÿÿÿÿ)r rM r
Úsort)r r% Ú name_matchess `` r Úget_latest_name_matchz"CheckPointer.get_latest_name_matchm sT øø€ ØwĞwĞwĞwĞwÍÌ
ĞSWÔShÑHiÔHiĞwÑwÔwˆ Ø×ÒĞ ;Ğ ;ĞÑ<Ô<Ğ<ؘBÔ Ğ r c ó — | j S rn )r ra s r Úget_name_prefixz CheckPointer.get_name_prefixr s € ØÔ!Ğ!r Úmsgc ó6 — | j rt |¦ « d S d S rn )r rK )r r s r r5 zCheckPointer.print_verboseu s% € Ø Œ<ğ İ �#‰JŒJˆJˆJˆJğ ğ r c óL — dd l }| | j › d�d¬¦ « d S )Nr z/Temp/T)Ú
ignore_errors)ÚshutilÚrmtreer
)r rƒ s r r zCheckPointer.cleanupy s3 € ؈
ˆ
ˆ
�
Š
˜Ô.Ğ6Ğ6Ğ6Àdˆ
ÑKÔKĞKĞKĞKr Úlossesc ó’ — t | j › d�d¦ « 5 }t j ||¦ « d d d ¦ « d S # 1 swxY w Y d S )Nz/training_losses.pklÚwb)Úopenr
ÚpickleÚdump)r r… Úfs r Ú save_losseszCheckPointer.save_losses} s‘ € İ
�TÔ*Ğ@Ğ@Ğ@À$Ñ
GÔ
Gğ #È1İ ŒK˜ Ñ "Ô "Ğ "ğ #ğ #ğ #ñ #ô #ğ #ğ #ğ #ğ #ğ #ğ #ğ #øøøğ #ğ #ğ #ğ #ğ #ğ #s ™<¼A ÁA )T)r
Nrn )!Ú__name__Ú
__module__Ú __qualname__r- Úboolr r r. ÚnnÚModuler2 Úfloatrf r r7 r: r> ÚtuplerE rH rR rX rZ r ri rl ro r| r~ r5 r ÚlistrŒ rr r r r r
sš € € € € € ğ&ğ & s¨T¡zğ &¸Dğ &ÈDğ &ğ &ğ &ğ &ğ3ğ 3ğ 3ğ 3ğ SWğ#ğ # 2¤5¤<ğ #¸2¼5¼<ğ #ĞTVÔT\ğ #Ø !ğ#à#&ğ#à>Dğ#àLOğ#à[_ğ#ğ #ğ #ğ #ğ$<¨b¬e¬lğ <À#ğ <È$ğ <ğ <ğ <ğ <ğ
=¨b¬e¬lğ =À#ğ =È$ğ =ğ =ğ =ğ =ğ
K¸ğ KÀğ Kğ Kğ Kğ Kğ?°sğ ?¸uğ ?ğ ?ğ ?ğ ?ğ
\ğ
\ğ
\ğ
\ğ3ğ 3ğ 3ğ 3ğ 6ğ 6ğ 6ğ 6ğ
Y¨3ğ Yğ Yğ Yğ Yğt¨Cğ tğ tğ tğ tğ<˜sğ <ğ <ğ <ğ <ğ% Cğ %ğ %ğ %ğ %ğ ¨#ğ ğ ğ ğ ğ
"ğ "ğ "ğ ğ ¨ğ ğ ğ ğ ğLğ Lğ Lğ# $ u¤+ğ #ğ #ğ #ğ #ğ #ğ #r r )
r r r‰ rB Útorchr. Ú mu_alpha_zero.AlphaZero.utilsr Úmu_alpha_zero.General.utilsr Úmu_alpha_zero.configr r rr r r ú<module>rš s§ ğØ
€
€
€
Ø € € € Ø
€
€
€
Ø
€
€
€
à Ğ Ğ Ğ à 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø 9Ğ 9Ğ 9Ğ 9Ğ 9Ğ 9Ø 'Ğ 'Ğ 'Ğ 'Ğ 'Ğ 'ğr#ğ r#ğ r#ğ r#ğ r#ñ r#ô r#ğ r#ğ r#ğ r#r | 11,468 | Python | .pyt | 35 | 326.571429 | 2,143 | 0.360679 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,561 | utils.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/utils.cpython-311.pyc | §
weŠfº* ã
ó4 — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dl m
Z
d dl m Z
d dlmZ d dlmZmZ G d„ de¦ « Zd ed
efd „Zd ej d
efd„Zd ej defd„Zd efd„Zd ej fd„Z dej dej d
ej fd„Z dej d
ej fd„Z defd„Z ded
efd„Z!d ed ed
efd „Z"d ed e#d!e#d"e#d#ef
d$„Z$d%ed
e
fd&„Z%d6d'ed
e&e
ej' j( ef fd(„Z)d)e#dedz fd*„Z*d+„ Z+d7d-ed.e,fd/„Z-d8d0e#d1edz fd2„Z.d3ed4ed efd5„Z/dS )9é N)Ú get_ipython)Ú AlphaZeroNet)ÚSAMPLE_AZ_ARGS)Ú MemBuffer)ÚConfigÚAlphaZeroConfigc ó — e Zd Zd„ Zd„ ZdS )ÚDotDictc ó — | | S ©N© )ÚselfÚnames úL/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/utils.pyÚ __getattr__zDotDict.__getattr__ s € Ø�DŒzĞó c ó — || |< d S r r
)r r Úvalues r Ú __setattr__zDotDict.__setattr__ s € ؈ˆT‰
ˆ
ˆ
r N)Ú__name__Ú
__module__Ú __qualname__r r r
r r r
r
s2 € € € € € ğğ ğ ğğ ğ ğ ğ r r
Úgame_experienceÚreturnc ó& — g }| D �]Š\ }}}}t j d„ | ¦ « D ¦ « ¦ « }| |||f¦ « t ddgddg¦ « D �]$\ }}t j | ¦ « |¬¦ « } t j | ¦ « ||¦ « |¬¦ « ¦ « }
| | |
|f¦ « ~ ~
t j | ¦ « |¬¦ « } t j | ¦ « ||¦ « |¬¦ « ¦ « }
| | |
|f¦ « �Œ&�ŒŒ|S )Nc ó — g | ]}|‘ŒS r
r
)Ú.0Úxs r ú
<listcomp>z6augment_experience_with_symmetries.<locals>.<listcomp> s € Ğ.Ğ.Ğ.˜Q�qĞ.Ğ.Ğ.r r é é ©Úk©Úaxis)
ÚnpÚarrayÚvaluesÚappendÚzipÚrot90ÚcopyÚreshapeÚflattenÚflip) r Ú
board_sizeÚgame_experience_ÚstateÚpiÚvÚ_r% r# Ústate_Úpi_s r Ú"augment_experience_with_symmetriesr8 sw € ØĞØ*ğ
6ñ
6‰ˆˆr�1�aİ
ŒXĞ.Ğ. "§)¢)¡+¤+Ğ.Ñ.Ô.Ñ
/Ô
/ˆØ× Ò ¨¨A Ñ/Ô/Ğ/ݘA˜q˜6 A q 6Ñ*Ô*ğ 6ñ 6‰GˆD�!İ”X˜eŸjšj™lœl¨aĞ0Ñ0Ô0ˆFİ”(˜2Ÿ7š7™9œ9×,Ò,¨Z¸ÑDÔDÈĞJÑJÔJ×RÒRÑTÔTˆCØ × #Ò # V¨S°!Ğ$4Ñ 5Ô 5Ğ 5ؘݔW˜UŸZšZ™\œ\°Ğ5Ñ5Ô5ˆFİ”'˜"Ÿ'š'™)œ)×+Ò+¨J¸
ÑCÔCÈ$ĞOÑOÔO×WÒWÑYÔYˆCØ × #Ò # V¨S°!Ğ$4Ñ 5Ô 5Ğ 5Ñ 5ñ 6ğ Ğr r2 r# c ó~ — t | j d ¦ « D ]!}t j | | |¬¦ « | |< Œ"| S )Nr r" )ÚrangeÚshaper& r+ )r2 r# Údims r Ú rotate_stackr= * sB € İ�U”[ ”^Ñ$Ô$ğ /ğ /ˆİ”X˜e Cœj¨AĞ.Ñ.Ô.ˆˆc‰
ˆ
Ø €Lr r% c ó~ — t | j d ¦ « D ]!}t j | | |¬¦ « | |< Œ"| S )Nr r$ )r: r; r& r/ )r2 r% r< s r Ú
flip_stackr? 0 sB € İ�U”[ ”^Ñ$Ô$ğ 4ğ 4ˆİ”W˜U 3œZ¨dĞ3Ñ3Ô3ˆˆc‰
ˆ
Ø €Lr c ór — g }| D ]1\ }}}}}t |¦ « }| |||||f¦ « Œ2|S r )Úmake_channels_from_singler) )r Ú
experiencer2 r3 r4 Úcurrent_playerÚmoves r Ú
make_channelsrE 6 sX € Ø€JØ.=ğ @ğ @Ñ*ˆˆr�1�n dİ)¨%Ñ0Ô0ˆØ×Ò˜5 " a¨ ¸Ğ>Ñ?Ô?Ğ?Ğ?à Ğr c óÒ — t j | dk dd¦ « }t j | dk dd¦ « }t j | dk dd¦ « }t j | |||gd¬¦ « S )Nr r éÿÿÿÿr$ )r& ÚwhereÚstack)r2 Úplayer_one_stateÚplayer_minus_one_stateÚ empty_states r rA rA ? sj € İ”x ¨¢
¨A¨qÑ1Ô1Ğİ œX e¨r¢k°1°aÑ 8Ô 8Ğİ”(˜5 Aš: q¨!Ñ,Ô,€Kİ
Œ8�UĞ ,Ğ.DÀkĞRĞYZĞ [Ñ [Ô [Ğ[r Ú
probabilitiesÚmaskc óŒ — d}| d|dz ¦ « | d|dz ¦ « z }| ¦ « }|dk r=|d|› �z
}t j |j dt j |j ¦ « z ¦ « }n||z }t
|¦ « dk r t |t dd¦ « ¬ ¦ « |S )
NÚ rG é r zISum of valid probabilities is 0. Creating a uniform probability...
Mask:
g ğ?zmasking_message.txtÚw)Úfile) r- Úsumr& Úfullr; ÚprodÚlenÚprintÚopen)rM rN r0 Úto_printÚvalidsÚ
valids_sums r Úmask_invalid_actionsr] F sÊ € Ø€Hğ ×
"Ò
" 2 z°Q¡Ñ
7Ô
7¸$¿,º,ÀrÈ:ĞYZÉ?Ñ:[Ô:[Ñ
[€FØ—’‘ ” €JØ�Q‚€ğ ĞhĞbfĞhĞhÑhˆİ”˜œ sR¬W°V´\Ñ-BÔ-BÑ'BÑCÔCˆˆà˜*Ñ$ˆå
ˆ8�}„}�qÒĞİ
ˆh�TĞ"7¸Ñ =Ô =Ğ>Ñ>Ô>Ğ>Ø €Mr Ústatesc óĞ — g }| D ]�}| ¦ « ¦ « ¦ « }t j |dk d|¦ « }t j |dk d|¦ « }t j |dk d|¦ « }| |¦ « Œ�t
j t j |¦ « t j ¬¦ «
d¦ « S )Nr éûÿÿÿr ©Údtype) ÚdetachÚcpuÚnumpyr& rH r) ÚthÚtensorr' Úfloat32Úsqueeze)r^ Úmasksr2 Únp_staterN s r Úmask_invalid_actions_batchrl Z s € Ø €EØğ ğ ˆØ—<’<‘>”>×%Ò%Ñ'Ô'×-Ò-Ñ/Ô/ˆİŒx˜ Aš
r¨8Ñ4Ô4ˆİŒx˜ š 1 dÑ+Ô+ˆİŒx˜ š
A tÑ,Ô,ˆØ
� Š �TÑÔĞĞå
Œ9•R”X˜e‘_”_B¬JĞ 7Ñ 7Ô 7× ?Ò ?ÀÑ BÔ BĞBr Úargsc óL — g d¢}|D ] }|| vrt d|› d|› d�¦ « ‚Œ d S )N)Únum_net_channelsÚnum_net_in_channelsÚ net_dropoutÚnet_action_sizeÚnum_simulationsÚself_play_gamesÚ num_itersÚepochsÚlrÚmax_buffer_sizeÚ
num_pit_gamesÚrandom_pit_freqr0 Ú
batch_sizeÚtauÚcÚcheckpoint_dirÚupdate_thresholdz Missing key z? in args dict. Please supply all required keys.
Required keys: ú.)ÚKeyError)rm Ú
required_keysÚkeys r Ú
check_argsr„ f st € ğvğ vğ v€Mğ ğ ?ğ ?ˆØ �dˆ?ˆ?İğ >¨#ğ >ğ >Ø-:ğ >ğ >ğ >ñ ?ô ?ğ
?ğ ğ?ğ ?r Únc ó� ‡— t | ‰¦ « | dz dz z dt ˆfd„t ‰| ¦ « D ¦ « ¦ « z z S )NrQ é c ó0 •— g | ]}t |‰¦ « ‘ŒS r
)Ú get_num_horizontal_conv_slides)r r r# s €r r z1calculate_board_win_positions.<locals>.<listcomp>s s$ ø€ ĞCĞCĞC°!Õ '¨¨1Ñ -Ô -ĞCĞCĞCr )r‰ rT r: )r… r# s `r Ú calculate_board_win_positionsrŠ q sa ø€ İ )¨!¨QÑ /Ô /°1°q±5¸1±9Ñ =ÀÅCØCĞCĞCĞCµu¸QÀ±{´{ĞCÑCÔCñEEô EEñ AEñ Eğ Er r0 Ú kernel_sizec ó — | |z
dz S )Nr r
)r0 r‹ s r r‰ r‰ v s € Ø ˜Ñ $¨Ñ )Ğ)r Ún_trialsÚ
init_net_pathÚstorageÚ
study_nameÚconfigc ó¦ ‡‡‡‡ ‡
— ˆˆ ˆˆˆ
fd„}ddl mŠ ddlmŠ |Š
d‰
_ t j ||¬¦ « }| || ¬¦ « dS ) aœ
Performs a hyperparameter search using optuna. This method is meant to be called using the start_jobs.py script.
For this method to work, a mysql database must be running on the storage address and an optuna study with the
given name and the 'maximize' direction must exist.
:param n_trials: num of trials to run the search for.
:param init_net_path: The path to the initial network to use for all trials.
:param storage: The mysql storage string. Specifies what database to use.
:param study_name: Name of the study to use.
:param game: The game instance to use.
:param config: The config to use for the search.
:return:
c óX •— | ddd¦ « }| ddd¦ « }| ddd ¦ « }| d
d d d
¬¦ « }| ddd¦ « }| dd d¦ « }| ddd¦ « }| dddd
¬¦ « }|‰_ |‰_ |‰_ |‰_ |‰_ |‰_ |‰_ d‰_ |‰_
‰ ‰ ¦ « ‰¦ « } ‰
‰‰‰| ¦ « }
t d| j › d�¦ « |
¦ « |
¦ « } | | | j ¦ « t d| j › d| › d�¦ « ~
| S ) NÚnum_mc_simulationsé< i@ Únum_self_play_gamesé2 éÈ Ú
num_epochséd i� rw g-C ëâ6?g{®Gáz„?T)ÚlogÚtempg à?g ø?Ú
arena_tempÚcpucté Ú log_epsilong»½×Ùß|Û=gH¯¼šò×z>zTrial z started.z finished with win freq r€ )Ú suggest_intÚ
suggest_floatrs rt rv rw r| r} Ú arena_tauru r Úmake_fresh_instanceÚfrom_state_dictrX ÚnumberÚtrainÚ get_arena_win_frequencies_meanÚreport)Útrialr” r– r™ rw rœ r� r� r Ú search_treeÚtrainerÚwin_freqÚ McSearchTreeÚTrainerÚgamer� Ú trial_configs €€€€€r Ú objectivez-az_optuna_parameter_search.<locals>.objective‰ sÁ ø€ Ø "× .Ò .Ğ/CÀRÈÑ NÔ NĞØ #× /Ò /Ğ0EÀrÈ3Ñ OÔ OĞØ×&Ò& |°S¸#Ñ>Ô>ˆ
Ø
×
Ò
t¨T°tĞ
Ñ
<Ô
<ˆØ×"Ò" 6¨3°Ñ4Ô4ˆØ×(Ò(¨ °t¸SÑAÔAˆ
Ø×#Ò# G¨S°!Ñ4Ô4ˆØ×)Ò)¨-¸ÀÈ$Ğ)ÑOÔOˆ à'9ˆ Ô$Ø':ˆ Ô$Ø (ˆ ÔØ ˆ ŒØ ˆ ÔØ ˆ ŒØ!+ˆ Ô Ø!"ˆ Ô Ø#.ˆ Ô Ø"�l 4×#;Ò#;Ñ#=Ô#=¸|ÑLÔLˆ Ø×)Ò)¨-¸ ÀtÈ[ÑYÔYˆİ
Ğ.�u”|Ğ.Ğ.Ğ.Ñ/Ô/Ğ/Ø�
Š
‰ŒˆØ×9Ò9Ñ;Ô;ˆØ
� Š �X˜uœ|Ñ,Ô,Ğ,İ
ĞH�u”|ĞHĞH¸XĞHĞHĞHÑIÔIĞIØ Øˆr r )r¯ )r® F)r� r� )r� N)Úmu_alpha_zero.trainerr¯ Ú+mu_alpha_zero.AlphaZero.MCTS.az_search_treer® Ú show_tqdmÚoptunaÚ
load_studyÚoptimize) r� r� r� r� r° r‘ r² Ústudyr® r¯ r± s ` ` @@@r Úaz_optuna_parameter_searchrº z s˜ øøøøø€ ğ ğ ğ ğ ğ ğ ğ ğ ğ ğ: .Ğ-Ğ-Ğ-Ğ-Ğ-ØHĞHĞHĞHĞHĞHØ€LØ "€LÔİ Ô ¨¸WĞ EÑ EÔ E€EØ ‡N‚N�9 x€NÑ0Ô0Ğ0Ğ0Ğ0r Ú
muzero_configc ó„ — t | j | j | j | j | j ¦ « }| |¦ « S r )r rp ro rq rr Úaz_net_linear_input_sizeÚto)r» ÚdeviceÚnetworks r Úbuild_net_from_configrÁ ® sB € ݘ=Ô<¸mÔ>\Ø(Ô4°mÔ6SØ(ÔAñCô C€Gğ �:Š:�fÑ Ô Ğ r Úmuzero_alphazero_configc óÖ — |€| j }|€| j }t | |¦ « }t j | ¦ « |¬¦ « }t |¬¦ « }|||fS )N)rw )Úmax_size)rw rx rÁ rf ÚoptimÚAdamÚ
parametersr )r r¿ rw Ú buffer_sizerÀ Ú optimizerÚmemorys r Úbuild_all_from_configrË µ sm € à €zØ
$Ô
'ˆØĞØ-Ô=ˆ İ#Ğ$;¸VÑDÔD€Gİ”—
’
˜g× 0Ò 0Ñ 2Ô 2°r�
Ñ:Ô:€Iİ
Ğ
,Ñ
,Ô
,€FØ �I˜vĞ %Ğ%r Úcheckpoint_pathc ó — |€t t ¦ « }t j t j ¦ « rdnd¦ « }t
||¦ « }t j | ¦ « }| |d ¦ « |S )NÚcudard Únet) r
Ú test_argsrf r¿ rÎ Ú is_availablerÁ ÚloadÚload_state_dict)rÌ rm r¿ rÏ Údatas r Úmake_net_from_checkpointrÕ Á su € Ø €|İ•yÑ!Ô!ˆİ
ŒY¥¤×!5Ò!5Ñ!7Ô!7ĞB�v�v¸UÑ
CÔ
C€Fİ
fÑ
-Ô
-€Cİ
Œ7�?Ñ #Ô #€DØ×Ò˜˜Uœ Ñ$Ô$Ğ$Ø €Jr c óŠ — t ¦ « j j } | dk rdS | dk rdS | dk rdS dS # t $ r Y dS w xY w)NÚZMQInteractiveShellTÚShellÚTerminalInteractiveShellF)r Ú __class__r Ú NameError)Úshells r Ú is_notebookrİ Ë so € ğ İ‘
”
Ô'Ô0ˆØ Ğ)Ò )Ğ )Ø�4Ø
�gÒ
Ğ
Ø�4Ø
Ğ0Ò
0Ğ
0Ø�5à�5øİ ğ ğ ğ ؈uˆuğøøøs ‚ 4 ¢4 ª4 ´
AÁAFÚfilesÚnot_notebook_okc ó( — t ¦ « }|s|rd S |st d¦ « ‚| D ]i}t j d¦ « s3t j d¦ « rt j d¦ « t
j |d¦ « Œjd S )Nz2This method should only be called from a notebook.z"/content/drive/MyDrive/Checkpointsz/content/drive/MyDrive)rİ Ú RuntimeErrorÚosÚpathÚexistsÚmkdirÚshutilr, )rŞ rß Úis_nbtrS s r Úupload_checkpoint_to_gdriverè Ú s¨ € İ
‰]Œ]€FØ ğ �oğ ØˆØ ğ QİĞOÑPÔPĞPàğ @ğ @ˆİŒw�~Š~Ğ BÑCÔCğ ;ÍÌÏÊĞWoÑHpÔHpğ ;İ ŒHĞ9Ñ :Ô :Ğ :İŒ �DĞ>Ñ?Ô?Ğ?Ğ?ğ@ğ @r Úoutput_file_nameÚ depth_limitc ó\ ‡‡— t ¦ « }d|j d< d|j d< d|j d< d|j d< ‰€t d¦ « Šd t j d
t fˆˆfd „ Š ‰| | |‰¬ ¦ « | d
¬¦ « | |› d�¦ « d S )NzMCTS visualizationÚlabelÚcircler; ÚblueÚcolorÚgoldÚinfÚgÚd_limitc
óú •— d }| j €:t t j dd|j j ¬¦ « ¦ « }nt | j ¦ « }| |¦ « || k r(| t |j ¦ « |¦ « | ¦ « r|dk rd S | j
¦ « D ])} ‰|| |‰t d¦ « k r|dz
n‰¬¦ « Œ*d S )Nr rŸ )ÚlowÚhighÚsizerñ r ©ró ) r2 Ústrr& ÚrandomÚrandintr; Úadd_nodeÚadd_edgeÚ was_visitedÚchildrenr( Úfloat)ÚnodeÚparentrò ró r6 Úchildrê Ú
make_graphs €€r r z"visualize_tree.<locals>.make_graphğ s ø€ àˆØ Œ:Ğ İ�œ×*Ò*¨q°q¸v¼|Ô?QĞ*ÑRÔRÑSÔSˆFˆF嘜‘_”_ˆFØ �
Š
�6ÑÔĞØ �TŠ>ˆ>Ø
�JŠJ•s˜6œ<Ñ(Ô(¨&Ñ 1Ô 1Ğ 1Ø× Ò Ñ!Ô!ğ W°¢\ \Ø ˆFğ ”]×)Ò)Ñ+Ô+ğ lğ lˆEØ ˆJ�u˜d A¸kÍUĞSXÉ\Ì\Ò>YĞ>Y¨w¸©{¨{Ğ_jĞ kÑ kÔ kĞ kĞ kğ lğ lr rø Údot)Úprogz.png) Ú
pygraphvizÚAGraphÚ
graph_attrÚ node_attrÚ edge_attrr ÚintÚlayoutÚdraw)Ú root_noderé rê Úgraphr s ` @r Úvisualize_treer ç så øø€ İ × Ò Ñ Ô €EØ 4€EÔ�WÑ Ø '€E„O�GÑ Ø %€E„O�GÑ Ø %€E„O�GÑ ØĞݘE‘l”lˆ ğl¥JÔ$5ğ lÅğ lğ lğ lğ lğ lğ lğ lğ* €Jˆy˜) U°KĞ@Ñ@Ô@Ğ@Ø ‡L‚L�e€LÑ Ô Ğ Ø ‡J‚JĞ"Ğ(Ğ(Ğ(Ñ)Ô)Ğ)Ğ)Ğ)r rÔ rÊ c óú — | D ]w}|D ]r\ }}}t j |t j ¬¦ « ||¦ « }t j |t j ¬¦ « }| |||f¦ « ŒsŒxd S )Nra )rf rg rh r- Úadd)rÔ rÊ r0 Ú game_datar2 r3 r4 s r Úcpp_data_to_memoryr
s� € ğ ğ 'ğ 'ˆ Ø %ğ 'ğ '‰LˆE�2�qİ”I˜e2¬:Ğ6Ñ6Ô6×>Ò>¸zÈ:ÑVÔVˆEİ”˜2¥R¤ZĞ0Ñ0Ô0ˆBØ �JŠJ˜˜r 1�~Ñ &Ô &Ğ &Ğ &ğ 'ğ'ğ 'r )NN)Fr )0râ ræ re r& r¶ Útorchrf ÚIPythonr Ú$mu_alpha_zero.AlphaZero.Network.nnetr Ú!mu_alpha_zero.AlphaZero.constantsr rĞ Úmu_alpha_zero.mem_bufferr Úmu_alpha_zero.configr r Údictr
Úlistr8 Úndarrayr r= r? rE rA r] rg rl r„ rŠ r‰ rù rº rÁ ÚtuplerÅ Ú OptimizerrË rÕ rİ Úboolrè r r r
r r ú<module>r" s� ğØ € € € Ø
€
€
€
Ø Ğ Ğ Ğ Ø
€
€
€
à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ ğ >Ğ =Ğ =Ğ =Ğ =Ğ =Ø IĞ IĞ IĞ IĞ IĞ IØ .Ğ .Ğ .Ğ .Ğ .Ğ .Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8ğğ ğ ğ ğ ˆdñ ô ğ ğ ¸ğ ÈTğ ğ ğ ğ ğ"˜œ
ğ sğ ğ ğ ğ ğ �b”jğ ¨ğ ğ ğ ğ ğ 4ğ ğ ğ ğ ğ\ R¤Zğ \ğ \ğ \ğ \ğ¨¬
ğ ¸"¼*ğ ĞUWÔU_ğ ğ ğ ğ ğ( C r¤yğ C°R´Yğ Cğ Cğ Cğ Cğ?�Tğ ?ğ ?ğ ?ğ ?ğE Sğ E¨Sğ Eğ Eğ Eğ Eğ
*¨sğ *Àğ *Èğ *ğ *ğ *ğ *ğ11¨ğ 11¸Sğ 11È3ğ 11Ğ\_ğ 11Ğo~ğ 11ğ 11ğ 11ğ 11ğh ¨ğ ¸Lğ ğ ğ ğ ğ &ğ &°6ğ &ĞafØ�"”(Ô$ iĞ/ôb1ğ &ğ &ğ &ğ &ğ¨cğ ¸À4¹ ğ ğ ğ ğ ğ ğ ğ ğ
@ğ
@ tğ
@¸dğ
@ğ
@ğ
@ğ
@ğ *ğ *°ğ *À#ÈÁ*ğ *ğ *ğ *ğ *ğF'˜Tğ '¨9ğ 'À#ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'r | 17,643 | Python | .pyt | 104 | 168.125 | 1,037 | 0.351559 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,562 | alpha_zero.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/alpha_zero.cpython-311.pyc | §
weŠf¤ ã óÀ — d dl Z d dlmZ d dlZd dlZd dlmZ d dl m
Z
d dl m Z d dl
mZ d dlmZ d dlmZ d d lmZ d d
lmZmZ d d lmZ d d lmZ G d
„ d¦ « Z dS )é N)ÚType)ÚArena)Ú NetPlayer)Ú McSearchTree)ÚTrainer)Ú
AlphaZeroGame)ÚGeneralMemoryBuffer)ÚGeneralNetwork)Ú net_not_noneÚfind_project_root)Ú HookManager)ÚAlphaZeroConfigc
óÒ — e Zd Zdefd„Z d dedee ded e d
e
pdd e f d „Z d dee d
e de d e d
e
pdd e f d„Z
d„ Zd dej dedefd„Z d de de dededede f d „ZdS ) Ú AlphaZeroÚ
game_instancec ó¾ — d | _ d | _ || _ t j t j ¦ « rdnd¦ « | _ d | _ d | _ d S )NÚcudaÚcpu) ÚtrainerÚnetÚgameÚthÚdevicer Ú is_availableÚalpha_zero_configÚtree)Úselfr s úQ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/alpha_zero.pyÚ__init__zAlphaZero.__init__ sR € ؈Œ ؈ŒØ!ˆŒ Ý”i"¬'×*>Ò*>Ñ*@Ô*@Ð K ÀeÑLÔLˆŒ Ø26ˆÔ Ø"&ˆŒ ˆ ˆ ó TNFr Ú
network_classÚmemoryÚheadlessÚ hook_managerÚcheckpointer_verbosec ó¦ — | ||¬¦ « | j ¦ « }t | j ¦ « |¦ « }|| _ t | j ¦ « fi ||dœ¤Ž} || _ t j
|| j ||| ||||¬¦ « | _ | j ¦ « | _
d S )N)r$ )ÚnetworkÚmonte_carlo_tree_search)r# r% Úmemory_overrider$ )Úmake_from_configÚtor r r Úmake_fresh_instancer r r r Úcreater Ú get_networkr )
r r r! r" r# r$ r% r' r Ú
net_players
r Ú
create_newzAlphaZero.create_new sÙ € ð ×0Ò0Ð1BÐQ]Ð0Ñ^Ô^×aÒaÐbfÔbmÑnÔnˆÝ˜DœI× 9Ò 9Ñ ;Ô ;Ð=NÑOÔOˆØˆŒ Ý ˜tœy× <Ò <Ñ >Ô >ÐxÐxÈgÐrvÐBwÐBwÐxÐxˆ
Ø!2ˆÔ Ý ”~Ð&7¸¼ÀGÈTÐS]ÐhpØ;OÐagØ3?ðAñ Aô AˆŒ ð ”<×+Ò+Ñ-Ô-ˆŒˆˆr ÚpathÚcheckpoint_dirc ó — t j |t t ||| j |||¬¦ « | _ | j ¦ « | _ | j ¦ « | _ | j
¦ « | _ d S )N)r# r$ r% ) r Úfrom_checkpointr r r r r. r Úget_treer Úget_argsÚargs)r r! r1 r2 r# r$ r% s r Úload_checkpointzAlphaZero.load_checkpoint) s € õ Ô.¨}½lÍIÐW[Ð]kÐmqÔmvØ8@È|ØDXðZñ Zô ZˆŒ ð ”<×+Ò+Ñ-Ô-ˆŒØ”L×)Ò)Ñ+Ô+ˆŒ Ø”L×)Ò)Ñ+Ô+ˆŒ ˆ ˆ r c ó` — t | j ¦ « | j ¦ « d S )N)r r r Útrain)r s r r: zAlphaZero.train3 s, € Ý�T”XÑ Ô Ð Ø Œ ×ÒÑ Ô Ð Ð Ð r r ÚxÚtauÚreturnc ój — t | j ¦ « |j | j d | j d | j d fk s J dt |j ¦ « z ¦ « ‚| j | j |d| j |¬¦ « \ }}| j || j
j ¬¦ « S )NÚ
board_sizeÚnum_net_in_channelszXInput shape is not correct. Expected (board_size, board_size, num_net_in_channels).Got: é )r< ) r r Úshaper7 Ústrr Úsearchr r Ú select_mover r< )r r; r< ÚpiÚ_s r ÚpredictzAlphaZero.predict7 s¹ € Ý�T”XÑ Ô Ð ØŒw˜4œ9 \Ô2°D´I¸lÔ4KÈTÌYØ !ôN#ð $ò $ð $ð $ð&-Ý/2°1´7©|¬|ñ&<ñ$ô $ð $ð ” × Ò ¤¨1¨a°´À#Ð ÑFÔF‰ˆˆAØŒy×$Ò$ R¨TÔ-CÔ-GÐ$ÑHÔHÐHr rA Úp1_nameÚp2_nameÚ num_gamesÚstartsÚswitch_playersc ó0 — t | j ¦ « | j | j ¦ « | j ¦ « | j ¦ « }t ||¦ « }| j ||j |j ddœ} t ¦ « dd¦ « d¦ « d }
t j |
› d� j | |fi | ¤Ž} t j |
› d� j | |fi | ¤Ž} | j ¦ « }
|
d¦ « t# |
|| j ¦ « }| | | ||j | || ¬¦ « \ }}}t) d|› d |› d
|› �¦ « d S ) Néÿÿÿÿ)r' r( Ú evaluate_fnÚdepthÚplayerú\ú/z.AlphaZero.Arena.playersF)Ú
one_playerÚ start_playerÚ
add_to_kwargszResults: Player 1 wins: z, Player 2 wins: z , Draws: )r r r+ r Úevalr r, r Ú
eval_boardÚ
minimax_depthr ÚreplaceÚsplitÚsysÚmodulesÚ__dict__Ú set_headlessr ÚpitÚnum_simulationsÚprint)r rI rJ rK r rL rM Úmanagerr ÚkwargsÚ path_prefixÚp1Úp2Ú
arena_managerÚarenaÚp1_wÚp2_wÚdss r ÚplayzAlphaZero.play? s¨ € å�T”XÑ Ô Ð Ø Œ� Š �D”KÑ Ô Ð Ø Œ�
Š
‰ŒˆØ”)×/Ò/Ñ1Ô1ˆÝ˜GÐ%6Ñ7Ô7ˆØ !œXÀ$ÐW^ÔWiØ,Ô:ÀbðJð Jˆå'Ñ)Ô)×1Ò1°$¸Ñ<Ô<×BÒBÀ3ÑGÔGÈÔKˆ Ý
Œ[˜KÐAÐAÐAÔ
BÔ
KÈGÔ
TÐU\Ð
gÐ
gÐ`fÐ
gÐ
gˆÝ
Œ[˜KÐAÐAÐAÔ
BÔ
KÈGÔ
TÐU\Ð
gÐ
gÐ`fÐ
gÐ
gˆØ œ ×5Ò5Ñ7Ô7ˆ
Ø×"Ò" 5Ñ)Ô)Ð)Ý�mÐ%6¸¼ ÑDÔDˆØ Ÿš 2 r¨9Ð6GÔ6WÐhvÐdvØ06Àfð #ñ Nô N‰ˆˆd�Bå
ÐS¨ÐSÐSÀÐSÐSÈrÐSÐSÑTÔTÐTÐTÐTr )TNF)r )rA T)Ú__name__Ú
__module__Ú __qualname__r r r r r
r Úboolr
r0 rC r8 r: ÚnpÚndarrayÚfloatÚintrH rn © r r r r s‹ € € € € € ð' mð 'ð 'ð 'ð 'ð lpØ05ð .ð .¨Oð .ÈDÐQ_ÔL`ð .Ø.ð .Ø:>ð .ØU`ÐUhÐdhð .à)-ð .ð .ð .ð .ð TXØ5:ð,ð ,¨T°.Ô-Að ,Èð ,Ð^að ,Ø"&ð,Ø=HÐ=PÈDð,à.2ð,ð ,ð ,ð ,ð ð ð ðIð I˜œð I¨%ð I¸ð Ið Ið Ið Ið rsØ$(ðUð U˜Cð U¨#ð U¸#ð UÐRað UÐknð UØ !ðUð Uð Uð Uð Uð Ur r ) r] Útypingr Únumpyrs Útorchr Ú#mu_alpha_zero.AlphaZero.Arena.arenar Ú%mu_alpha_zero.AlphaZero.Arena.playersr Ú+mu_alpha_zero.AlphaZero.MCTS.az_search_treer Úmu_alpha_zero.trainerr Ú mu_alpha_zero.General.az_gamer Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr
Úmu_alpha_zero.General.utilsr r Ú mu_alpha_zero.Hooks.hook_managerr
Úmu_alpha_zero.configr r rw r r ú<module>r… s. ðØ
€
€
€
Ø Ð Ð Ð Ð Ð à Ð Ð Ð Ø Ð Ð Ð à 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø ;Ð ;Ð ;Ð ;Ð ;Ð ;Ø DÐ DÐ DÐ DÐ DÐ DØ )Ð )Ð )Ð )Ð )Ð )Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7Ø <Ð <Ð <Ð <Ð <Ð <Ø 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø GÐ GÐ GÐ GÐ GÐ GÐ GÐ GØ 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø 0Ð 0Ð 0Ð 0Ð 0Ð 0ð=Uð =Uð =Uð =Uð =Uñ =Uô =Uð =Uð =Uð =Ur | 7,308 | Python | .pyt | 31 | 234.677419 | 1,057 | 0.377851 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,563 | constants.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/constants.cpython-311.pyc | §
føf!
ã
ó — d Z i dd“dd“dd“de dz “d d
“d d “d
d“dd“dd“dd“dd“dd“de “dd“d d“d d “d d“d d!d"d#d$d$d%d$d&d'd d(œ ¥Zi dd“dd“dd“de dz “d d
“d d“d
d“dd)“dd“dd“dd“dd“de “dd*“d d+“d d,“d d-“d d!d"d#d$d"d.d'd d/œ ¥Zi dd“d0d“dd“dd“dd1“d2d3“d d4“d d$“d5d$“d6d7“d8d9“d:d"“d;d<“d
d“dd=“dd>“dd?“i dd“dd“de “dd@“d d“d dA“d d“dBdC“dDdE“dFd “dGd!“dHd “dIdJ“dKd$“dLd$“dMd%“dNd$“¥ddOdPdJdQdRd'd dSœ¥Zd S )Té Únum_net_channelsi Únum_net_in_channelsé Ú net_dropoutg333333Ó?Únet_action_sizeé Únum_simulationsi% Úself_play_gamesi, Ú num_itersé2 Úepochsiô Úlrg{äL?´œj?Úmax_buffer_sizei † Ú
num_pit_gamesé( Úrandom_pit_freqé Ú
board_sizeÚ
batch_sizeé ÚtauÚ arena_taugëö‹ÓG1¥?ÚcNg333333ã?é Té g•Ö&è .>i H ÚLogs) Úcheckpoint_dirÚupdate_thresholdÚ
minimax_depthÚ show_tqdmÚ num_workersÚ
num_to_winÚ log_epsilonÚzero_tau_afterÚaz_net_linear_input_sizeÚlog_dirÚpushbullet_tokeni@ é€ g ğ?é gffffffò?g»"¸hkN>) r r r r r! r" r# r& r' Únum_net_out_channelsé Únet_latent_sizeé$ éğ ÚKÚgammag�•C‹lçï?Úframe_buffer_sizeé Ú
frame_skipÚ num_stepsi� éd gü©ñÒMbP?ip éÿ g{®Gáz„?Úc2iÄL Úalphagš™™™™™é?r r r r Fr! r" r# r$ zALE/Asteroids-v5z Pickles/Data)é` r9 i )ÚbetaÚenv_idÚ
pickle_dirÚuse_javaÚtarget_resolutionr% r& r' )r ÚSAMPLE_AZ_ARGSÚTRAINED_AZ_NET_ARGSÚSAMPLE_MZ_ARGS© ó úP/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/constants.pyú<module>rE sî ğğ €
ğ ؘğ à˜1ğ ğ �3ğ ğ �z Q‘ğ ğ
�tğ ğ �sğ
ğ �ğ ğ
ˆcğ ğ Ğ
ğ ğ �wğ ğ �Rğ ğ �qğ ğ �*ğ ğ �#ğ ğ
ˆ1ğ ğ Ğ$ğ! ğ" ˆğ# ğ$ ØØØØØØØØ %ØØ ğ9 ğ ğ €ğ>ؘğà˜1ğğ �3ğğ �z Q‘ğ ğ
�tğ ğ �qğ
ğ �ğğ
ˆcğğ Ğ
ğğ �wğğ �Rğğ �qğğ �*ğğ �#ğ ğ
ˆ3ğ ğ �ğ!ğ" ˆğ#ğ$ ØØØØØØ)ØØ ğ5ğ ğ Ğ ğ:+ؘğ+à˜Cğ+ğ ˜1ğ+ğ �3ğ +ğ
�rğ +ğ �rğ
+ğ �sğ+ğ �qğ+ğ ˆğ+ğ ˆUğ+ğ ˜ğ+ğ �!ğ+ğ �ğ+ğ �ğ +ğ
ˆcğ +ğ ˆ%ğ!+ğ" �vğ#+ğ +ğ$ �Rğ%+ğ& �qğ'+ğ( �*ğ)+ğ* �#ğ++ğ,
ˆ1ğ-+ğ. �ğ/+ğ0 ˆğ1+ğ2 ˆ%ğ3+ğ4 ˆSğ5+ğ6 �dğ7+ğ8 ˜ğ9+ğ: �Tğ;+ğ< �ğ=+ğ> �1ğ?+ğ@ �!ğA+ğB �4ğC+ğD �ağE+ğ +ğF
Ø Ø ØØ!Ø $ØØ ğU+ğ +ğ +€€€rC | 2,697 | Python | .pyt | 19 | 140.789474 | 589 | 0.454274 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,564 | logger.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/__pycache__/logger.cpython-311.pyc | §
føfî ã ór — d dl Z d dlZd dlZd dlZd dlmZ d dlmZ G d„ d¦ « Z G d„ d¦ « Z dS )é N)ÚAPI©Úfind_project_rootc óŠ — e Zd Zddepddepdddfd„Zddededdfd „Zdepdfd
„Zdepdddfd „Zdded
eddfd„Zd„ Z dd„Z
dS )ÚLoggerNÚlogdirÚtokenÚreturnc ó — | |¦ « | _ t j | j d¬¦ « t j d¦ « | _ | j t j ¦ « t j | j › dt j
¦ « d¦ « › d�¦ « | _
| j
t j ¦ « | j | j
¦ « d| _ t! ¦ « | _ | |¦ « t j d¦ « }| j
|¦ « t+ j | j ¦ « d S ) NT)Úexist_okÚAlphaZeroLoggerú/z%Y-%m-%d_%H-%M-%Sz.logFz)[%(asctime)s - %(levelname)s] %(message)s)Ú init_logdirr ÚosÚmakedirsÚloggingÚ getLoggerÚloggerÚsetLevelÚDEBUGÚ FileHandlerÚdatetimeÚnowÚstrftimeÚ file_handlerÚ
addHandlerÚ is_token_setr ÚapiÚinit_api_tokenÚ FormatterÚ setFormatterÚatexitÚregisterÚcleanup)Úselfr r Ú formatters úM/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/logger.pyÚ__init__zLogger.__init__ s= € Ø×&Ò& vÑ.Ô.ˆŒ İ
Œ �D”K¨$Ğ/Ñ/Ô/Ğ/İ Ô'Ğ(9Ñ:Ô:ˆŒ Ø Œ × Ò �Wœ]Ñ+Ô+Ğ+İ #Ô /ØŒ{Ğ XĞ X�XÔ .× 2Ò 2Ñ 4Ô 4× =Ò =Ğ>QÑ RÔ RĞ XĞ XĞ Xñ Zô ZˆÔà Ô×"Ò"¥7¤=Ñ1Ô1Ğ1Ø Œ × Ò ˜tÔ 0Ñ1Ô1Ğ1Ø !ˆÔİ‘5”5ˆŒØ ×Ò˜EÑ"Ô"Ğ"İÔ%Ğ&QÑRÔRˆ Ø Ô×&Ò& yÑ1Ô1Ğ1İŒ˜œ Ñ%Ô%Ğ%Ğ%Ğ%ó ÚdebugÚmsgÚlevelc óB — t | j |¦ « |¦ « d S ©N)Úgetattrr )r% r+ r, s r' Úlogz
Logger.log s$ € Ø#��” ˜UÑ#Ô# CÑ(Ô(Ğ(Ğ(Ğ(r) c ó, — |€t ¦ « › d�S |S )Nz/Logs/ProgramLogsr )r% r s r' r zLogger.init_logdir s! € Ø ˆ>İ'Ñ)Ô)Ğ<Ğ<Ğ<Ğ <àˆMr) c óP — |€d S | j |¦ « d| _ d S )NT)r Ú set_tokenr )r% r s r' r zLogger.init_api_token% s0 € Ø ˆ=Ø ˆFØ Œ×Ò˜5Ñ!Ô!Ğ!Ø ˆÔĞĞr) ÚMuZeroÚ algorithmc ó¤ — | j sd S | j |› d�|¦ « d S # t $ r}t |¦ « Y d }~d S d }~ww xY w)Nz training notification.)r r Ú send_noteÚ ExceptionÚprint)r% r+ r5 Úes r' Úpushbullet_logzLogger.pushbullet_log+ st € ØÔ ğ Ø ˆFğ Ø ŒH× Ò )Ğ DĞ DĞ DÀcÑ JÔ JĞ JĞ JĞ Jøİğ ğ ğ İ �!‰HŒHˆHˆHˆHˆHˆHˆHˆHøøøøğ øøøs ‹ + «
AµA
Á
Ac óz — t j | j ¦ « D ] }t j | j › d|› �¦ « Œ!d S )Nr )r Úlistdirr Úremove)r% Ú file_names r' Ú clear_logdirzLogger.clear_logdir3 sJ € İœ D¤KÑ0Ô0ğ 4ğ 4ˆIİ ŒI˜œĞ2Ğ2 yĞ2Ğ2Ñ 3Ô 3Ğ 3Ğ 3ğ 4ğ 4r) c óv — | j ¦ « | j | j ¦ « d S r. )r Úcloser Ú
removeHandler)r% s r' r$ zLogger.cleanup7 s6 € Ø Ô× Ò Ñ!Ô!Ğ!Ø Œ ×!Ò! $Ô"3Ñ4Ô4Ğ4Ğ4Ğ4r) r. )r* )r4 )r
N) Ú__name__Ú
__module__Ú __qualname__Ústrr( r0 r r r; r@ r$ © r) r' r r s € € € € € ğ&ğ &˜s˜{ dğ &°3°;¸$ğ &È$ğ &ğ &ğ &ğ &ğ )ğ )�sğ ) 3ğ )°Tğ )ğ )ğ )ğ )ğ ¨ğ ğ ğ ğ ğ ! C K¨4ğ !°Dğ !ğ !ğ !ğ !ğ ğ #ğ °#ğ ÀTğ ğ ğ ğ ğ4ğ 4ğ 4ğ5ğ 5ğ 5ğ 5ğ 5ğ 5r) r c
óÈ — e Zd Zedededefd„¦ « Zededededededef d „¦ « Zedefd
„¦ « Zedededed e fd „¦ « Z
ed
efd„¦ « Z ede fd„¦ « Z
ede de fd„¦ « Zede de fd„¦ « Zedefd„¦ « Zedefd„¦ « Zededefd„¦ « Zededefd „¦ « Zed efd „¦ « Zed „ ¦ « Zd S )!ÚLoggingMessageTemplatesÚname1Úname2Ú num_gamesc ó — d| › d|› d|› d�S )NzStarting pitting between ú and z for ú games.rH )rK rL rM s r' Ú
PITTING_STARTz%LoggingMessageTemplates.PITTING_START> s# € àU¨5ĞUĞU°uĞUĞUÀ9ĞUĞUĞUĞUr) Úwins1Úwins2ÚtotalÚdrawsc ó2 — d| › d|› d||z › d||z › d|› d� S )NzPitting ended between rO z. Player 1 win frequency: ú. Player 2 win frequency: ú . Draws: ú.rH )rK rL rR rS rT rU s r' Ú PITTING_ENDz#LoggingMessageTemplates.PITTING_ENDB se € ğL¨ğ Lğ L°Uğ Lğ LØ+0°5©=ğLğ Là+0°5©=ğLğ LàCHğLğ Lğ Lğ Mr) c ó — d| › d�S )NzStarting self play for rP rH )rM s r' ÚSELF_PLAY_STARTz'LoggingMessageTemplates.SELF_PLAY_STARTH s € à;¨Ğ;Ğ;Ğ;Ğ;r) Ú not_zero_fnc ór — | �|�|€dS d| || |z |z ¦ « z › d| || |z |z ¦ « z › d|› d�S )NzTSelf play ended. Results not available (This is expected if you are running MuZero).z)Self play ended. Player 1 win frequency: rW rX rY rH )rR rS rU r] s r' Ú
SELF_PLAY_ENDz%LoggingMessageTemplates.SELF_PLAY_ENDL sˆ € à ˆ=˜E˜M¨U¨]ØiĞiğk¸EÀ[À[ĞQVĞY^ÑQ^ĞafÑQfÑEgÔEgÑ<hğ kğ kØ+0°K°KÀÈÁ
ĞPUÑ@UÑ4VÔ4VÑ+Wğkğ kØbgğkğ kğ kğ lr) Ú num_epchsc ó — d| › d�S )Nz Starting network training for z epochs.rH )r` s r' ÚNETWORK_TRAINING_STARTz.LoggingMessageTemplates.NETWORK_TRAINING_STARTS s € àC° ĞCĞCĞCĞCr) Ú mean_lossc ó — d| › �S )Nz#Network training ended. Mean loss: rH )rc s r' ÚNETWORK_TRAINING_ENDz,LoggingMessageTemplates.NETWORK_TRAINING_ENDW s € à@°YĞ@Ğ@Ğ@r) Únum_winsÚupdate_thresholdc ó — d| › d|› d�S )Nz:!!! Model rejected, restoring previous version. Win rate: ú. Update threshold: ú !!!rH ©rf rg s r' Ú MODEL_REJECTz$LoggingMessageTemplates.MODEL_REJECT[ s* € ğ<ÈXğ <ğ <Ø%5ğ<ğ <ğ <ğ =r) c ó — d| › d|› d�S )Nz7!!! Model accepted, keeping current version. Win rate: ri rj rH rk s r' Ú MODEL_ACCEPTz$LoggingMessageTemplates.MODEL_ACCEPT` s* € ğ
Àhğ
ğ
Ğdtğ
ğ
ğ
ğ r) Ú num_itersc ó — d| › d�S )NzStarting training for z iterations.rH )ro s r' ÚTRAINING_STARTz&LoggingMessageTemplates.TRAINING_STARTf s € à?¨ Ğ?Ğ?Ğ?Ğ?r) Ú args_usedc ój — d}| ¦ « D ]\ }}||› d|› d�z
}Œd|d d… › �S )NÚ z: z, zTraining ended. Args used: éşÿÿÿ)Úitems)rr Ú
args_used_strÚkeyÚvalues r' Ú TRAINING_ENDz$LoggingMessageTemplates.TRAINING_ENDj sX € àˆ
Ø#Ÿ/š/Ñ+Ô+ğ 1ğ 1‰JˆC�Ø Ğ 0Ğ 0 uĞ 0Ğ 0Ğ 0Ñ 0ˆMˆMØA¨]¸3¸B¸3Ô-?ĞAĞAĞAr) Útype_Úpathc ó — d| › d|› �S )NzSaved z to rH ©r{ r| s r' ÚSAVEDz LoggingMessageTemplates.SAVEDq s € à)˜Ğ)Ğ) 4Ğ)Ğ)Ğ)r) c ó — d| › d|› �S )Nz Restored z from rH r~ s r' ÚLOADEDz LoggingMessageTemplates.LOADEDu s € à.˜5Ğ.Ğ.¨Ğ.Ğ.Ğ.r) Úiterc ó — d| › d�S )Nz
Iteration z$ of the algorithm training finished!rH )r‚ s r' ÚITER_FINISHED_PSBz)LoggingMessageTemplates.ITER_FINISHED_PSBy s € àF˜DĞFĞFĞFĞFr) c ó — dS )Nz;Algorithm Training finished, you can collect the results :)rH rH r) r' ÚTRAINING_END_PSBz(LoggingMessageTemplates.TRAINING_END_PSB} s € àLĞLr) N)rD rE rF Ú staticmethodrG ÚintrQ rZ r\ Úcallabler_ rb Úfloatre rl rn rq Údictrz r r� r„ r† rH r) r' rJ rJ < s € € € € € àğV˜Sğ V¨ğ V¸ğ Vğ Vğ Vñ „\ğVğ ğM˜3ğ M sğ M°3ğ M¸sğ MÈ3ğ MĞWZğ Mğ Mğ Mñ „\ğMğ
ğ< 3ğ <ğ <ğ <ñ „\ğ<ğ ğl˜Sğ l¨ğ l°Sğ lÀxğ lğ lğ lñ „\ğlğ ğD¨#ğ Dğ Dğ Dñ „\ğDğ ğA¨ğ Ağ Ağ Añ „\ğAğ ğ=˜uğ =¸ğ =ğ =ğ =ñ „\ğ=ğ ğ˜uğ ¸ğ ğ ğ ñ „\ğğ
ğ@ #ğ @ğ @ğ @ñ „\ğ@ğ ğB ğ Bğ Bğ Bñ „\ğBğ ğ*�Sğ * ğ *ğ *ğ *ñ „\ğ*ğ ğ/�cğ / ğ /ğ /ğ /ñ „\ğ/ğ ğG ğ Gğ Gğ Gñ „\ğGğ ğMğ Mñ „\ğMğ Mğ Mr) rJ )
r" r r r Ú
pushbulletr Úmu_alpha_zero.General.utilsr r rJ rH r) r' ú<module>r� s¼ ğØ
€
€
€
Ø €€€Ø €€€Ø € € € à Ğ Ğ Ğ Ğ Ğ à 9Ğ 9Ğ 9Ğ 9Ğ 9Ğ 9ğ.5ğ .5ğ .5ğ .5ğ .5ñ .5ô .5ğ .5ğbCMğ CMğ CMğ CMğ CMñ CMô CMğ CMğ CMğ CMr) | 9,693 | Python | .pyt | 20 | 483.6 | 3,727 | 0.398284 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,565 | players.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Arena/__pycache__/players.cpython-311.pyc | §
�Œf$ ã ó — d dl Z d dlZd dlmZmZ d dlZd dlZd dlm Z d dl
m Z d dl m
Z
G d„ de¦ « Z G d„ d e¦ « Z G d
„ d e¦ « Z G d „ d
e¦ « Z G d„ de¦ « Z G d„ de¦ « Z G d„ de¦ « ZdS )é N)ÚABCÚabstractmethod)Ú AlphaZeroNet)ÚTicTacToeGameManager)Ú
AlphaZeroGamec óš — e Zd ZdZed„ ¦ « Zedej dee e f fd„¦ « Z
ed„ ¦ « Z de fd„Z
ed „ ¦ « Zd
S ) ÚPlayerz�
To create a custom player, extend this class and implement the choose_move method.
You can see different implementations below.
c ó — d S ©N© ©ÚselfÚ game_managerÚkwargss úT/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Arena/players.pyÚ__init__zPlayer.__init__ ó € à ˆó ÚboardÚreturnc ó — d S r r )r r r s r Ú choose_movezPlayer.choose_move r r c ó — d S r r ©r s r Úmake_fresh_instancezPlayer.make_fresh_instance r r r c ób — | ¦ « D ]}t | ||| ¦ « Œd S r )ÚkeysÚsetattr)r r Úkeys r Ú init_kwargszPlayer.init_kwargs s< € Ø—;’;‘=”=ğ ,ğ ,ˆCİ �D˜#˜v cœ{Ñ +Ô +Ğ +Ğ +ğ ,ğ ,r c ó — d S r r ©r r s r Úset_game_managerzPlayer.set_game_manager# r r N)Ú__name__Ú
__module__Ú __qualname__Ú__doc__r r ÚnpÚndarrayÚtupleÚintr r Údictr r# r r r r r
s € € € € € ğğ ğ
ğ
ğ
ñ „^ğ
ğ ğ
¤ğ
¸%ÀÀSÀ¼/ğ
ğ
ğ
ñ „^ğ
ğ ğ
ğ
ñ „^ğ
ğ, $ğ ,ğ ,ğ ,ğ ,ğ ğ
ğ
ñ „^ğ
ğ
ğ
r r c óP — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z dS ) Ú RandomPlayerr c ón — || _ | j j | _ || _ | |¦ « d S r ©r Ú __class__r$ Únamer r r
s r r zRandomPlayer.__init__) ó7 € Ø (ˆÔØ”NÔ+ˆŒ Ø ˆŒ Ø ×Ò˜Ñ Ô Ğ Ğ Ğ r r r c óª — | j j |fi |¤�}d| ¦ « v r |d }nd}|rt |¦ « nt |¦ « S )NÚunravelT)r Úget_random_valid_actionr r* r+ )r r r Úmover5 s r r zRandomPlayer.choose_move/ s^ € Ø8ˆtÔ Ô8¸ĞIĞIÀ&ĞIĞIˆØ ˜Ÿ š ™
œ
Ğ %Ğ %Ø ˜YÔ'ˆGˆGàˆGØ %Ğ4�u�T‰{Œ{ˆ{3¨t©9¬9Ğ4r c óT — t | j ¦ « fi | j ¤�S r )r. r r r r s r r z RandomPlayer.make_fresh_instance7 s) € ݘDÔ -× AÒ AÑ CÔ CĞSĞSÀtÄ{ĞSĞSĞSr c ó — || _ d S r ©r r" s r r# z RandomPlayer.set_game_manager: ó € Ø (ˆÔĞĞr N© r$ r% r& r r r( r) r* r+ r r r# r r r r. r. ( s} € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ 5 ¤ğ 5¸%ÀÀSÀ¼/ğ 5ğ 5ğ 5ğ 5ğTğ Tğ Tğ)ğ )ğ )ğ )ğ )r r. c óV — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z d„ Z d S )
Ú NetPlayerr c ón — || _ | j j | _ || _ | |¦ « d S r r0 r
s r r zNetPlayer.__init__? r3 r r r c ó — |d }|d }|d }n # t $ r t d¦ « ‚w xY w| j | j ||||¬¦ « \ }}| j ||¬¦ « }| j d ¦ « d| ¦ « v r |d } nd} | r| j |¦ « S t |¦ « S )NÚcurrent_playerÚdeviceÚtauúKMissing keyword argument. Please supply kwargs: current_player, device, tau©rC r5 T)
ÚKeyErrorÚmonte_carlo_tree_searchÚsearchÚnetworkr Ú select_moveÚ step_rootr Únetwork_to_boardr+ )
r r r rA rB rC ÚpiÚ_r7 r5 s
r r zNetPlayer.choose_moveE s € ğ "Ø #Ğ$4Ô 5ˆNؘHÔ%ˆFؘ”-ˆCˆCøİğ "ğ "ğ "İğ !ñ "ô "ğ
"ğ "øøøğ Ô,×3Ò3°D´LÀ%È ĞY_ĞehĞ3ÑiÔi‰ˆˆAØÔ ×,Ò,¨R°CĞ,Ñ8Ô8ˆØ Ô$×.Ò.¨tÑ4Ô4Ğ4Ø ˜Ÿ š ™
œ
Ğ %Ğ %Ø ˜YÔ'ˆGˆGàˆGØ ğ <ØÔ$×5Ò5°dÑ;Ô;Ğ ;İ�4‰yŒyĞó ‚ ›5c ó¬ — t | j ¦ « fi t j | j ¦ « | j ¦ « dœ¤�S )N)rI rG )r> r r ÚcopyÚdeepcopyrI rG r s r r z NetPlayer.make_fresh_instanceY sv € ݘÔ*×>Ò>Ñ@Ô@ğ Uğ UÕPTÔP]Ğ^bÔ^jÑPkÔPkØ`dÔ`|÷ aQò aQñ aSô aSğETğ ETğ Uğ Uğ Ur c ó — || _ d S r )rI )r rI s r Ú set_networkzNetPlayer.set_network] s
€ Ø ˆŒ ˆ ˆ r c ó — || _ d S r r: r" s r r# zNetPlayer.set_game_manager` r; r N)
r$ r% r& r r r( r) r* r+ r r rT r# r r r r> r> > sŒ € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ(Uğ Uğ Uğ ğ ğ ğ)ğ )ğ )ğ )ğ )r r> c ód — e Zd Zdededefd„Zdefd„Zdej de
e e f fd„Z d „ Z
d
„ Zd S ) ÚTrainingNetPlayerrI r Úargsc ó — t d¦ « ‚)Nz8Don't use this class yet, it produces incorrect results.)
ÚNotImplementedErrorr1 r$ r2 Ú _TrainingNetPlayer__init_argsrX rI r ÚtraceÚ traced_path)r rI r rX s r r zTrainingNetPlayer.__init__e s € İ!Ğ"\Ñ]Ô]Ğ]r r c ó‚ — dD ];} | |¦ « Œ# t $ r t d|› d�¦ « Y Œ8w xY w|S )N)Úcheckpoint_dirÚ max_depthzKey z
not present.)ÚpoprF Úprint)r rX r s r Ú __init_argsz TrainingNetPlayer.__init_argsm sg € Ø2ğ 1ğ 1ˆCğ
1Ø—’˜‘
”
�
�
øİğ
1ğ
1ğ
1İĞ/˜SĞ/Ğ/Ğ/Ñ0Ô0Ğ0Ğ0Ğ0ğ
1øøøàˆ s † œ <»<r c ó, — |d }|d }|d }n # t $ r t d¦ « ‚w xY wt |||| j | j ¦ « }| j ||¬¦ « }| j |¦ « S )NrA rB rC rD rE )rF Ú
CpSelfPlayÚ CmctsSearchrX r] r rJ rL )r r r rA rB rC rM r7 s r r z TrainingNetPlayer.choose_moveu s € ğ "Ø #Ğ$4Ô 5ˆNؘHÔ%ˆFؘ”-ˆCˆCøİğ "ğ "ğ "İğ !ñ "ô "ğ
"ğ "øøøõ ×
#Ò
# E¨>¸3ÀÄ È4ÔK[Ñ
\Ô
\ˆØÔ ×,Ò,¨R°CĞ,Ñ8Ô8ˆØÔ ×1Ò1°$Ñ7Ô7Ğ7rO c ó — t ‚r )rZ r s r r z%TrainingNetPlayer.make_fresh_instance� s € İ!Ğ!r c ó — || _ d S r r: r" s r r# z"TrainingNetPlayer.set_game_manager„ r; r N)r$ r% r& r r r, r r[ r( r) r* r+ r r r# r r r rW rW d s« € € € € € ğG ğ GĞ<Pğ GĞX\ğ Gğ Gğ Gğ Gğ 4ğ ğ ğ ğ ğ
8 ¤ğ
8¸%ÀÀSÀ¼/ğ
8ğ
8ğ
8ğ
8ğ"ğ "ğ "ğ)ğ )ğ )ğ )ğ )r rW c óP — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z dS ) Ú HumanPlayerr c ón — | j j | _ || _ || _ | |¦ « d S r )r1 r$ r2 r r r r
s r r zHumanPlayer.__init__‰ s7 € Ø”NÔ+ˆŒ Ø (ˆÔØ ˆŒ Ø ×Ò˜Ñ Ô Ğ Ğ Ğ r r r c óp — | j j rt d¦ « ‚| j |¦ « }|S )Nz1Cannot play with a human player in headless mode.)r ÚheadlessÚ RuntimeErrorÚget_human_input)r r r r7 s r r zHumanPlayer.choose_move� s; € Ø Ô Ô %ğ Tİ Ğ RÑSÔSĞ SØÔ ×0Ò0°Ñ7Ô7ˆØˆ r c óT — t | j ¦ « fi | j ¤�S r )rj r r r r s r r z HumanPlayer.make_fresh_instance• s) € ݘ4Ô,×@Ò@ÑBÔBĞRĞRÀdÄkĞRĞRĞRr c ó — || _ d S r r: r" s r r# z HumanPlayer.set_game_manager˜ r; r Nr< r r r rj rj ˆ s} € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ Sğ Sğ Sğ)ğ )ğ )ğ )ğ )r rj c óV — e Zd Zdefd„Zdej deeef fd„Z d„ Z
d„ Z d„ Z d S )
ÚJavaMinimaxPlayerr c óÄ — t j d¦ « t j ¦ « || _ | j j | _ || _ t j | j
¦ « d S )NzFC:\Users\Skyr\IdeaProjects\Minimax\build\libs\Minimax-1.0-SNAPSHOT.jar) ÚjpypeÚ addClassPathÚstartJVMr r1 r$ r2 r ÚatexitÚregisterÚ on_shutdownr
s r r zJavaMinimaxPlayer.__init__� sV € İ
ÔĞdÑeÔeĞeİ
ŒÑÔĞØ (ˆÔØ”NÔ+ˆŒ Ø ˆŒ İŒ˜Ô(Ñ)Ô)Ğ)Ğ)Ğ)r r r c ór — |d }|d }n # t $ r t d¦ « ‚w xY wt j d¦ « } |¦ « }t | j j ¦ « t j |¦ « }| ||||d| j j ¦ « }t |¦ « S )NÚdepthÚplayerú=Missing keyword argument. Please supply kwargs: depth, playerzdev.skyr.MinimaxT)
rF ru ÚJClassrb r Ú
num_to_winÚJArrayÚofÚrunr* ) r r r r| r} ÚMinimaxÚminimaxÚjar7 s r r z JavaMinimaxPlayer.choose_move¦ sº € ğ \ؘ7”OˆEؘHÔ%ˆFˆFøİğ \ğ \ğ \İĞZÑ[Ô[Ğ [ğ \øøøå”,Ğ 1Ñ2Ô2ˆØ�'‘)”)ˆİ
ˆdÔ Ô*Ñ+Ô+Ğ+İ
Œ\�_Š_˜UÑ
#Ô
#ˆØ�{Š{˜5 &¨&°%¸¸tÔ?PÔ?[Ñ\Ô\ˆİ�T‰{Œ{Ğó ‚ “-c ó — d S r r r s r r z%JavaMinimaxPlayer.make_fresh_instance³ s € Ø ˆr c ó, — t j ¦ « d S r )ru Ú shutdownJVMr s r rz z JavaMinimaxPlayer.on_shutdown¶ s € İ
ÔÑÔĞĞĞr c ó — || _ d S r r: r" s r r# z"JavaMinimaxPlayer.set_game_manager¹ r; r N)
r$ r% r& r r r( r) r* r+ r r rz r# r r r rs rs œ sˆ € € € € € ğ* ]ğ *ğ *ğ *ğ *ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ
ğ
ğ
ğ ğ ğ ğ)ğ )ğ )ğ )ğ )r rs c ó¢ — e Zd Zdefd„Zdej deeef fd„Z e
d¦ « e
d¦ « fdej dede d edef
d
„Z d „ Z
d „ Zd
S )Ú
MinimaxPlayerr c ón — || _ | j j | _ || _ | |¦ « d S r r0 r
s r r zMinimaxPlayer.__init__¾ s7 € à (ˆÔØ”NÔ+ˆŒ Ø ˆŒ Ø ×Ò˜Ñ Ô Ğ Ğ Ğ r r r c óŞ — |d }|d }n # t $ r t d¦ « ‚w xY w| | ¦ « |d|¦ « d }t |¦ « S )Nr| r} r~ Té )rF r… rQ r* )r r r r| r} r7 s r r zMinimaxPlayer.choose_moveÅ s{ € ğ \ؘ7”OˆEؘHÔ%ˆFˆFøİğ \ğ \ğ \İĞZÑ[Ô[Ğ [ğ \øøøà�|Š|˜EŸJšJ™LœL¨%°°vÑ>Ô>¸qÔAˆİ�T‰{Œ{Ğr‡ Úinfr| Úis_maxr} c óŠ — | |¦ « }|�|d fS |dk r| |d¬¦ « d fS |rÇt d¦ « }d } | j |¦ « D ]–}
|||
d |
d < | | ¦ « |dz
d| ||¦ « d } d||
d |
d < | |k r|
} t
| |¦ « }t
||¦ « }||k r nŒ—|| fS t d¦ « }d } | j |¦ « D ]‘}
|||
d |
d < | | ¦ « |dz
d| ||¦ « d } | €ŒRd||
d |
d < t | |¦ « }t ||¦ « }||k r nŒ’|| fS )Nr F)Ú check_endr‘ r� T)Ú evaluate_fnÚfloatr Úget_valid_movesr… rQ ÚmaxÚmin) r r r| r’ r} ÚalphaÚbetaÚeval_Ú
best_scoreÚ best_mover7 Úscores r r… zMinimaxPlayer.minimaxÎ s € ğ × Ò Ñ'Ô'ˆØ Рؘ$�;Ğ Ø �AŠ:ˆ:Ø×#Ò# E°UĞ#Ñ;Ô;¸TĞAĞ Aà ğ # )İ ™,œ,˜ ˆJØ ˆIØ Ô)×9Ò9¸%Ñ@Ô@ğ
ğ
�Ø*0��d˜1”g”˜t AœwÑ'Ø Ÿ š U§Z¢Z¡\¤\°5¸1±9¸eÀfÀWÈeĞUYÑZÔZĞ[\Ô]�à*+��d˜1”g”˜t AœwÑ'ؘ:Ò%Ğ%Ø $�Iİ ¨
Ñ 3Ô 3�
ݘE :Ñ.Ô.�à˜D’=�=Ø�Eğ !ğ ˜yĞ(Ğ (å ˜u™ œ ˆJØ ˆIØ Ô)×9Ò9¸%Ñ@Ô@ğ
ğ
�Ø*0��d˜1”g”˜t AœwÑ'Ø Ÿ š U§Z¢Z¡\¤\°5¸1±9¸dÀVÀGÈUĞTXÑYÔYĞZ[Ô\�Ø�=Ø à*+��d˜1”g”˜t AœwÑ'İ ¨
Ñ 3Ô 3�
ݘ4 Ñ,Ô,�à˜5’=�=Ø�Eğ !ğ ˜yĞ(Ğ (r c óT — t | j ¦ « fi | j ¤�S r )r� r r r r s r r z!MinimaxPlayer.make_fresh_instanceü s) € İ ˜TÔ .× BÒ BÑ DÔ DĞTĞTÈÌ ĞTĞTĞTr c ó — || _ d S r r: r" s r r# z MinimaxPlayer.set_game_managerÿ r; r N)r$ r% r& r r r( r) r* r+ r r– Úboolr… r r# r r r r� r� ½ sŞ € € € € € ğ!Ğ%9ğ !ğ !ğ !ğ !ğ ¤ğ ¸%ÀÀSÀ¼/ğ ğ ğ ğ ğ X]ĞW\Ğ]bÑWcÔWcĞVcØ�U˜5‘\”\ğ,)ğ ,)˜RœZğ ,)°ğ ,)¸Tğ ,)È3ğ ,)Ø&+ğ,)ğ ,)ğ ,)ğ ,)ğ\Uğ Uğ Uğ)ğ )ğ )ğ )ğ )r r� )rx rQ Úabcr r ru Únumpyr( Ú$mu_alpha_zero.AlphaZero.Network.nnetr Ú!mu_alpha_zero.Game.tictactoe_gamer Ú mu_alpha_zero.General.az_gamer r r. r> rW rj rs r� r r r ú<module>r¨ s· ğØ
€
€
€
Ø € € € Ø #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #à € € € Ø Ğ Ğ Ğ à =Ğ =Ğ =Ğ =Ğ =Ğ =Ø BĞ BĞ BĞ BĞ BĞ BØ 7Ğ 7Ğ 7Ğ 7Ğ 7Ğ 7ğ
ğ
ğ
ğ
ğ
ˆSñ
ô
ğ
ğ6)ğ )ğ )ğ )ğ )�6ñ )ô )ğ )ğ,#)ğ #)ğ #)ğ #)ğ #)�ñ #)ô #)ğ #)ğL!)ğ !)ğ !)ğ !)ğ !)˜ñ !)ô !)ğ !)ğH)ğ )ğ )ğ )ğ )�&ñ )ô )ğ )ğ( )ğ )ğ )ğ )ğ )˜ñ )ô )ğ )ğBC)ğ C)ğ C)ğ C)ğ C)�Fñ C)ô C)ğ C)ğ C)ğ C)r | 15,725 | Python | .pyt | 55 | 284.672727 | 1,489 | 0.309936 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,566 | arena.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Arena/__pycache__/arena.cpython-311.pyc | §
kŒ¢fH ã ó~ — d dl Z d dlZd dlmZ d dlmZ d dlmZ d dlm Z d dl
m Z m Z d dl
mZ G d„ d e¦ « ZdS )
é N)ÚPlayer)Ú GeneralArena)Ú HookManager)ÚHookAt)Ú scale_actionÚ
resize_obs)ÚAlphaZeroConfigc óv — e Zd Z ddedepddefd„Z dded ed
ed ed ed
ede pddede
eeef fd„Z d„ Z dS )ÚArenaNFÚalpha_zero_configÚ hook_managerÚ
state_managedc ól — || _ || _ || _ |�|n
t ¦ « | _ || _ d S )N)Ú game_managerr Údevicer r
r )Úselfr r r r
r s úR/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Arena/arena.pyÚ__init__zArena.__init__ s< € à (ˆÔØ *ˆÔØ ˆŒ Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ!2ˆÔ Ğ Ğ ó é Úplayer1Úplayer2Únum_games_to_playÚnum_mc_simulationsÚ
one_playerÚ start_playerÚ
add_to_kwargsÚdebugÚreturnc
óÔ
— ddddœ} | j j }
|r|} n|dz } | j rl|j dk r| j |j _ |j dk r| j |j _ | | j ¦ « | | j ¦ « t |¦ « D �]²} | | k r|}
n| }
||
| j |
| j j dœ}|�|
|¦ « | j j s| j ¦ « }n| j |
¬¦ « }| j j
r�| j j r„ |j j | j |d¦ « d¦ « |j j | j |d ¦ « d ¦ « n# t$ $ r Y nw xY w|j d
k r|j d¦ « |j d
k r|j d¦ « | j ¦ « |
dk r |j |fi |¤�}n |j |fi |¤�}| j | | j j t4 t6 j |||
f¬ ¦ « | j s8| j |||
¦ « }| j |
|¦ « }n;| j ||
¦ « d }| j |
¦ « }| j ¦ « | j j
�ret? |t@ ¦ « r'tC || j " ¦ « ¦ « nCtC |d |j# d z |d z | j " ¦ « ¦ « } |
dk rg|j dk r!|j j $ ||d ¦ « |j j $ | j |d¦ « |d¦ « nf|j dk r!|j j $ ||d¦ « |j j $ | j |d ¦ « |d ¦ « n# t$ $ r Y nw xY w|�Ã|dk r(|
dk r| d
xx dz
cc< nO| dxx dz
cc< n>|d k r(|
dk r| dxx dz
cc< n!| d
xx dz
cc< n| dxx dz
cc< |j dk s |j dk r|rtK j& d¦ « tO j( | d
| d | d dœ¦ « n |
d z }
|
|d< �ŒM�Œ´| d
| d | d fS )aÌ
Pit two players against each other for a given number of games and gather the results.
:param start_player: Which player should start the game.
:param one_player: If True always only the first player will start the game.
:param player1:
:param player2:
:param num_games_to_play:
:param num_mc_simulations:
:return: number of wins for player1, number of wins for player2, number of draws
r )Úwins_p1Úwins_p2Údrawsé Ú NetPlayer)Únum_simulationsÚcurrent_playerr ÚtauÚunravelN)Úplayerr éÿÿÿÿÚ
NetworkPlayerT)Úargsr! r" r# Ú HumanPlayergš™™™™™É?r' ))r Ú arena_taur Únamer Úmonte_carlo_tree_searchÚset_game_managerÚranger r) ÚupdateÚrequires_player_to_resetÚresetÚarena_running_muzeroÚenable_frame_bufferÚbufferÚ init_bufferÚ get_state_for_passive_playerÚAttributeErrorÚ step_rootÚrenderÚ choose_mover
Úprocess_hook_executesÚpitÚ__name__Ú__file__r ÚMIDDLEÚget_next_stateÚ game_resultÚ
isinstanceÚintr Úget_num_actionsÚshapeÚ add_frameÚtimeÚsleepÚwandbÚlog)r r r r r r r r r Úresultsr( Únum_games_per_playerÚgamer' ÚkwargsÚstateÚmoveÚstatuss r rA z Arena.pit sš € ğ ¨A¸Ğ:Ğ:ˆØÔ$Ô.ˆØ ğ :Ø#4Ğ Ğ à#4¸Ñ#9Ğ à Ô ğ 8ØŒ|˜{Ò*Ğ*Ø?CÔ?P�Ô/Ô<ØŒ|˜{Ò*Ğ*Ø?CÔ?P�Ô/Ô<à × $Ò $ TÔ%6Ñ 7Ô 7Ğ 7Ø × $Ò $ TÔ%6Ñ 7Ô 7Ğ 7åĞ+Ñ,Ô,ğ V :ñ V :ˆDØĞ*Ò*Ğ*Ø!-��à". �à);È~ĞimÔitØ ¨TÔ-CÔ-KğMğ MˆFà Ğ(Ø—
’
˜mÑ,Ô,Ğ,ØÔ)ÔBğ
GØ Ô)×/Ò/Ñ1Ô1��à Ô)×/Ò/°~Ğ/ÑFÔF�àÔ%Ô:ğ
¸tÔ?UÔ?iğ
ğØÔ3Ô:×FÒFØ Ô)×FÒFÀuÈaÑPÔPØñô ğ ğ Ô3Ô:×FÒFØ Ô)×FÒFÀuÈbÑQÔQĞSUñWô Wğ Wğ Wøå%ğ ğ ğ Ø�Dğøøøğ Œ|˜Ò.Ğ.ØÔ/×9Ò9¸$Ñ?Ô?Ğ?ØŒ|˜Ò.Ğ.ØÔ/×9Ò9¸$Ñ?Ô?Ğ?ğ8
:ØÔ!×(Ò(Ñ*Ô*Ğ*Ø! QÒ&Ğ&Ø.˜7Ô.¨uĞ?Ğ?¸Ğ?Ğ?�D�Dà.˜7Ô.¨uĞ?Ğ?¸Ğ?Ğ?�DØÔ!×7Ò7¸¸d¼hÔ>OÕQYÕ[aÔ[hØ>BÀFÈNĞ=[ğ 8ñ ]ô ]ğ ]àÔ)ğ KØ Ô -× <Ò <¸UÀDÈ.Ñ YÔ Y�EØ !Ô .× :Ò :¸>È5Ñ QÔ Q�F�Fà Ô -× <Ò <¸TÀ>Ñ RÔ RĞSTÔ U�EØ !Ô .× :Ò :¸>Ñ JÔ J�FØÔ!×(Ò(Ñ*Ô*Ğ*ØÔ)Ô>ñ İV`ĞaeÕgjÑVkÔVkğ n�<¨¨dÔ.?×.OÒ.OÑ.QÔ.QÑRÔRĞRİ$ T¨!¤W¨u¬{¸1¬~Ñ%=ÀÀQÄÑ%GÈÔIZ×IjÒIjÑIlÔIlÑmÔmğ ğ Ø)¨QÒ.Ğ.Ø &œ|¨{Ò :Ğ :Ø 'Ô ?Ô F× PÒ PĞQVĞW[Ğ\^Ñ _Ô _Ğ _Ø #Ô ;Ô B× LÒ LÈTÔM^×M{ÒM{ğ }Bğ DEñ NFô NFğ GKğ LMñ Nô Nğ Nğ Nà &œ|¨{Ò :Ğ :Ø 'Ô ?Ô F× PÒ PĞQVĞW[Ğ\]Ñ ^Ô ^Ğ ^Ø #Ô ;Ô B× LÒ LÈTÔM^×M{ÒM{ğ }Bğ DFñ NGô NGğ HLğ MOñ Pô Pğ Pøøå)ğ ğ ğ à ˜ğ øøøğ Ğ%Ø ’{�{Ø)¨QÒ.Ğ.Ø # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à 2š ˜ Ø)¨QÒ.Ğ.Ø # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à # IĞ .Ğ .Ô .°!Ñ 3Ğ .Ğ .Ñ .Ğ .à Ğ(Ğ(Ô(¨AÑ-Ğ(Ğ(Ñ(ğ œ ¨
Ò5Ğ5¸¼ È Ò9VĞ9VĞ\aĞ9Vİ œ
3™œ˜å”I¨'°)Ô*<ÈĞQZÔI[ĞfmĞnuÔfvĞ wĞ wÑxÔxĞxØà "Ñ$�Ø+9�Ğ'Ñ(ñq8
:ñj ğ �yÔ! 7¨9Ô#5°w¸wÔ7GĞGĞGs& Ä6A2F)Æ)
F6Æ5F6Î CQ0Ñ0
Q=Ñ<Q=c ó$ — t d¦ « }|S )Nz Press any key to continue...)Úinput)r Úinpts r Ú
wait_keypresszArena.wait_keypress� s € İĞ3Ñ4Ô4ˆØˆ r )NF)Fr NF)
rB Ú
__module__Ú __qualname__r r Úboolr r rH ÚdictÚtuplerA rZ © r r r r sí € € € € € àQVğ3ğ 3¸ğ 3Ø *Ğ 2¨dğ3ØJNğ3ğ 3ğ 3ğ 3ğ puğuHğ uH˜6ğ uH¨Fğ uHÀsğ uHĞ`cğ uHØ ğuHØ47ğuHØLPÈLĞTXğuHØhlğuHà �#�s˜C�-Ô ğuHğ uHğ uHğ uHğnğ ğ ğ ğ r r )rL rN Ú%mu_alpha_zero.AlphaZero.Arena.playersr Úmu_alpha_zero.General.arenar Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr Úmu_alpha_zero.MuZero.utilsr r Úmu_alpha_zero.configr r r` r r ú<module>rg sÇ ğØ € € € à € € € à 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8à 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ø 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0ğBğ Bğ Bğ Bğ BˆLñ Bô Bğ Bğ Bğ Br | 7,879 | Python | .pyt | 29 | 268.103448 | 1,827 | 0.361992 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,567 | az_node.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/MCTS/__pycache__/az_node.cpython-311.pyc | §
’Ô†f‚ ã óD — d dl Z d dlZd dlZd dlmZ G d„ d¦ « ZdS )é N)ÚTicTacToeGameManagerc óP — e Zd ZdZdd„Zdd„Zd„ Zd„ Zdd
„Zdd „Z d „ Z
d
„ Z d„ Z dS )Ú
AlphaZeroNodez€
This class defines a node in the search tree. It stores all the information needed for DeepMind's AlphaZero algorithm.
r Nc ó¸ — || _ |dk | _ i | _ |�t j |¦ « nd | _ || _ d | _ || _ d | _ d| _
d S ©Nr ) Ú
times_visitedÚwas_init_with_zero_visitsÚchildrenÚweakrefÚrefÚparentÚselect_probabilityÚqÚcurrent_playerÚstateÚ total_value)Úselfr r r
Útimes_visited_inits úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/MCTS/az_node.pyÚ__init__zAlphaZeroNode.__init__
se € Ø /ˆÔØ);¸qÒ)@ˆÔ&؈Œ
Ø-3Ğ-?•g”k &Ñ)Ô)Ğ)ÀTˆŒ Ø"4ˆÔ ؈ŒØ ,ˆÔ؈Œ
Ø ˆÔĞĞó Úreturnc óª — | ¦ « | _ t |¦ « D ])\ }}t | j dz || ¬¦ « }|| j |< Œ*dS )a6
Expands the newly visited node with the given action probabilities and state.
:param state: np.ndarray of shape (board_size, board_size) representing the state current game board.
:param action_probabilities: list of action probabilities for each action.
:return: None
éÿÿÿÿ)r r
N)Úcopyr Ú enumerater r r
)r r Úaction_probabilitiesÚactionÚ probabilityÚnodes r ÚexpandzAlphaZeroNode.expand sg € ğ —Z’Z‘\”\ˆŒ
å#,Ğ-AÑ#BÔ#Bğ )ğ )Ñ ˆF�Kİ Ô!4¸Ñ!;ĞP[ĞdhĞiÑiÔiˆDØ$(ˆDŒM˜&Ñ !Ğ !ğ )ğ )r c ó2 — t | j ¦ « dk S r )Úlenr
©r s r Ú was_visitedzAlphaZeroNode.was_visited& s € İ�4”=Ñ!Ô! AÒ%Ğ%r c ój — | j € || _ d S | j | j z |z | j dz z | _ d S )Né )r r )r Úvs r Úupdate_qzAlphaZeroNode.update_q) s; € à Œ6ˆ>؈DŒFˆFˆFàÔ(¨4¬6Ñ1°AÑ5¸$Ô:LÈqÑ:PÑQˆDŒFˆFˆFr ç ø?c
óx — t d¦ « }d }d }t j | j dk d| j ¦ « }t j |dk d|¦ « }g }| j ¦ « D ]h\ }}t j || j j ¦ « } || dk rŒ1| |¬¦ « }
| |
¦ « |
|k r|
}|}|}Œid„ | j
¦ « D ¦ « } |€<t d| j d|| | ¦ « |t d d
¦ « ¬ ¦ « ||fS ) NÚinfr éûÿÿÿr' )Úcc óB — g | ] }|j |j |j |j g‘Œ S © )r r r r )Ú.0Úchilds r ú
<listcomp>z0AlphaZeroNode.get_best_child.<locals>.<listcomp>B s8 € ğ 6ğ 6ğ 6Ğhm˜uœ{¨EÔ,?ÀÄÈ%ÔJbĞ cğ 6ğ 6ğ 6r z-Best child is None. Possibly important info:
Ú
zimportant_info.txtÚw)Úfile)ÚfloatÚnpÚwherer r
ÚitemsÚ
unravel_indexÚshapeÚ
calculate_utcÚappendÚvaluesÚprintr% Úopen) r r. Úbest_utcÚ
best_childÚ best_actionÚvalids_for_stateÚutcsr r2 Úaction_Ú child_utcÚprintable_childrens r Úget_best_childz AlphaZeroNode.get_best_child0 so € ݘ%‘L”L�=ˆØˆ
؈ İ œ8 D¤J°!¢O°R¸¼ÑDÔDĞİ œ8Ğ$4¸Ò$9¸1Ğ>NÑOÔOĞØˆØ !œ]× 0Ò 0Ñ 2Ô 2ğ %ğ %‰MˆF�EİÔ& v¨t¬zÔ/?Ñ@Ô@ˆGØ Ô(¨AÒ-Ğ-ØØ ×+Ò+¨aĞ+Ñ0Ô0ˆIØ �KŠK˜ Ñ "Ô "Ğ "ؘ8Ò#Ğ#Ø$�Ø "�
Ø $� øğ 6ğ 6Ø "œm× 2Ò 2Ñ 4Ô 4ğ 6ñ 6ô 6Ğà Ğ å ĞBÀDÄJĞPTØ"Ğ$6¸×8HÒ8HÑ8JÔ8JÈDİğ #Ø$'ñ)ô )ğ
*ñ
*ô
*ğ
*ğ
˜;Ğ&Ğ&r c óú — | ¦ « }| j €(|| j z t j |j dz ¦ « z }n7| j || j t j |j ¦ « d| j z z z z z }|S )Ng:Œ0â�yE>r' )r
r r ÚmathÚsqrtr )r r. r
Úutcs r r= zAlphaZeroNode.calculate_utcM s€ € Ø—’‘ ” ˆØ Œ6ˆ>à�dÔ-Ñ-µ´ ¸&Ô:NĞQUÑ:UÑ0VÔ0VÑVˆCˆCà”&˜1ØÔ+µ´ ¸&Ô:NÑ0OÔ0OĞTUĞX\ÔXjÑTjÑ/kÑlñ nñ nˆCğ ˆ
r c ó: — | j dk r| j | j z ndS r )r r r$ s r Úget_self_valuez AlphaZeroNode.get_self_valueX s& € Ø8<Ô8JÈQÒ8NĞ8NˆtÔ $Ô"4Ñ4Ğ4ĞTUĞUr c óp — | j }i }| j ¦ « D ]\ }}|j |z ||< Œ|S ©N)r r
r: )r Útotal_times_visitedÚ action_probsr r2 s r Ú get_self_action_probabilitiesz+AlphaZeroNode.get_self_action_probabilities[ sO € à "Ô 0Ğ؈ Ø !œ]× 0Ò 0Ñ 2Ô 2ğ Mğ M‰MˆF�EØ#(Ô#6Ğ9LÑ#LˆL˜Ñ Ğ àĞr c ó — | j S rR )r r$ s r Ú
get_latentzAlphaZeroNode.get_latentd s
€ ØŒzĞr )r Nr )r N)r* )
Ú__name__Ú
__module__Ú __qualname__Ú__doc__r r! r% r) rJ r= rP rU rW r0 r r r r s¿ € € € € € ğğ ğ ğ ğ ğ ğ )ğ )ğ )ğ )ğ &ğ &ğ &ğRğ Rğ Rğ'ğ 'ğ 'ğ 'ğ: ğ ğ ğ ğVğ Vğ Vğ ğ ğ ğğ ğ ğ ğ r r )rL r Únumpyr8 Ú!mu_alpha_zero.Game.tictactoe_gamer Ú GameManagerr r0 r r ú<module>r_ sq ğØ € € € Ø €€€Ø Ğ Ğ Ğ à QĞ QĞ QĞ QĞ QĞ Qğ]ğ ]ğ ]ğ ]ğ ]ñ ]ô ]ğ ]ğ ]ğ ]r | 5,913 | Python | .pyt | 35 | 166.514286 | 775 | 0.387651 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,568 | az_search_tree.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/MCTS/__pycache__/az_search_tree.cpython-311.pyc | §
’Ô†fe( ã ó¼ — d dl Z d dlmZ d dlZd dlZd dlmZ d dl m
Z
m Z d dl m
Z
d dlmZ d dlmZ d dlmZ d d lmZ d d
lmZ d d lmZ G d „ d
e¦ « Zd„ ZdS )é N)ÚPool)Ú
AlphaZeroNode)Ú"augment_experience_with_symmetriesÚmask_invalid_actions)ÚTicTacToeGameManager)ÚGeneralMemoryBuffer)ÚGeneralNetwork)Ú
SearchTree)Ú HookManager)ÚHookAt)ÚAlphaZeroConfigc
óø — e Zd Z ddededepdfd„Zdedej de
e e e e f fd „Z
dd
„Zd „ Zd e dz ddfd
„Zd„ Zdedej de dede
e e e f f
d„Zede de dedej de de f d„¦ « ZdS )Ú McSearchTreeNÚ game_managerÚalpha_zero_configÚ hook_managerc ó^ — || _ || _ |�|n
t ¦ « | _ d | _ d S ©N)r r r r Ú root_node)Úselfr r r s úZ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/MCTS/az_search_tree.pyÚ__init__zMcSearchTree.__init__ s2 € à (ˆÔØ!2ˆÔ Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ ˆŒˆˆó ÚnetworkÚdeviceÚreturnc ó¨ ‡ ‡
— | j ¦ « }dŠ g }ddddœ} | ||‰ |¦ « \ }}| j || j j ¬¦ « }| d¦ « | |‰ z |d‰ f¦ « | j || j |¦ « ‰ ¦ « }| j
‰ |¦ « Š
‰
�w‰
‰ k r|dxx dz
cc< n1d‰
cxk rdk rn n|d xx dz
cc< n|d
xx dz
cc< d‰
cxk rdk rn nˆ
fd „|D ¦ « }nˆ ˆ
fd „|D ¦ « }n‰ dz Š �ŒAt || j j ¦ « }| j
| | j j t" t$ j ||f¬
¦ « ||d |d
|d fS )a¸
Plays a single game using the Monte Carlo Tree Search algorithm.
Args:
network: The neural network used for searching and evaluating moves.
device: The device (e.g., CPU or GPU) on which the network is located.
Returns:
A tuple containing the game history, and the number of wins, losses, and draws.
The game history is a list of tuples, where each tuple contains:
- The game state multiplied by the current player
- The policy vector (probability distribution over moves)
- The game result (1 for a win, -1 for a loss, 0 for a draw)
- The current player (-1 or 1)
é r )Ú1ú-1ÚDT)ÚtauNr éÿÿÿÿr! r c ó@ •— g | ]}|d |d ‰|d f‘ŒS ©r r é © )Ú.0ÚxÚrs €r ú
<listcomp>z.McSearchTree.play_one_game.<locals>.<listcomp>A s. ø€ Ğ#PĞ#PĞ#P¸a Q q¤T¨1¨Q¬4°°A°a´DĞ$9Ğ#PĞ#PĞ#Pr c óX •— g | ]&}|d |d ‰‰z |d z |d f‘Œ'S r% r' )r( r) Úcurrent_playerr* s €€r r+ z.McSearchTree.play_one_game.<locals>.<listcomp>C s= ø€ Ğ#hĞ#hĞ#hĞVW Q q¤T¨1¨Q¬4°°^Ñ1CÀaÈÄdÑ1JÈAÈaÌDĞ$QĞ#hĞ#hĞ#hr ©Úargs)r ÚresetÚsearchÚ select_mover r" Ú step_rootÚappendÚget_next_stateÚnetwork_to_boardÚ game_resultr Ú
board_sizer Úprocess_hook_executesÚ
play_one_gameÚ__name__Ú__file__r ÚTAIL) r r r ÚstateÚ game_historyÚresultsÚpiÚ_Úmover- r* s @@r r: zMcSearchTree.play_one_game s/ øø€ ğ" Ô!×'Ò'Ñ)Ô)ˆØˆØˆ Ø ¨Ğ+Ğ+ˆğ !Ø—K’K ¨°ÀÑGÔG‰EˆB�ØÔ$×0Ò0°¸Ô8NÔ8RĞ0ÑSÔSˆDà �NŠN˜4Ñ Ô Ğ à × Ò ¨ Ñ!7¸¸TÀ>Ğ RÑ SÔ SĞ SØÔ%×4Ò4°U¸DÔ<M×<^Ò<^Ğ_cÑ<dÔ<dĞftÑuÔuˆEØÔ!×-Ò-¨n¸eÑDÔDˆA؈}ؘÒ&Ğ&ؘC�L�L”L AÑ%�L�L‘L�Lؘ!�Z�Z’Z�Z˜a’Z�Z�Z�Z�ZؘC�L�L”L AÑ%�L�L‘L�Là˜D�M�M”M QÑ&�M�M‘Mà˜�:�:’:�:˜A’:�:�:�:�:Ø#PĞ#PĞ#PĞ#PÀ<Ğ#PÑ#PÔ#P�L�Là#hĞ#hĞ#hĞ#hĞ#hĞ[gĞ#hÑ#hÔ#h�LØØ ˜bÑ ˆNñ- !õ2 :¸,ÈÔHYÔHdÑeÔeˆ Ø Ô×/Ò/°°dÔ6HÔ6QÕS[Õ]cÔ]hØ6BÀGĞ5Lğ 0ñ Nô Nğ Nà˜W Sœ\¨7°4¬=¸'À#¼,ĞFĞFr c ó� — | j j }|€ | j j }| j €t |d¬¦ « | _ | j ||¦ « }t j |t j |¬¦ «
d¦ « }| |d¬¦ « \ }} |t j
| j j g| j j z ¦ « dd¦ « z }t% || j | ¦ « |¦ « | j j ¦ « }| ¦ « ¦ « }| j ||¦ « t3 |¦ « D �]n}
| j } | g} d}
| ¦ « r¢| | j j ¬ ¦ « \ } }
| €Tt j | j d
¦ « t j | ¦ « d |› d �¦ « t? d
¦ « ‚| | ¦ « | ¦ « °¢| j ! | " ¦ « j# | j $ |
¦ « | " ¦ « j% ¦ « }| j || j% ¦ « }| j & | j% |¦ « } | €ït j |t j |¬¦ «
d¦ « }| |d¬¦ « \ }} t% || j || j% ¦ « | j j ¦ « }| ¦ « ¦ « d } | ¦ « ¦ « }| ||¦ « | ' | | ¦ « �Œp| j( ) | | j* j+ tX tZ j. || j f¬¦ « | j / ¦ « dfS )zó
Perform a Monte Carlo Tree Search on the current state starting with the current player.
:param tau:
:param network:
:param state:
:param current_player:
:param device:
:return:
Nr )Útimes_visited_init)Údtyper F)Úmuzeror r# )Úcz root_node.ptÚnetwork_none_checkpoint_z.ptzcurrent_node is Noner. )0r Únum_simulationsr" r r r Úget_canonical_formÚthÚtensorÚfloat32Ú unsqueezeÚpredictÚnpÚrandomÚ dirichletÚdirichlet_alphaÚnet_action_sizeÚreshaper Úget_invalid_actionsÚcopyr8 ÚflattenÚtolistÚexpandÚrangeÚ was_visitedÚget_best_childrH ÚsaveÚ
state_dictÚ
ValueErrorr4 r5 Úparentr> r6 r- r7 Úbackpropr r9 r1 r; r< r r= Ú get_self_action_probabilities)r r r> r- r r" rJ Ústate_Ú
probabilitiesÚvÚ
simulationÚ current_nodeÚpathÚactionÚ
next_stateÚ next_state_s r r1 zMcSearchTree.searchM s
€ ğ Ô0Ô@ˆØ ˆ;ØÔ(Ô,ˆCØ Œ>Ğ !İ *¨>ÈaĞ PÑ PÔ PˆDŒNØÔ"×5Ò5°e¸^ÑLÔLˆå”˜6¬¸FĞCÑCÔC×MÒMÈaÑPÔPˆØ"Ÿ?š?¨6¸%˜?Ñ@Ô@ш
�qØ%¬ ×(;Ò(;Ø
Ô
#Ô
3Ğ 4°tÔ7MÔ7]Ñ ]ñ)_ô )_ß_fÒ_fĞghØgiñ`kô `kñkˆ
õ -¨]Ø-1Ô->×-RÒ-RĞSX×S]ÒS]ÑS_ÔS_ĞaoÑ-pÔ-pØ-1Ô->Ô-IñKô Kˆ
ğ &×-Ò-Ñ/Ô/×6Ò6Ñ8Ô8ˆ
Ø Œ× Ò ˜e ]Ñ3Ô3Ğ3İ Ñ0Ô0ğ #ñ #ˆJØ œ>ˆLØ �>ˆD؈FØ ×*Ò*Ñ,Ô,ğ
*Ø'3×'BÒ'BÀTÔE[ÔE]Ğ'BÑ'^Ô'^Ñ$� ˜fØ Ğ'İ”G˜DœN¨NÑ;Ô;Ğ;İ”G˜G× .Ò .Ñ 0Ô 0Ğ2`È^Ğ2`Ğ2`Ğ2`ÑaÔaĞaİ$Ğ%;Ñ<Ô<Ğ<Ø— ’ ˜LÑ)Ô)Ğ)ğ
×*Ò*Ñ,Ô,ğ
*ğ Ô*×9Ò9¸,×:MÒ:MÑ:OÔ:OÔ:UØ:>Ô:K×:\Ò:\Ğ]cÑ:dÔ:dØ:F×:MÒ:MÑ:OÔ:OÔ:^ñ`ô `ˆJğ Ô+×>Ò>¸zÈ<ÔKfÑgÔgˆKØÔ!×-Ò-¨lÔ.IÈ:ÑVÔVˆA؈yå œi¨ ½2¼:ÈfĞ UÑ UÔ U× _Ò _Ğ`aÑ bÔ b� Ø#*§?¢?°;Àu ?Ñ#MÔ#MÑ �
˜qİ 4°]ÀDÔDU×DiÒDiĞjtØjvô kFñEGô EGà59Ô5FÔ5Qñ!Sô !S�
ğ —I’I‘K”K×&Ò&Ñ(Ô(¨Ô+�Ø -× 5Ò 5Ñ 7Ô 7× >Ò >Ñ @Ô @�
Ø ×#Ò# J°
Ñ>Ô>Ğ>à �MŠM˜!˜TÑ "Ô "Ğ "Ñ "à Ô×/Ò/°°d´kÔ6JÍHÕV\ÔVaØ69¸4¼>Ğ5Jğ 0ñ Lô Lğ LàŒ~×;Ò;Ñ=Ô=¸tĞCĞCr c ó� — t |¦ « D ]<}|dz }|xj |z
c_ | |¦ « |xj dz
c_ Œ=dS )a
Backpropagates the value `v` through the search tree, updating the relevant nodes.
Args:
v (float): The value to be backpropagated.
path (list): The path from the leaf node to the root node.
Returns:
None
r# r N)ÚreversedÚ total_valueÚupdate_qÚ
times_visited)r rg rj Únodes r rc zMcSearchTree.backprop‹ si € õ ˜T‘N”Nğ $ğ $ˆDØ
�‰GˆAØ Ğ Ô Ñ !Ğ Ô Ø �MŠM˜!Ñ Ô Ğ Ø Ğ Ô !Ñ #Ğ Ô Ğ ğ $ğ $r Úactionsc ó´ — |�N| j �E| j ¦ « sd S |D ]}| j j | | _ Œd | j _ d S d S d | _ d S r )r r] Úchildrenrb )r rt rk s r r3 zMcSearchTree.step_rootœ st € Ø Ğ ØŒ~Ğ)Ø”~×1Ò1Ñ3Ô3ğ Ø�FØ %ğ Eğ E�FØ%)¤^Ô%<¸VÔ%D�D”N�NØ(,�”Ô%Ğ%Ğ%ğ *Ğ)ğ "ˆDŒNˆNˆNr c óZ — t | j ¦ « | j ¦ « S r )r r Úmake_fresh_instancer )r s r rx z McSearchTree.make_fresh_instance¨ s$ € ݘDÔ -× AÒ AÑ CÔ CÀTÔE[Ñ\Ô\Ğ\r ÚnetÚ num_gamesÚmemoryc óº — d\ }}}t |¦ « D ]A}| ||¦ « \ } }
} } ||
z
}|| z
}|| z
}| | ¦ « ŒB|||fS ©N©r r r )r\ r: Úadd_list)
r ry r rz r{ Úwins_p1Úwins_p2ÚdrawsÚgameÚ game_resultsÚwins_p1_Úwins_p2_Údraws_s
r Ú self_playzMcSearchTree.self_play« s‚ € à")Ñ ˆ�˜%ݘ)Ñ$Ô$ğ *ğ *ˆDØ7;×7IÒ7IÈ#ÈvÑ7VÔ7VÑ 4ˆL˜( H¨fØ �xÑ ˆGØ �xÑ ˆGØ �V‰OˆEØ �OŠO˜LÑ )Ô )Ğ )Ğ )à˜ Ğ&Ğ&r ÚnetsÚtreesÚnum_jobsc
ó( ‡ ‡‡‡‡‡— t ‰¦ « 5 }‰j sF| t ˆˆ ˆˆˆfd„t t ‰ ¦ « ¦ « D ¦ « ¦ « }nF| t ˆˆˆ ˆˆˆfd„t t ‰ ¦ « ¦ « D ¦ « ¦ « }d d d ¦ « n # 1 swxY w Y d\ }} }
|D ]E} || d z
}| | d z
} |
| d z
}
‰j s‰ | d ¦ « ŒF|| |
fS )Nc ó` •— g | ]*}‰| ‰| t j ‰¦ « ‰‰z d f‘Œ+S r ©rX Údeepcopy)r( Úir r‰ rz r‹ rŠ s €€€€€r r+ z3McSearchTree.parallel_self_play.<locals>.<listcomp>½ sK ø€ ğ %7ğ %7ğ %7Ğqr d¨1¤g¨u°Q¬x½¼ ÀvÑ9NÔ9NĞPYĞ]eÑPeĞgkĞ%lğ %7ğ %7ğ %7r c ó„ •— g | ]<}‰| ‰| t j ‰¦ « ‰‰z t j ‰¦ « f‘Œ=S r' r� )r( r� r r{ r‰ rz r‹ rŠ s €€€€€€r r+ z3McSearchTree.parallel_self_play.<locals>.<listcomp>À sU ø€ ğ 2&ğ 2&ğ 2&Øqr�T˜!”W˜e Aœh¬
°fÑ(=Ô(=¸yÈHÑ?TÕVZÔVcĞdjÑVkÔVkĞlğ2&ğ 2&ğ 2&r r~ r r é r& )r Úis_diskÚstarmapÚ p_self_playr\ Úlenr ) r‰ rŠ r{ r rz r‹ Úpr@ r€ r� r‚ Úresults `````` r Úparallel_self_playz McSearchTree.parallel_self_play· sª øøøøøø€ õ �(‰^Œ^ğ '˜qØ”>ğ
'ØŸ)š)¥Kğ%7ğ %7ğ %7ğ %7ğ %7ğ %7ğ %7ğ %7İ%*3¨t©9¬9Ñ%5Ô%5ğ%7ñ %7ô %7ñ8ô 8��ğ Ÿ)š)¥Kğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&å�#˜d™)œ)Ñ$Ô$ğ2&ñ 2&ô 2&ñ 'ô '�ğ
'ğ 'ğ 'ñ 'ô 'ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'øøøğ 'ğ 'ğ 'ğ 'ğ #*Ñ ˆ�˜%Ø ğ +ğ +ˆFØ �v˜a”yÑ ˆGØ �v˜a”yÑ ˆGØ �V˜A”YÑ ˆEØ”>ğ
+Ø—’ q¤ Ñ*Ô*Ğ*øؘ Ğ&Ğ&s –BB6Â6B:Â=B:r )r; Ú
__module__Ú __qualname__r r
r r r rL r ÚtupleÚlistÚintr: r1 rc r3 rx r rˆ Ú staticmethodr™ r' r r r r sŒ € € € € € à59ğ ğ Ğ%9ğ Èoğ Ø *Ğ 2¨dğ ğ ğ ğ ğ1G ^ğ 1G¸R¼Yğ 1GÈ5ĞQUĞWZĞ\_ĞadĞQdÔKeğ 1Gğ 1Gğ 1Gğ 1Gğf<Dğ <Dğ <Dğ <Dğ|$ğ $ğ $ğ"
" ¨¡ğ
"°ğ
"ğ
"ğ
"ğ
"ğ]ğ ]ğ ]ğ
'˜^ğ
'°R´Yğ
'È3ğ
'ĞXkğ
'ĞpuØ ˆS�#ˆ
ôqğ
'ğ
'ğ
'ğ
'ğ ğ' ğ '¨dğ 'Ğ<Oğ 'ĞY[ÔYbğ 'Ğorğ 'Ø%(ğ'ğ 'ğ 'ñ „\ğ'ğ 'ğ 'r r c óş — d\ }}}g }t |¦ « D ]Y} | | |¦ « \ }
} } }
|| z
}|| z
}||
z
}|�| |
¦ « ŒD| |
¦ « ŒZ|€||||fS |||fS r} )r\ r: r Úextend)ry Útreer rz r{ r€ r� r‚ Údatarƒ r„ Úwp1Úwp2Údss r r• r• Í s´ € Ø %Ñ€GˆW�eØ
€Dİ�iÑ Ô ğ &ğ &ˆØ%)×%7Ò%7¸¸VÑ%DÔ%DÑ"ˆ �c˜3 Ø�3‰ˆØ�3‰ˆØ
�‰ ˆØ Ğ Ø �OŠO˜LÑ )Ô )Ğ )Ğ )à �KŠK˜ Ñ %Ô %Ğ %Ğ %Ø
€~ؘ ¨Ğ,Ğ,Ø �G˜UĞ "Ğ"r ) rX Úmultiprocessingr ÚnumpyrQ ÚtorchrL Ú$mu_alpha_zero.AlphaZero.MCTS.az_noder Ú mu_alpha_zero.AlphaZero.utilsr r Ú!mu_alpha_zero.Game.tictactoe_gamer Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr Ú!mu_alpha_zero.General.search_treer
Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr Úmu_alpha_zero.configr
r r• r' r r ú<module>r³ s, ğØ € € € Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ø Ğ Ğ Ğ à >Ğ >Ğ >Ğ >Ğ >Ğ >Ø bĞ bĞ bĞ bĞ bĞ bĞ bĞ bØ BĞ BĞ BĞ BĞ BĞ BØ <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0ğx'ğ x'ğ x'ğ x'ğ x'�:ñ x'ô x'ğ x'ğv#ğ #ğ #ğ #ğ #r | 14,571 | Python | .pyt | 85 | 167.270588 | 1,422 | 0.350549 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,569 | nnet.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Network/__pycache__/nnet.cpython-311.pyc | §
*¤fZD ã óî — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dl mc m
Z d dl Z d dl m
Z
d dlmZ d dlmZ d dlmZ d dlmZ d dlmZmZ d dlmZ d d lm Z G d
„ d ej e¦ « Z G d „ d
ej e¦ « Z G d„ dej j ¦ « Z G d„ dej j ¦ « Z! G d„ dej j ¦ « Z" G d„ dej j ¦ « Z# G d„ dej ¦ « Z$dS )é N)Úmse_loss)ÚT)ÚGeneralAlphZeroNetwork)Ú HookManager)ÚHookAt)ÚAlphaZeroConfigÚConfig)Ú MemBuffer)Úinvert_scale_reward_valuec
ó ‡ — e Zd Z d dedededededepdf ˆ fd„
Zd d
efd „Z e j
¦ « d d „¦ « Z d defd„Z dede
fd„Zd d„Zd„ Zed dedepdfd„¦ « Zdedefd„Zde j fd„Zd„ Zd „ Zˆ xZS )!Ú AlphaZeroNetNÚ in_channelsÚ num_channelsÚdropoutÚ action_sizeÚlinear_input_sizeÚ hook_managerc ó •— t t | ¦ « ¦ « || _ || _ || _ || _ || _ |�|n
t ¦ « | _ t j ||dd¬¦ « | _ t j
|¦ « | _ t j ||dd¬¦ « | _ t j
|¦ « | _ t j ||dd¬¦ « | _ t j
|¦ « | _ t j ||d¦ « | _ t j
|¦ « | _ t j | j d¦ « | _ t j d¦ « | _ t j dd¦ « | _ t j d¦ « | _ t j |¦ « | _ t j d|¦ « | _ t j dd¦ « | _ t? j | j! ¦ « d S )Né é ©Úpaddingi i )"Úsuperr
Ú__init__r r Ú dropout_pr r r r ÚnnÚConv2dÚconv1Ú BatchNorm2dÚbn1Úconv2Úbn2Úconv3Úbn3Úconv4Úbn4ÚLinearÚfc1Ú BatchNorm1dÚfc1_bnÚfc2Úfc2_bnÚDropoutr ÚpiÚvÚatexitÚregisterÚ clear_traces)Úselfr r r r r r Ú __class__s €úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Network/nnet.pyr zAlphaZeroNet.__init__ s¡ ø€ å
�l˜DÑ!Ô!×*Ò*Ñ,Ô,Ğ,Ø&ˆÔØ (ˆÔØ ˆŒØ&ˆÔØ!2ˆÔ Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔå”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1Ñ=Ô=ˆŒ
İ”> ,Ñ/Ô/ˆŒõ ”9˜TÔ 3°TÑ:Ô:ˆŒİ”n TÑ*Ô*ˆŒ İ”9˜T 3Ñ'Ô'ˆŒİ”n SÑ)Ô)ˆŒ å”z 'Ñ*Ô*ˆŒ õ ”)˜C Ñ-Ô-ˆŒİ”˜3 Ñ"Ô"ˆŒİŒ˜Ô)Ñ*Ô*Ğ*Ğ*Ğ*ó TÚmuzeroc ó6 — |s| d¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | |
|¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j |
| |¦ « ¦ « ¦ « }| |¦ « }t j | | |¦ « ¦ « ¦ « }| |¦ « }t j | |¦ « d¬¦ « }t j | |¦ « ¦ « }||fS )Nr r éÿÿÿÿ©Údim)Ú unsqueezeÚFÚrelur r r" r! r$ r# r& r% ÚreshapeÚsizer* r( r r, r+ Ú log_softmaxr. Útanhr/ ©r3 Úxr7 r. r/ s r5 ÚforwardzAlphaZeroNet.forward7 sr € Øğ Ø— ’ ˜A‘”ˆAİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆà
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆå
Œ]˜4Ÿ7š7 1™:œ:¨1Ğ
-Ñ
-Ô
-ˆİ
ŒF�4—6’6˜!‘9”9Ñ Ô ˆà�1ˆuˆ r6 c ó> — | ||¬¦ « \ }}t j |¦ « }| ¦ « ¦ « ¦ « | ¦ « ¦ « ¦ « fS ©N©r7 ©rE ÚthÚexpÚdetachÚcpuÚnumpyrC s r5 ÚpredictzAlphaZeroNet.predictL óo € à— ’ ˜Q v� Ñ.Ô.‰ˆˆAİ
ŒV�B‰ZŒZˆØ�yŠy‰{Œ{�ŠÑ Ô ×&Ò&Ñ(Ô(¨!¯(ª(©*¬*¯.ª.Ñ*:Ô*:×*@Ò*@Ñ*BÔ*BĞBĞBr6 é
Ú
board_sizec ó� — t j | t j dd||¦ « ¦ « ¦ « S )Nr é )rJ ÚjitÚtraceÚrandÚcuda)r3 rR s r5 Úto_traced_scriptz AlphaZeroNet.to_traced_scriptR s4 € İŒv�|Š|˜D¥"¤'¨!¨S°*¸jÑ"IÔ"I×"NÒ"NÑ"PÔ"PÑQÔQĞQr6 Úreturnc ó` — | |¬¦ « }d}| |¦ « |S )N)rR z traced.pt)rY Úsave)r3 rR ÚtracedÚpaths r5 rV zAlphaZeroNet.traceU s4 € Ø×&Ò&°*Ğ&Ñ=Ô=ˆØˆØ� Š �DÑÔĞ؈ r6 c ó~ — ddl m} t j |¦ « › d�¦ « D ]}t j |¦ « Œd S )Nr )Úfind_project_rootz/Checkpoints/Traces/*.pt)Úmu_alpha_zero.General.utilsr` ÚglobÚosÚremove)r3 r` Ú
trace_files r5 r2 zAlphaZeroNet.clear_traces[ s] € ØAĞAĞAĞAĞAĞAİ œ)Ğ'8Ğ'8Ñ':Ô':Ğ$TĞ$TĞ$TÑUÔUğ "ğ "ˆJİ ŒI�jÑ !Ô !Ğ !Ğ !ğ "ğ "r6 c óZ — t | j | j | j | j | j ¦ « S ©N)r
r r r r r ©r3 s r5 Úmake_fresh_instancez AlphaZeroNet.make_fresh_instance` s. € ݘDÔ ,¨dÔ.?ÀÄ ĞQUÔQaØ Ô 2ñ4ô 4ğ 4r6 Úconfigc ó^ — t | j | j | j | j | j |¬¦ « S )N)r )r
Únum_net_in_channelsÚnum_net_channelsÚ net_dropoutÚnet_action_sizeÚaz_net_linear_input_size)rj r s r5 Úmake_from_configz AlphaZeroNet.make_from_configd s; € å˜FÔ 6¸Ô8OĞQWÔQcØ "Ô 2°FÔ4SĞbnğpñ pô pğ pr6 Ú
memory_bufferÚalpha_zero_configc ó” — ddl m} t j t j ¦ « rdnd¦ « }g }t j | ¦ « |j ¬¦ « }|
¦ « t |j ¦ « D �]Ñ} ||j
¦ « D �]¼}t |¦ « dk rŒt |� \ } }
} t j t# j | ¦ « t j |¬¦ « } t j t# j |
¦ « t j |¬¦ « }
t j | t j |¬¦ « d¦ « } | | d¬ ¦ « \ } }
|| ¦ « }t- |
| ¦ « | | |
||¦ « z }| | ¦ « ¦ « | ¦ « | ¦ « | ¦ « | j | | j j tB tD j# || ¦ « |f¬
¦ « �Œ¾�ŒÓ| j | | j j tB tD j$ |f¬
¦ « tK |¦ « t |¦ « z |fS ) Nr ©Úmask_invalid_actions_batchrX rM )Úlrr ©ÚdtypeÚdeviceFrH ©Úargs)&Ú mu_alpha_zero.AlphaZero.utilsrv rJ rz rX Ú is_availableÚoptimÚAdamÚ
parametersrw ÚshuffleÚrangeÚepochsÚ
batch_sizeÚlenÚzipÚtensorÚnpÚarrayÚfloat32r< rE r Úpi_lossÚappendÚitemÚ zero_gradÚbackwardÚstepr Úprocess_hook_executesÚ train_netÚ__name__Ú__file__r ÚMIDDLEÚTAILÚsum)r3 rr rs rv rz ÚlossesÚ optimizerÚepochÚexperience_batchÚstatesr. r/ Úpi_predÚv_predÚmasksÚlosss r5 r“ zAlphaZeroNet.train_neti sƒ € ØLĞLĞLĞLĞLĞLİ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆØˆİ”H—M’M $§/¢/Ñ"3Ô"3Ğ8IÔ8L�MÑMÔMˆ Ø× Ò Ñ Ô Ğ İĞ,Ô3Ñ4Ô4ğ eñ eˆEØ$1 MĞ2CÔ2NÑ$OÔ$Oğ
eñ
eĞ İĞ'Ñ(Ô(¨AÒ-Ğ-Ø İ #Ğ%5Ğ 6‘
�˜˜Aİœ¥2¤8¨FÑ#3Ô#3½2¼:ÈfĞUÑUÔU�İ”Y�rœx¨™|œ|µ2´:ÀfĞMÑMÔM�İ”I˜a¥r¤z¸&ĞAÑAÔA×KÒKÈAÑNÔN�Ø"&§,¢,¨v¸e ,Ñ"DÔ"D‘�˜Ø2Ğ2°6Ñ:Ô:�İ ¨Ñ*Ô*¨T¯\ª\¸'À2ÀuÈfÑ-UÔ-UÑU�Ø—
’
˜dŸiši™kœkÑ*Ô*Ğ*ğ ×#Ò#Ñ%Ô%Ğ%Ø—
’
‘”�Ø—’Ñ Ô Ğ ØÔ!×7Ò7¸¸d¼nÔ>UÕW_ÕagÔanØ>NĞPT×PYÒPYÑP[ÔP[Ğ]bĞ=cğ 8ñ eô eğ eñ eñ!
eğ&
Ô×/Ò/°°d´nÔ6MÍxÕY_ÔYdĞlrĞktĞ/ÑuÔuĞuİ�6‰{Œ{�S ™[œ[Ñ(¨&Ğ0Ğ0r6 rz c óÌ — | |j ¦ « |¦ « }||z }t j ||z ¦ « | ¦ « d z S ©Nr ©r? ÚshapeÚtorJ r˜ r@ ©r3 Úy_hatÚyr rz Ú masked_y_hats r5 rŒ zAlphaZeroNet.pi_loss† óU € Ø—
’
˜eœkÑ*Ô*×-Ò-¨fÑ5Ô5ˆØ ˜u‘}ˆ İ”�q˜<Ñ'Ñ(Ô(Ğ(¨1¯6ª6©8¬8°A¬;Ñ6Ğ6r6 c ó\ — | ¦ « D ]}| ¦ « Œd S rg )r� Ú
share_memory_)r3 Úparams r5 Úto_shared_memoryz AlphaZeroNet.to_shared_memory‹ s: € Ø—_’_Ñ&Ô&ğ "ğ "ˆEØ × Ò Ñ !Ô !Ğ !Ğ !ğ "ğ "r6 c ór — | j | | j j t t
j ¦ « d S rg )r r’ Úrun_at_training_endr” r• r ÚALLrh s r5 r± z AlphaZeroNet.run_at_training_end� s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr6 rg ©T)rQ )rZ N)r” Ú
__module__Ú __qualname__ÚintÚfloatr r ÚboolrE rJ Úno_gradrO rY ÚstrrV r2 ri Ú staticmethodr rq r
r“ rz rŒ r¯ r± Ú
__classcell__©r4 s @r5 r
r
sÒ ø€ € € € € à59ğ +ğ + Cğ +°sğ +ÀUğ +ĞY\ğ +Ğqtğ +Ø *Ğ 2¨dğ +ğ +ğ +ğ +ğ +ğ +ğBğ ğ ğ ğ ğ ğ* €R„Z�\„\ğCğ Cğ Cñ „\ğCğ
Rğ R¨3ğ Rğ Rğ Rğ Rğ ğ ¨ğ ğ ğ ğ ğ "ğ "ğ "ğ "ğ
4ğ 4ğ 4ğ ğpğ p ğ pÀ Ğ@SÈtğ pğ pğ pñ „\ğpğ1 yğ 1À_ğ 1ğ 1ğ 1ğ 1ğ:7¨r¬yğ 7ğ 7ğ 7ğ 7ğ
"ğ "ğ "ğoğ oğ oğ oğ oğ oğ or6 r
c óT ‡ — e Zd Zddgddddfdedededed ee d
ed ed ed
ededee depddededef ˆ fd„
Zd$dedefd„Z e
j ¦ « d%d„¦ « Z de
j
fd„Zd„ Zed&d ed ddepdfd „¦ « Zd ed eeee f fd!„Zd ed dfd"„Zd#„ Zˆ xZS )'ÚOriginalAlphaZerNetworké Né Fr r r r r Ú support_sizeÚstate_linear_layersÚpi_linear_layersÚv_linear_layersÚlinear_head_hidden_sizeÚ latent_sizer Ú
num_blocksr7 Ú is_dynamicsc ó •‡— t t | ¦ « ¦ « || _ ‰| _ || _ || _ || _ || _ |
| _ || _
| | _ || _ || _
|| _ | | _ |
| _ d | _ | �| n
t% ¦ « | _ t) j |‰dd¬¦ « | _ t) j ‰¦ « | _ t) j |¦ « | _ t) j ˆfd„t9 |
¦ « D ¦ « ¦ « | _ t= ||d |‰| |
¦ « | _ |r tA |d ‰| ||
¦ « | _! d S tE ||d ‰||
¦ « | _! d S )Nr r r c ó0 •— g | ]}t ‰‰¦ « ‘ŒS © )ÚOriginalAlphaZeroBlock)Ú.0Ú_r s €r5 ú
<listcomp>z4OriginalAlphaZerNetwork.__init__.<locals>.<listcomp>° s% ø€ Ğ$sĞ$sĞ$sĞ\]Õ%;¸LÈ,Ñ%WÔ%WĞ$sĞ$sĞ$sr6 r é )#r r¿ r r r r r r r rÈ r7 rÇ rÉ rà rÄ rÅ rÆ rš r r r r r r r r- r Ú
ModuleListrƒ ÚblocksÚ ValueHeadÚ
value_headÚ StateHeadÚ policy_headÚ
PolicyHead)r3 r r r r r r rà rÄ rÅ rÆ rÇ r rÈ r7 rÉ r4 s ` €r5 r z OriginalAlphaZerNetwork.__init__• s¡ øø€ õ Õ% tÑ,Ô,×5Ò5Ñ7Ô7Ğ7Ø&ˆÔØ (ˆÔØ ˆŒØ&ˆÔØ!2ˆÔ Ø (ˆÔØ$ˆŒØ ˆŒ Ø&ˆÔØ&ˆÔØ#6ˆÔ Ø 0ˆÔ Ø .ˆÔ Ø'>ˆÔ$Ø ˆŒØ,8Ğ,D˜L˜LÍ+É-Ì-ˆÔå”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”z 'Ñ*Ô*ˆŒ İ”mĞ$sĞ$sĞ$sĞ$sÕafĞgqÑarÔarĞ$sÑ$sÔ$sÑtÔtˆŒ İ# FĞ,=¸aÔ,@À,ĞP\Ğ^mØ$;ñ=ô =ˆŒà ğ Cİ (Ğ):¸1Ô)=¸|È[ĞZmØ)@ñ Bô BˆDÔ Ğ Ğ õ *¨+Ğ7HÈÔ7KÈ\Ğ[kØ*Añ Cô CˆDÔ Ğ Ğ r6 Úreturn_supportc ó� — |s| d¦ « }t j | | |¦ « ¦ « ¦ « }| j D ]
} ||¦ « }Œ| |¦ « }| |¦ « }| |¦ « }|sœt j
|¦ « }t j | j | j dz dt j
|j ¬¦ « d¦ « }t j ||z d¬¦ « }t! |¦ « }|| d¦ « fS ||fS )Nr rx r r: )r< r= r> r r rÓ r rÕ r× rJ rK Úaranger r‹ rz r˜ r ) r3 rD r7 rÙ ÚblockÚ val_h_outputÚ pol_h_outputÚ
support_rangeÚoutputs r5 rE z OriginalAlphaZerNetwork.forwardº s6 € Øğ Ø— ’ ˜A‘”ˆAİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆØ”[ğ ğ ˆEØ��a‘”ˆAˆAØ �LŠL˜‰OŒOˆØ—’ qÑ)Ô)ˆ Ø×'Ò'¨Ñ*Ô*ˆ Ø ğ 5å œ6 ,Ñ/Ô/ˆLİ œI tÔ'8Ğ&8¸$Ô:KÈaÑ:OĞQRÕZ\ÔZdØ-.¬Xğ 7ñ 7ô 7ß7@²yÀ±|´|ğ
å”V˜L¨=Ñ 8¸aĞ@Ñ@Ô@ˆFİ.¨vÑ6Ô6ˆFØ ×!1Ò!1°!Ñ!4Ô!4Ğ4Ğ 4ؘ\Ğ)Ğ)r6 Tc ó> — | ||¬¦ « \ }}t j |¦ « }| ¦ « ¦ « ¦ « | ¦ « ¦ « ¦ « fS rG rI rC s r5 rO z OriginalAlphaZerNetwork.predictÍ rP r6 rz c óÌ — | |j ¦ « |¦ « }||z }t j ||z ¦ « | ¦ « d z S r£ r¤ r§ s r5 rŒ z OriginalAlphaZerNetwork.pi_lossÓ r« r6 c óÔ — t | j | j | j | j | j | j | j | j | j | j
| j | j | j
| j | j ¬¦ « S )N)r rÈ r7 rÉ )r¿ r r r r r r rà rÄ rÅ rÆ rÇ r rÈ r7 rÉ rh s r5 ri z+OriginalAlphaZerNetwork.make_fresh_instanceØ sn € İ& tÔ'7¸Ô9JÈDÌNĞ\`Ô\lØ'+Ô'=¸tÔ?PØ'+Ô'?ÀÔAVĞX\ÔXlØ'+Ô'CØ'+Ô'7Ø48Ô4EØ26´/È$Ì+ĞcgÔcsğ
uñ uô uğ ur6 rj Ú game_managerc ó¾ — t |j |j |j |j |j ||j |j |j |j |j
|j |j |j
¬¦ « S )N) r rà rÄ rÅ rÆ rÈ r7 r rÇ )r¿ rl rm rn ro rp rà rÄ rÅ rÆ rÈ r7 r Únet_latent_size)Úclsrj rä r s r5 rq z(OriginalAlphaZerNetwork.make_from_configá si € å& vÔ'AÀ6ÔCZĞ\bÔ\nØ'-Ô'=¸vÔ?^Ø4@Ø;AÔ;UØ8>Ô8OØ7=Ô7MØ?EÔ?]Ø28Ô2CÈFÌMØ4:Ô4GĞU[ÔUkğmñ mô mğ mr6 Úmuzero_alphazero_configrZ c ó> — g }| j €Ct j | ¦ « |j |j ¬¦ « | _ | ¦ « t |j ¦ « D �]N} ||j
¦ « D �]9}t |¦ « dk rŒ| ||¦ « \ }}}|
| ¦ « ¦ « t j | ¦ « | ¦ « | ¦ « dœ¦ « | j ¦ « | ¦ « | j ¦ « | j | | j j t0 t2 j || ¦ « |f¬¦ « �Œ;�ŒP| j | | j j t0 t2 j |f¬¦ « t9 |¦ « t |¦ « z |fS )N©rw Ú weight_decayr )Ú
combined_lossÚloss_vÚloss_pir{ ) rš rJ r r€ r� rw Úl2r‚ rƒ r„ r… r† Úcalculate_lossr� r� ÚwandbÚlogr� r� r‘ r r’ r“ r” r• r r– r— r˜ ) r3 rr rè r™ r› rœ r¡ Úv_lossrŒ s r5 r“ z!OriginalAlphaZerNetwork.train_netí sõ € ØˆØ Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<ĞAXÔA[Ø8OÔ8Rğ +ñ Tô TˆDŒNà× Ò Ñ Ô Ğ İĞ2Ô9Ñ:Ô:ğ eñ eˆEØ$1 MĞ2IÔ2TÑ$UÔ$Uğ
eñ
eĞ İĞ'Ñ(Ô(¨AÒ-Ğ-Ø Ø(,×(;Ò(;Ğ<LĞNeÑ(fÔ(fÑ%��f˜gØ—
’
˜dŸiši™kœkÑ*Ô*Ğ*İ” ¨D¯IªI©K¬KÀ6Ç;Â;Á=Ä=Ğ]d×]iÒ]iÑ]kÔ]kĞlĞlÑmÔmĞmØ”×(Ò(Ñ*Ô*Ğ*Ø—
’
‘”�Ø”×#Ò#Ñ%Ô%Ğ%ØÔ!×7Ò7¸¸d¼nÔ>UÕW_ÕagÔanØ>NĞPT×PYÒPYÑP[ÔP[Ğ]bĞ=cğ 8ñ eô eğ eñ eñ
eğ
Ô×/Ò/°°d´nÔ6MÍxÕY_ÔYdĞlrĞktĞ/ÑuÔuĞuİ�6‰{Œ{�S ™[œ[Ñ(¨&Ğ0Ğ0r6 c ó — | j €Ct j | ¦ « |j |j ¬¦ « | _ | d¬¦ « t |j ¦ « D ]“} ||j
d¬¦ « D ]~}t |¦ « dk rŒ| ||¦ « \ }}}t j | ¦ « | ¦ « | ¦ « dœ¦ « ŒŒ”d S )Nrê T)Úis_evalr )Úeval_combined_lossÚ eval_loss_vÚ eval_loss_pi)rš rJ r r€ r� rw rï r‚ rƒ Ú eval_epochsr… r† rğ rñ rò r� )r3 rr rè r› rœ r¡ ró rŒ s r5 Úeval_netz OriginalAlphaZerNetwork.eval_net s( € Ø Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<ĞAXÔA[Ø8OÔ8Rğ +ñ Tô TˆDŒNà× Ò dĞ Ñ+Ô+Ğ+İĞ2Ô>Ñ?Ô?ğ wğ wˆEØ$1 MĞ2IÔ2TĞ^bĞ$cÑ$cÔ$cğ
wğ
wĞ İĞ'Ñ(Ô(¨AÒ-Ğ-Ø Ø(,×(;Ò(;Ğ<LĞNeÑ(fÔ(fÑ%��f˜gİ” Ø+/¯9ª9©;¬;ÀvÇ{Â{Á}Ä}Ğfm×frÒfrÑftÔftĞuĞuñwô wğ wğ wğ
wğ wğ wr6 c ó| — ddl m} t j t j ¦ « rdnd¦ « }t
|� \ }}}t j t j |¦ « t j
|¬¦ « }t j t j |¦ « t j
|¬¦ « }t j |t j
|¬¦ « d¦ « }| ||j
¬¦ « \ }} ||¦ « }
t | |¦ « } | |||
|¦ « } | | z }
|
| | fS )Nr ru rX rM rx r rH )r} rv rJ rz rX r~ r‡ rˆ r‰ rŠ r‹ r< rE r7 r rŒ )r3 rœ rè rv rz r� r. r/ r� rŸ r ró rŒ r¡ s r5 rğ z&OriginalAlphaZerNetwork.calculate_loss s € ØLĞLĞLĞLĞLĞLİ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİĞ -Ğ.‰
ˆ��Aİ”�2œ8 FÑ+Ô+µ2´:ÀfĞMÑMÔMˆİ
ŒY•r”x ‘|”|2¬:¸fĞ
EÑ
EÔ
Eˆİ ŒI�a�rœz°&Ğ 9Ñ 9Ô 9× CÒ CÀAÑ FÔ FˆØ Ÿ,š, vĞ6MÔ6T˜,ÑUÔU‰ˆ�Ø*Ğ*¨6Ñ2Ô2ˆİ˜& !Ñ$Ô$ˆØ—,’,˜w¨¨E°6Ñ:Ô:ˆØ˜Ñ ˆØ�V˜WĞ$Ğ$r6 )FFr³ rg )r” r´ rµ r¶ r· Úlistr r¸ r rE rJ r¹ rO rz rŒ ri Ú classmethodr rq Útupler“ rú rğ r¼ r½ s @r5 r¿ r¿ “ s3 ø€ € € € € ğ
,-¨a¨&Ø59ÈQĞ_dØ%*ğ #Cğ #C Cğ #C°sğ #CÀUğ #CĞY\ğ #CØ$(¨¤Iğ#CØ=@ğ#Cà&)ğ#Cà=@ğ#CàSVğ#Càqtğ#Cğ # 3œiğ#Cğ +Ğ 2¨dğ #Cğ HKğ #Cğ Y]ğ #Cğ
#ğ #Cğ #Cğ #Cğ #Cğ #Cğ #CğJ*ğ * ğ *¸tğ *ğ *ğ *ğ *ğ& €R„Z�\„\ğCğ Cğ Cñ „\ğCğ
7¨r¬yğ 7ğ 7ğ 7ğ 7ğ
uğ uğ uğ ğ mğ m fğ m¸Dğ mĞP[ĞPcĞ_cğ mğ mğ mñ „[ğ mğ1Àğ 1È5ĞQVĞX\Ğ]bÔXcĞQcÔKdğ 1ğ 1ğ 1ğ 1ğ, w¸vğ wÈ$ğ wğ wğ wğ wğ %ğ %ğ %ğ %ğ %ğ %ğ %r6 r¿ c ó. ‡ — e Zd Zdedefˆ fd„ Zd„ Zˆ xZS )rÍ r r c ó. •— t t | ¦ « ¦ « t j ||dd¬¦ « | _ t j |¦ « | _ t j ||dd¬¦ « | _ t j |¦ « | _ d S )Nr r r )
r rÍ r r r r r r r! r" )r3 r r r4 s €r5 r z OriginalAlphaZeroBlock.__init__ sz ø€ İ
Õ$ dÑ+Ô+×4Ò4Ñ6Ô6Ğ6İ”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒˆˆr6 c óü — |}t j | | |¦ « ¦ « ¦ « }| | |¦ « ¦ « }||z
}t j |¦ « S rg )r= r> r r r" r! )r3 rD Úx_skips r5 rE z OriginalAlphaZeroBlock.forward' s^ € ؈İ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆØ �HŠH�T—Z’Z ‘]”]Ñ #Ô #ˆØ ˆV‰ ˆİŒv�a‰yŒyĞr6 ©r” r´ rµ r¶ r rE r¼ r½ s @r5 rÍ rÍ sZ ø€ € € € € ğ0 Cğ 0°sğ 0ğ 0ğ 0ğ 0ğ 0ğ 0ğğ ğ ğ ğ ğ ğ r6 rÍ c ó> ‡ — e Zd Zdedededededef ˆ fd„ Zd„ Zˆ xZS ) rÔ r7 r r r Ú
num_layersÚlinear_hidden_sizec ó •— t t | ¦ « ¦ « t j |dd¦ « | _ t j d¦ « | _ t j |d¦ « | _ t dd||¦ « | _ |r<t j dd|z dz ¦ « | _ t j
d¬¦ « | _ d S t j dd¦ « | _ t j ¦ « | _ d S )Nr rT rÑ r: )r rÔ r r r Úconvr Úbnr' r( Ú
HeadLinearÚfcr+ Ú
LogSoftmaxÚactÚTanh)r3 r7 r r r r r r4 s €r5 r zValueHead.__init__0 sĞ ø€ å
�i˜Ñ Ô ×'Ò'Ñ)Ô)Ğ)İ”I˜l¨A¨qÑ1Ô1ˆŒ İ”. Ñ#Ô#ˆŒİ”9Ğ .°Ñ4Ô4ˆŒİ ˜S # zĞ3EÑFÔFˆŒØ ğ !İ”y a¨,Ñ&6¸Ñ&:Ñ;Ô;ˆDŒHİ”}¨Ğ+Ñ+Ô+ˆDŒHˆHˆHå”y aÑ(Ô(ˆDŒHİ”w‘y”yˆDŒHˆHˆHr6 c ó¸ — t j | | |¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j | |¦ « ¦ « }t j | |¦ « ¦ « }| |¦ « }| |¦ « S ©Nr r9 )
r= r> r r r? r@ r( r r+ r
©r3 rD s r5 rE zValueHead.forward> s• € İ
ŒF�4—7’7˜4Ÿ9š9 Q™<œ<Ñ(Ô(Ñ )Ô )ˆØ
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆİ
ŒF�4—8’8˜A‘;”;Ñ Ô ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ �HŠH�Q‰KŒKˆØ�xŠx˜‰{Œ{Ğr6 )r” r´ rµ r¸ r¶ r rE r¼ r½ s @r5 rÔ rÔ / sy ø€ € € € € ğ !˜tğ !¸ğ !È3ğ !Ğ^ağ !Ğorğ !Ø%(ğ !ğ !ğ !ğ !ğ !ğ !ğ ğ ğ ğ ğ ğ ğ r6 rÔ c
ó: ‡ — e Zd Zdededededef
ˆ fd„ Zd„ Zˆ xZS )rØ r Úlinear_input_size_policyr r r c óT •— t t | ¦ « ¦ « t j |dd¦ « | _ t j d¦ « | _ t j |d¦ « | _ t dd||¦ « | _ t j d|¦ « | _ d S )NrÑ r rT )
r rØ r r r r r r r' r( r
r r+ )r3 r r r r r r4 s €r5 r zPolicyHead.__init__H s† ø€ å
�j˜$Ñ Ô ×(Ò(Ñ*Ô*Ğ*İ”I˜l¨A¨qÑ1Ô1ˆŒ İ”. Ñ#Ô#ˆŒİ”9Ğ 5°sÑ;Ô;ˆŒİ ˜S # zĞ3EÑFÔFˆŒİ”9˜S +Ñ.Ô.ˆŒˆˆr6 c óº — t j | | |¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j | |¦ « ¦ « }t j | |¦ « ¦ « }| |¦ « }t j |d¬¦ « S )Nr r9 r r: )
r= r> r r r? r@ r( r r+ rA r s r5 rE zPolicyHead.forwardQ sš € İ
ŒF�4—7’7˜4Ÿ9š9 Q™<œ<Ñ(Ô(Ñ )Ô )ˆØ
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆİ
ŒF�4—8’8˜A‘;”;Ñ Ô ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ �HŠH�Q‰KŒKˆİŒ}˜Q AĞ&Ñ&Ô&Ğ&r6 r r½ s @r5 rØ rØ G sr ø€ € € € € ğ/ Cğ /À3ğ /ĞVYğ /Ğgjğ /Ø%(ğ/ğ /ğ /ğ /ğ /ğ /ğ'ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'r6 rØ c
óF ‡ — e Zd Zdededee dedef
ˆ fd„ Zd„ Zˆ xZS )rÖ r Ú out_channelsrÇ r r c óD •— t t | ¦ « ¦ « t j |dd¦ « | _ t j d¦ « | _ t ||||¦ « | _ t j
||d |d z |z ¦ « | _ d S )Né r r ) r rÖ r r r r r r r
r r' Úfc3)r3 r r rÇ r r r4 s €r5 r zStateHead.__init__[ s… ø€ å
�i˜Ñ Ô ×'Ò'Ñ)Ô)Ğ)İ”I˜l¨A¨qÑ1Ô1ˆŒ İ”. Ñ#Ô#ˆŒİ Ğ .° ¸jĞJ\Ñ]Ô]ˆŒİ”9˜\¨;°q¬>¸KȼNÑ+JÈ\Ñ+YÑZÔZˆŒˆˆr6 c óD — t j | | |¦ « ¦ « ¦ « }| | d¦ « d¦ « }t j | |¦ « ¦ « }| |¦ « }|S r )r= r> r r r? r@ r r r s r5 rE zStateHead.forwardc sq € İ
ŒF�4—7’7˜4Ÿ9š9 Q™<œ<Ñ(Ô(Ñ )Ô )ˆØ
�IŠI�a—f’f˜Q‘i”i Ñ $Ô $ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ �HŠH�Q‰KŒKˆØˆr6 )r” r´ rµ r¶ rü r rE r¼ r½ s @r5 rÖ rÖ Z s� ø€ € € € € ğ[¨#ğ [¸Sğ [ÈtĞTWÌyğ [Ğfiğ [Ø%(ğ[ğ [ğ [ğ [ğ [ğ [ğğ ğ ğ ğ ğ ğ r6 rÖ c ó6 ‡ — e Zd Zdedededefˆ fd„ Zd„ Zˆ xZS )r
r r r Ú hidden_sizec ó •‡— t t | ¦ « ¦ « t j |‰¦ « | _ t j ˆfd„t |¦ « D ¦ « � | _ t j ‰|¦ « | _ d S )Nc ó: •— g | ]}t j ‰‰¦ « ‘ŒS rÌ )r r' )rÎ rÏ r s €r5 rĞ z'HeadLinear.__init__.<locals>.<listcomp>o s% ø€ Ğ!aĞ!aĞ!aÈ!¥"¤)¨K¸Ñ"EÔ"EĞ!aĞ!aĞ!ar6 )
r r
r r r' r( Ú
Sequentialrƒ r r+ )r3 r r r r r4 s `€r5 r zHeadLinear.__init__l sv øø€ İ
�j˜$Ñ Ô ×(Ò(Ñ*Ô*Ğ*İ”9˜[¨+Ñ6Ô6ˆŒİ”-Ğ!aĞ!aĞ!aĞ!aÍuĞU_ÑO`ÔO`Ğ!aÑ!aÔ!aĞbˆŒİ”9˜[¨,Ñ7Ô7ˆŒˆˆr6 c óÈ — t j | |¦ « ¦ « }t j | |¦ « ¦ « }| |¦ « S rg )r= r> r( r r+ r s r5 rE zHeadLinear.forwardr sD € İ
ŒF�4—8’8˜A‘;”;Ñ Ô ˆİ
ŒF�4—7’7˜1‘:”:Ñ Ô ˆØ�xŠx˜‰{Œ{Ğr6 r r½ s @r5 r
r
k si ø€ € € € € ğ8 Cğ 8°sğ 8Èğ 8ĞZ]ğ 8ğ 8ğ 8ğ 8ğ 8ğ 8ğ ğ ğ ğ ğ ğ ğ r6 r
)%r0 rb rc rN r‰ ÚtorchrJ Útorch.nnr Útorch.nn.functionalÚ
functionalr= rñ r Útorch.nn.modules.moduler Ú mu_alpha_zero.General.networkr Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr Úmu_alpha_zero.configr r Úmu_alpha_zero.mem_bufferr
Úmu_alpha_zero.MuZero.utilsr ÚModuler
r¿ rÍ rÔ rØ rÖ r
rÌ r6 r5 ú<module>r. sj ğØ
€
€
€
Ø € € € Ø € € € à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ø € € € Ø (Ğ (Ğ (Ğ (Ğ (Ğ (Ø %Ğ %Ğ %Ğ %Ğ %Ğ %à @Ğ @Ğ @Ğ @Ğ @Ğ @Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø .Ğ .Ğ .Ğ .Ğ .Ğ .Ø @Ğ @Ğ @Ğ @Ğ @Ğ @ğ{oğ {oğ {oğ {oğ {o�2”9Ğ 4ñ {oô {oğ {oğ|I%ğ I%ğ I%ğ I%ğ I%˜bœiĞ)?ñ I%ô I%ğ I%ğX
ğ
ğ
ğ
ğ
˜RœUœ\ñ
ô
ğ
ğ ğ ğ ğ ğ �”” ñ ô ğ ğ0'ğ 'ğ 'ğ 'ğ '�”” ñ 'ô 'ğ 'ğ&ğ ğ ğ ğ �”” ñ ô ğ ğ"
ğ
ğ
ğ
ğ
�”ñ
ô
ğ
ğ
ğ
r6 | 31,519 | Python | .pyt | 95 | 330.715789 | 1,929 | 0.273636 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,570 | trainer.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/AlphaZero/Network/__pycache__/trainer.cpython-311.pyc | §
°#Ûeu> ã óJ — d dl mZ d dlmZ edd¬¦ « d dlmZ d dlmZ d dlZ d d l
m Z d d
l m Z d d l
mZ d d lmZmZmZ d d
lmZ d dlmZ d dlmZ d dlmZmZ d dl m Z d dl m Z d dl m!Z! d dl"m#Z# d dl$m%Z% d dl&m'Z' d dl(m)Z) d dl*m+Z+ d dl,m-Z- G d„ d¦ « Z.dS ) é )Úset_start_method)Únot_zeroÚspawnT)Úforce)Údeepcopy)ÚTypeN)Ú
SummaryWriter)Útqdm)ÚArena)Ú RandomPlayerÚPlayerÚ NetPlayer)Ú McSearchTree)Ú AlphaZeroNet)Ú CheckPointer)ÚLoggingMessageTemplatesÚLogger)Úbuild_all_from_config)Ú GeneralArena)Ú
AlphaZeroGame)ÚGeneralMemoryBuffer)ÚGeneralNetwork)Ú
SearchTree)Ú JavaManager)ÚConfig)Ú MemBufferc ó" — e Zd Z d0dededej dedede d e
d
e d e d ej
j pdd
epddeddfd„Ze d1dee dee
dee dededed e de d
epdfd„¦ « Ze d2dededed e
d
e d e de d
epddepddepdfd„¦ « Ze d1dededed e
d e de depdfd „¦ « Zdefd „Zd edee fd „Z d edee fd „Z d!„ Z d"„ Z d3d#„Z!d$„ Z"d%„ Z#d&„ Z$d'ed(efd)„Z%d'ed(ed*efd+„Z&d,ed-ed.e'd*efd/„Z(dS )4ÚTrainerTNÚnetworkÚgameÚ optimizerÚmemoryÚmuzero_alphazero_configÚ checkpointerÚ search_treeÚ
net_playerÚheadlessÚopponent_network_overrideÚarena_overrideÚ java_managerÚreturnc óÎ — || _ | | _ |
| _ || _ || _ || _ || _ | €| j ¦ « n| | _ || _ || _
|
| _ t d¦ « | _
| € t | j | j | j ¦ « n| | _ || _ t# | j j | j j ¬¦ « | _ g | _ g | _ d S )NzLogs/AlphaZero)ÚlogdirÚtoken)r# Údevicer' Ú game_managerÚmctsr& r Úmake_fresh_instanceÚopponent_networkr! r" r* r Úsummary_writerr Úarenar$ r Úlog_dirÚpushbullet_tokenÚloggerÚarena_win_frequenciesÚlosses)Úselfr r r! r" r# r$ r% r& r/ r' r( r) r* s úV/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/AlphaZero/Network/trainer.pyÚ__init__zTrainer.__init__" sû € ğ (?ˆÔ$Ø ˆŒ Ø ˆŒ
Ø ˆÔØ ˆŒ Ø$ˆŒØ ˆŒ ØF_ĞFg ¤ × @Ò @Ñ BÔ BĞ Bğ nGˆÔ Ø"ˆŒØ ˆŒ Ø (ˆÔİ +Ğ,<Ñ =Ô =ˆÔà+9Ğ+Aõ ˜4Ô,¨dÔ.JØ œ;ñ(ô (ğ (ØGUğ
Œ
à (ˆÔİ DÔ$@Ô$HØ#'Ô#?Ô#PğRñ Rô RˆŒ à%'ˆÔ"؈Œ ˆ ˆ ó FÚ net_classÚ
tree_classÚnet_player_classÚcheckpoint_pathÚcheckpoint_dirÚcheckpointer_verbosec
óö — t j t j ¦ « rdnd¦ « }
t ||¬¦ « } | |¦ « \ } }
}}}}t
j |¦ « } || ¦ « |¦ « }| d j d |_
| |¦ « }| ¦ « }t j
| ¦ « |¬¦ « } || ¦ « fi ||dœ¤�}| | ¦ « | |¦ « | |
¦ « | |||||| |||
|| ¬¦ « S ) NÚcudaÚcpu©Úverbosez
fc1.weighté )Úlr©r Úmonte_carlo_tree_search)r' r) )Úthr/ rF Ú is_availabler Úload_checkpoint_from_pathr Ú from_argsr2 ÚshapeÚaz_net_linear_input_sizeÚmake_from_configÚoptimÚAdamÚ
parametersÚload_state_dict)Úclsr? r@ rA rB rC r r' rD r) r/ r$ Ú network_dictÚoptimizer_dictr" rK ÚargsÚ
opponent_dictÚconfÚtreer r3 r! r& s r< Úfrom_checkpointzTrainer.from_checkpoint> s– € õ ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ# NĞ<PĞQÑQÔQˆ àHT×HnÒHnØ ñI ô I ÑEˆ �n f¨b°$¸
åÔ Ñ%Ô%ˆØˆz˜$×2Ò2Ñ4Ô4°dÑ;Ô;ˆØ(4°\Ô(BÔ(HÈÔ(KˆÔ%Ø×,Ò,¨TÑ2Ô2ˆØ"×6Ò6Ñ8Ô8Ğİ”H—M’M '×"4Ò"4Ñ"6Ô"6¸2�MÑ>Ô>ˆ à%Ğ% d×&>Ò&>Ñ&@Ô&@ğ _ğ _Ø4;ĞX\Ğ(]Ğ(]ğ_ğ _ˆ
à× Ò Ñ-Ô-Ğ-Ø×(Ò(¨ Ñ7Ô7Ğ7Ø×!Ò! .Ñ1Ô1Ğ1؈s�7˜D )¨V°T¸<ÈÈzĞ[aĞltØ"0ğ2ñ 2ô 2ğ 2r> Úmemory_overridec óø — t j t j ¦ « rdnd¦ « } t || ¦ « \ } }
}| €|n| }t |j |¬¦ « } | |||
|||||| |||
¬¦ « S )NrF rG rH )r' r) r* )rN r/ rF rO r r rC )rY r# r r r% r& r' rD r) ra r* r/ Ú_r! Úmemr" r$ s r< ÚcreatezTrainer.createY s¡ € õ ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ 1Ğ2IÈ6Ñ RÔ Rшˆ9�cØ'Ğ/��°_ˆİ#Ğ$;Ô$JĞThĞiÑiÔiˆ ؈s�7˜D )¨VĞ5LÈlĞ\gĞisØØ $°^ĞR^ğ`ñ `ô `ğ `r> Úpathc
ó| — t j t j ¦ « rdnd¦ « }t ||¦ « \ } }
} t |j |¬¦ « } | t j |¦ « ¦ « t |
¦ « fi | |dœ¤�}
| | ||
| || ||
|||¬¦ « S )NrF rG rH rL )r' r* ) rN r/ rF rO r r rC rX Úloadr r2 )rY rf r# r r% r' rD r* r/ Únetr! r" r$ r& s r< Úfrom_state_dictzTrainer.from_state_dictj sÔ € õ ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ!6Ğ7NĞPVÑ!WÔ!WÑ ˆˆY˜İ#Ğ$;Ô$JĞThĞiÑiÔiˆ Ø ×Ò�BœG D™MœMÑ*Ô*Ğ*İ ˜t× 7Ò 7Ñ 9Ô 9ĞvĞvÈĞitĞ=uĞ=uĞvĞvˆ
؈s�3˜˜i¨Ğ1HÈ,ĞXcĞeoĞqwØ $Ø ,ğ.ñ .ô .ğ .r> c óš — | j | j ¦ « | j t j | j j ¦ « ¦ « | j | j
¦ « ¦ « | j j }| j j }| j j
}| j j }| j
¦ « | t# |¦ « dd¦ « D �] }|| j j k r d| j _ t) j ¦ « 5 | j j }| j t j |¦ « ¦ « | j j dk r˜| j | |¦ « | |¦ « | j | j ||¦ « \ }}} t; | j t< ¦ « r*| j j r | j ¦ « | _ n0| j ! | j
| j || j ¦ « \ }}} | j t j" ||| tF ¦ « ¦ « | j $ d¦ « d d d ¦ « n # 1 swxY w Y | j% & | j
¦ « | j t j' d| j% ( ¦ « ¦ « ¦ « | j% ) | j ¦ « | j t j* d| j% ( ¦ « ¦ « ¦ « | j
+ ¦ « | j t j, |¦ « ¦ « | j
- | j | j ¦ « \ }
} | j t j. |
¦ « ¦ « | j $ d|
› dt_ | ¦ « › d ta | ¦ « › �¦ « | j1 2 | ¦ « | j% 3 | j1 ¦ « | j% 4 | j
| j | j5 | j j6 || j d
¬ ¦ « | j j7 } | j j8 }
| 9 || ¦ « \ }}}}| : |||
|¦ « | ; || |¦ « �Œ | j j | j j | j j< | j j8 | j j
| j j= | j j> | j j7 d œ}| j t j? |¦ « ¦ « | j $ t j@ ¦ « ¦ « | j
S )
NzTraining Progressr rJ zFinished self-play.ztemp checkpointzopponent networkz Mean loss: z , Max loss: z , Min loss: Úlatest_trained_net)Úname)ÚnumItersÚnumSelfPlayGamesÚtauÚupdateThresholdÚ
mcSimulationsÚcÚmaxDepthÚ numPitGames)Ar3 Útor/ r8 Úlogr ÚTRAINING_STARTr# Ú num_itersrX r Ú
state_dictÚepochsÚnum_simulationsÚself_play_gamesÚevalÚ
make_tqdm_barÚrangeÚzero_tau_afterÚ arena_taurN Úno_gradÚ num_workersÚSELF_PLAY_STARTr1 Úparallel_self_playÚmake_n_networksÚ make_n_treesr" Ú
isinstancer Úis_diskr2 Ú self_playÚ
SELF_PLAY_ENDr Úpushbullet_logr$ Úsave_temp_net_checkpointÚSAVEDÚ
get_temp_pathÚload_temp_net_checkpointÚLOADEDÚtrainÚNETWORK_TRAINING_STARTÚ train_netÚNETWORK_TRAINING_ENDÚmaxÚminr: ÚextendÚ save_lossesÚsave_checkpointr! rK Ú
num_pit_gamesÚupdate_thresholdÚ run_pittingÚ check_modelÚrun_pitting_randomrp rs Ú max_depthÚ TRAINING_ENDÚTRAINING_END_PSB)r; ry r{ r| r} ÚiÚn_jobsÚwins_p1Úwins_p2Ú
game_drawsÚ mean_lossr: Ú num_gamesr� Úp1_winsÚp2_winsÚdrawsÚ
wins_totalÚimportant_argss r< r“ z
Trainer.trainw sn € Ø Ô × Ò ¤Ñ-Ô-Ğ-Ø Œ �ŠÕ/Ô>¸tÔ?[Ô?eÑfÔfÑgÔgĞgØ Ô ×-Ò-¨d¬l×.EÒ.EÑ.GÔ.GÑHÔHĞHàÔ0Ô:ˆ ØÔ-Ô4ˆØ Ô6ÔFˆØ Ô6ÔFˆØ Œ ×ÒÑÔĞà×#Ò#¥E¨)Ñ$4Ô$4Ğ6IÈ1ÑMÔMğ ( Cñ ( CˆAØ�DÔ0Ô?Ò?Ğ?Ø9:�Ô,Ô6İ”‘ ” ğ
Bğ
BØ Ô5ÔA�Ø” —’Õ 7Ô GÈÑ XÔ XÑYÔYĞYàÔ/Ô;¸aÒ?Ğ?Ø37´9×3OÒ3OĞPT×PdÒPdĞekÑPlÔPlØPT×PaÒPaĞbhÑPiÔPiØPTÔP[Ğ]aÔ]hØP_ØPVñ 4Xô 4XÑ0�G˜W jõ
" $¤+yÑ9Ô9ğ H¸d¼kÔ>Qğ HØ&*¤k×&EÒ&EÑ&GÔ&G˜œ øà37´9×3FÒ3FÀtÄ|ĞUYÔU`ĞbqØGKÄ{ñ4Tô 4TÑ0�G˜W jà” —’Õ 7Ô EÀgÈwĞXbÕdlÑ mÔ mÑnÔnĞnØ” ×*Ò*Ğ+@ÑAÔAĞAğ!
Bğ
Bğ
Bñ
Bô
Bğ
Bğ
Bğ
Bğ
Bğ
Bğ
Bøøøğ
Bğ
Bğ
Bğ
Bğ$
Ô × 6Ò 6°t´|Ñ DÔ DĞ DØ ŒK�OŠOÕ 3Ô 9Ğ:KÈTÔM^×MlÒMlÑMnÔMnÑ oÔ oÑ pÔ pĞ pØ Ô × 6Ò 6°tÔ7LÑ MÔ MĞ MØ ŒK�OŠOÕ 3Ô :Ğ;MÈtÔO`×OnÒOnÑOpÔOpÑ qÔ qÑ rÔ rĞ rØ ŒL× Ò Ñ Ô Ğ Ø ŒK�OŠOÕ 3Ô JÈ6Ñ RÔ RÑ SÔ SĞ SØ $¤ × 6Ò 6°t´{ÀDÔD`Ñ aÔ aÑ ˆI�vØ ŒK�OŠOÕ 3Ô HÈÑ SÔ SÑ TÔ TĞ TØ ŒK× &Ò &Ğ'r°YĞ'rĞ'rÍCĞPVÉKÌKĞ'rĞ'rÕehĞioÑepÔepĞ'rĞ'rÑ sÔ sĞ sØ ŒK× Ò ˜vÑ &Ô &Ğ &Ø Ô × )Ò )¨$¬+Ñ 6Ô 6Ğ 6Ø Ô × -Ò -¨d¬l¸DÔ<QĞSWÔSaØ.2Ô.JÔ.MÈqĞRVÔRnØ3Gğ
.ñ
Iô
Iğ
Iğ Ô4ÔBˆIØ #Ô ;Ô LĞ Ø26×2BÒ2BÀ?ĞT]Ñ2^Ô2^Ñ /ˆG�W˜e ZØ × Ò ˜W jĞ2BÀAÑ FÔ FĞ FØ × #Ò # O°YÀÑ BÔ BĞ BÑ Bğ Ô4Ô>Ø $Ô <Ô LØÔ/Ô3Ø #Ô ;Ô LØ !Ô 9Ô IØÔ-Ô/Ø Ô4Ô>Ø Ô7ÔEğ
ğ
ˆğ
Œ �ŠÕ/Ô<¸^ÑLÔLÑMÔMĞMØ Œ ×"Ò"Õ#:Ô#KÑ#MÔ#MÑNÔNĞNØŒ|Ğs Ä&E JÊJ ÊJ Únc ó: ‡ — ˆ fd„t |¦ « D ¦ « S )z«
Make n identical copies of self.network using deepcopy.
:param n: The number of copies to make.
:return: A list of n identical networks.
c ó8 •— g | ]}t ‰j ¦ « ‘ŒS © )r r )Ú.0rc r; s €r< ú
<listcomp>z+Trainer.make_n_networks.<locals>.<listcomp> s# ø€ Ğ9Ğ9Ğ9¨1•˜œ Ñ&Ô&Ğ9Ğ9Ğ9r> )r€ )r; r° s ` r< r‡ zTrainer.make_n_networks» s% ø€ ğ :Ğ9Ğ9Ğ9µ°a±´Ğ9Ñ9Ô9Ğ9r> c óŠ — g }t |¦ « D ]0}| j ¦ « }| |¦ « Œ1|S )zŠ
Make n new search trees.
:param n: The number of trees to create.
:return: A list of n new search trees.
)r€ r1 r2 Úappend)r; r° Útreesr¤ r_ s r< rˆ zTrainer.make_n_treesÄ sK € ğ ˆİ�q‘”ğ ğ ˆAà”9×0Ò0Ñ2Ô2ˆDØ �LŠL˜Ñ Ô Ğ Ğ Øˆ r> c ón — t | j ¦ « t t | j ¦ « ¦ « z S ©N)Úsumr9 r Úlen©r; s r< Ú get_arena_win_frequencies_meanz&Trainer.get_arena_win_frequencies_meanÑ s+ € İ�4Ô-Ñ.Ô.µ½#¸dÔ>XÑ:YÔ:YÑ1ZÔ1ZÑZĞZr> c ód — | j ¦ « }| j ¦ « }| j ¦ « }t j || j | j j ||| j ¦ « dœ|¦ « t d |¦ « ¦ « d S )N)r! r" rK ri Úopponent_state_dictr\ z Saved latest model data to {}) r rz r3 r! rN Úsaver" r# rK Úto_dictÚprintÚformat)r; rf rz rÀ Úoptimizer_state_dicts r< Ú save_latestzTrainer.save_latestÔ s¯ € Ø”\×,Ò,Ñ.Ô.ˆ
Ø "Ô 3× >Ò >Ñ @Ô @ĞØ #œ~× 8Ò 8Ñ :Ô :Ğ İ
ŒØ-Ø”kØÔ.Ô1Ø Ø#6ØÔ0×8Ò8Ñ:Ô:ğ
ğ
ğ ñ ô ğ õ Ğ-×4Ò4°TÑ:Ô:Ñ;Ô;Ğ;Ğ;Ğ;r> c óD — | j j rt ||||¬¦ « S |S )N)ÚdescÚpositionÚleave)r# Ú show_tqdmr
)r; ÚiterablerÈ rÉ rÊ s r< r zTrainer.make_tqdm_barâ s, € Ø Ô 'Ô 1ğ ݘ t°hÀeĞLÑLÔLĞ LàˆOr> c ó — | j S rº )r r½ s r< Ú get_networkzTrainer.get_networkè s
€ ØŒ|Ğr> c ó — | j S rº )r1 r½ s r< Úget_treezTrainer.get_treeë s
€ ØŒyĞr> c ó — | j S rº )r# r½ s r< Úget_argszTrainer.get_argsî s € ØÔ+Ğ+r> r| rª c
óÈ — | j ¦ « | j ¦ « | j ¦ « }| | j ¦ « | j ¦ « }| | j ¦ « | j t j |j
|j
|¦ « ¦ « | j ||||d¬¦ « \ }}}t ||z ¦ « }| j t j |j
|j
||||¦ « ¦ « | j ||z ¦ « ||||fS )NF)Únum_mc_simulationsÚ
one_player)r r~ r3 r& r2 Ú set_networkr8 rw r Ú
PITTING_STARTrm r5 Úpitr Ú PITTING_ENDr9 r· ) r; r| rª Úp1Úp2r« r¬ r r® s r< r� zTrainer.run_pittingñ sN € Ø Œ ×ÒÑÔĞØ Ô ×"Ò"Ñ$Ô$Ğ$Ø
Œ_×
0Ò
0Ñ
2Ô
2ˆØ
�Š�t”|Ñ$Ô$Ğ$Ø
Œ_×
0Ò
0Ñ
2Ô
2ˆØ
�Š�tÔ,Ñ-Ô-Ğ-Ø Œ �ŠÕ/Ô=¸b¼gÀrÄwĞPYÑZÔZÑ[Ô[Ğ[Ø"&¤*§.¢.°°R¸ĞWfØ<Ağ #1ñ #Cô #CÑ ˆ�˜%å ˜g¨Ñ /Ñ0Ô0ˆ
Ø Œ �ŠÕ/Ô;¸B¼GÀRÄWÈgØ<CÀZĞQVñXô Xñ Yô Yğ Yà Ô"×)Ò)¨'°IÑ*=Ñ>Ô>Ğ>ؘ ¨
Ğ2Ğ2r> r¤ c ó¶ — || j j z dk rd S | j ¦ « t j ¦ « 5 t
| j ¦ « fi i ¤�}| j ¦ « }|
| j ¦ « | j t j |j |j |¦ « ¦ « | j ||||¬¦ « \ }}}t% ||z ¦ « } | j t j |j |j ||| |¦ « ¦ « d d d ¦ « d S # 1 swxY w Y d S )Nr )rÔ )r# Úrandom_pit_freqr r~ rN rƒ r r0 r2 r& rÖ r8 rw r r× rm r5 rØ r rÙ )
r; r| rª r¤ Ú
random_playerrÚ Úp1_wins_randomÚp2_wins_randomÚ draws_randomr® s
r< r zTrainer.run_pitting_random s¬ € Ø ˆtÔ+Ô;Ñ ;¸qÒ @Ğ @Ø ˆFØ Œ ×ÒÑÔĞİ
ŒZ‰\Œ\ğ _ğ _İ (¨Ô):×)NÒ)NÑ)PÔ)PĞ WĞ WĞTVĞ WĞ WˆMà” ×4Ò4Ñ6Ô6ˆBØ �NŠN˜4œ<Ñ (Ô (Ğ (Ø ŒK�OŠOÕ 3Ô AÀ"Ä'È=ÔK]Ğ_hÑ iÔ iÑ jÔ jĞ jØ;?¼:¿>º>È"ÈmĞ]fØ]lğ <Jñ <nô <nÑ 8ˆN˜N¨Lå! .°>Ñ"AÑBÔBˆJØ ŒK�OŠOİ'Ô3°B´G¸]Ô=OĞQ_Ø4BÀJĞP\ñ^ô ^ñ
_ô
_ğ
_ğ _ğ _ğ _ñ _ô _ğ _ğ _ğ _ğ _ğ _ğ _ğ _øøøğ _ğ _ğ _ğ _ğ _ğ _s ÁC?EÅEÅEr« r® r� c óú — ||z |k r²| j t j ||z |¦ « ¦ « | j | j | j | j | j j
|| j ¦ « | j t j d| j ¦ « ¦ « ¦ « n“| j t j
||z |¦ « ¦ « | j | j ¦ « | j t j d| j ¦ « ¦ « ¦ « | j t j |¦ « ¦ « d S )Nzaccepted model checkpointzprevious version checkpoint)r8 rw r Ú MODEL_ACCEPTr$ r› r r3 r! r# rK r� Úget_checkpoint_dirÚ MODEL_REJECTr‘ r’ r� r� ÚITER_FINISHED_PSB)r; r« r® r� r¤ s r< rŸ zTrainer.check_model s� € Ø �ZÑ Ğ"2Ò 2Ğ 2Ø ŒK�OŠOÕ 3Ô @ÀÈ:ÑAUØAQñ Sô Sñ
Tô
Tğ
Tà Ô × -Ò -¨d¬l¸DÔ<QĞSWÔSaØ.2Ô.JÔ.MÈqØ.2Ô.Jñ
Lô
Lğ
Lğ
ŒK�OŠOÕ 3Ô 9Ğ:UØ:>Ô:K×:^Ò:^Ñ:`Ô:`ñ bô bñ
cô
cğ
cğ
cğ
ŒK�OŠOÕ 3Ô @ÀÈ:ÑAUØAQñ Sô Sñ
Tô
Tğ
Tà Ô × 6Ò 6°t´|Ñ DÔ DĞ DØ ŒK�OŠOÕ 3Ô :Ğ;XØ;?Ô;L×;ZÒ;ZÑ;\Ô;\ñ ^ô ^ñ
_ô
_ğ
_à Œ ×"Ò"Õ#:Ô#LÈQÑ#OÔ#OÑPÔPĞPĞPĞPr> )TNNN)TFN)TFNNN)T))Ú__name__Ú
__module__Ú __qualname__r r rN rU r r r r r
ÚboolÚnnÚModuler r r= Ú classmethodr Ústrr` re rj r r“ ÚintÚlistr‡ r rˆ r¾ rÆ r rÎ rĞ rÒ r� r ÚfloatrŸ r³ r> r< r r ! sÁ € € € € € ğ
+/ØCGØ8<Ø-1ğğ ğ °mğ Ø œHğØ.Ağà*0ğà@Lğğ )ğğ 7=ğğ $(ğ ğ
-/¬E¬LĞ,@¸Dğ ğ ".Ğ!5°ğ
ğ +ğğ 7;ğğ ğ ğ ğ8 ğ ?CØ5:Ø?Cğ 2ğ 2¨¨^Ô(<ğ 2È$ÈzÔJZğ 2Ø*.¨v¬,ğ2à),ğ2à>Ağ2ğ ,ğ2ğ 8<ğ2ğ /3ğ 2ğ
)5Ğ(<¸ğ 2ğ 2ğ 2ñ „[ğ2ğ4 ğ !%Ø,1Ø6:Ø>BØ37ğ`ğ `¨Vğ `¸=ğ `ĞSağ `Ø &ğ`à!ğ`ğ ğ`ğ &*ğ `ğ
,Ğ 3¨tğ `ğ !4Ğ ;°tğ
`ğ )Ğ 0¨Dğ`ğ `ğ `ñ „[ğ`ğ à)-Ø`dğ
.ğ
. 3ğ
.Àğ
.È}ğ
.Ğkuğ
.Ø"&ğ
.à.2ğ
.àJUĞJ]ĞY]ğ
.ğ
.ğ
.ñ „[ğ
.ğB �|ğ B ğ B ğ B ğ B ğH: ğ :¨¨lÔ);ğ :ğ :ğ :ğ :ğ ˜cğ d¨<Ô&8ğ ğ ğ ğ ğ[ğ [ğ [ğ <ğ <ğ <ğ ğ ğ ğ ğ ğ ğ ğğ ğ ğ,ğ ,ğ ,ğ3¨3ğ 3¸3ğ 3ğ 3ğ 3ğ 3ğ _°#ğ _À#ğ _È#ğ _ğ _ğ _ğ _ğ"Q 3ğ Q°Cğ QÈ5ğ QĞUXğ Qğ Qğ Qğ Qğ Qğ Qr> r )/Útorch.multiprocessingr Úmu_alpha_zero.General.utilsr Úcopyr Útypingr ÚtorchrN Útorch.utils.tensorboardr r
Ú#mu_alpha_zero.AlphaZero.Arena.arenar Ú%mu_alpha_zero.AlphaZero.Arena.playersr r
r Ú+mu_alpha_zero.AlphaZero.MCTS.az_search_treer Ú$mu_alpha_zero.AlphaZero.Network.nnetr Ú$mu_alpha_zero.AlphaZero.checkpointerr Ú mu_alpha_zero.AlphaZero.loggerr r Ú mu_alpha_zero.AlphaZero.utilsr Úmu_alpha_zero.General.arenar Ú mu_alpha_zero.General.az_gamer Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.networkr Ú!mu_alpha_zero.General.search_treer Ú-mu_alpha_zero.MuZero.JavaGateway.java_managerr Úmu_alpha_zero.configr Úmu_alpha_zero.mem_bufferr r r³ r> r< ú<module>r sù ğØ 2Ğ 2Ğ 2Ğ 2Ğ 2Ğ 2à 0Ğ 0Ğ 0Ğ 0Ğ 0Ğ 0à Ğ � Ğ %Ñ %Ô %Ğ %Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø Ğ Ğ Ğ Ğ Ğ Ø 5Ğ 5Ğ 5Ğ 5Ğ 5Ğ 5Ø PĞ PĞ PĞ PĞ PĞ PĞ PĞ PĞ PĞ PØ DĞ DĞ DĞ DĞ DĞ DØ =Ğ =Ğ =Ğ =Ğ =Ğ =Ø =Ğ =Ğ =Ğ =Ğ =Ğ =Ø JĞ JĞ JĞ JĞ JĞ JĞ JĞ JØ ?Ğ ?Ğ ?Ğ ?Ğ ?Ğ ?Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 7Ğ 7Ğ 7Ğ 7Ğ 7Ğ 7Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø EĞ EĞ EĞ EĞ EĞ EØ 'Ğ 'Ğ 'Ğ 'Ğ 'Ğ 'Ø .Ğ .Ğ .Ğ .Ğ .Ğ .ğ @Qğ @Qğ @Qğ @Qğ @Qñ @Qô @Qğ @Qğ @Qğ @Qr> | 21,360 | Python | .pyt | 95 | 223.126316 | 1,279 | 0.33802 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,571 | tictactoe_game.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/__pycache__/tictactoe_game.cpython-311.pyc | §
Uxf"D ã ó~ — d dl Z d dlZd dlZd dlmZ d dlZd dlZ d dl
Z d dl Z
d dlmZ d dlmZ G d„ de¦ « ZdS )é N)ÚImage)Ú
AlphaZeroGamec óŞ — e Zd ZdZd?dededdfd„Zdededdfd „Zd
edz defd „Z d „ Z
d
e j de
fd„Zd„ Zde j dede
fd„Zd@de j dededz fd„Zde j defd„Zde j fd„Zd?dedefd„Zd?dedefd„Zd?dededefd „Zd?dededefd „Zd?dededeeef ez fd „ZdAded edefd!„Z d"„ Z ded#edefd$„Z ded#edefd%„Z d?defd&„Z d'„ Z!d(„ Z"d)„ Z#defd*„Z$dBd+„Z%dBd,„Z&d-„ Z'dedefd.„Z(d?d/e j depdde
fd0„Z)defd1„Z*d2„ Z+de j fd3„Z,d4„ Z-d5„ Z.d6„ Z/e0de j fd7„¦ « Z1de j d8epedede j fd9„Z2d:efd;„Z3d<e j defd=„Z4d>„ Z5dS )CÚTicTacToeGameManagerzV
This class is the game manager for the game of Tic Tac Toe and its variants.
NÚ
board_sizeÚheadlessÚreturnc óØ — d| _ d| _ || _ | ¦ « | _ || _ | |¦ « | _ | |¦ « | _ d S )Né éÿÿÿÿ)
ÚplayerÚ enemy_playerr Úinitialize_boardÚboardr Úinit_num_to_winÚ
num_to_winÚ
create_screenÚscreen)Úselfr r r s úP/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Game/tictactoe_game.pyÚ__init__z TicTacToeGameManager.__init__ s` € àˆŒ Ø ˆÔØ$ˆŒØ×*Ò*Ñ,Ô,ˆŒ
Ø ˆŒ
Ø ×.Ò.¨zÑ:Ô:ˆŒØ×(Ò(¨Ñ2Ô2ˆŒ ˆ ˆ ó r
Úindexc ó — || j |< d S ©N©r )r r
r s r ÚplayzTicTacToeGameManager.play s € Ø "ˆŒ
�5ÑĞĞr r c óL — |€| j }|| j k rt d¦ « ‚|S )Nz+Num to win can't be greater than board size)r Ú Exception)r r s r r z$TicTacToeGameManager.init_num_to_win! s1 € Ø Ğ Ø œ ˆJØ ˜œÒ 'Ğ 'İĞ IÑJÔJĞ JØĞr c ó^ — t j | j | j ft j ¬¦ « }|S )N©Údtype)ÚnpÚzerosr Úint8©r r s r r z%TicTacToeGameManager.initialize_board( s& € İ”˜$œ/¨4¬?Ğ;Å2Ä7ĞKÑKÔKˆØˆ r Ú observationsc ó˜ — | |¦ « }t |¦ « dk rt d¦ « ‚t j |¦ « S )Nr zNo valid moves)Úget_valid_movesÚlenr ÚrandomÚchoice)r r' ÚkwargsÚ valid_movess r Úget_random_valid_actionz,TicTacToeGameManager.get_random_valid_action, sG € Ø×*Ò*¨<Ñ8Ô8ˆ İ ˆ{Ñ Ô ˜qÒ Ğ İĞ ,Ñ-Ô-Ğ -İŒ}˜[Ñ)Ô)Ğ)r c óÈ — |rd S t j ¦ « t j d¦ « | j dz }t j ||f¦ « }|S )Nz Tic Tac Toeéd )ÚpgÚinitÚdisplayÚ set_captionr Úset_mode)r r Úboard_rect_sizer s r r z"TicTacToeGameManager.create_screen2 s[ € Ø ğ Ø ˆFå
Œ‰ Œ ˆ İ
Œ
× Ò ˜}Ñ-Ô-Ğ-Ø œ/¨CÑ/ˆİ”×$Ò$ o°Ğ%GÑHÔHˆØˆ
r ÚarrÚfuncc ó^ ‡— g }| d¦ « ‰D ]3}| || d¦ « ¦ « ¦ « Œ4| d¦ « ‰j D ]3}| || d¦ « ¦ « ¦ « Œ4ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }| |¦ « t
|¦ « D ]}\ }} |dk r| d¦ « n+|t |¦ « d z k r| d
¦ « | || d¦ « ¦ « ¦ « Œ~|S ) an
This function iterates over all rows, columns and the two main diagonals of supplied array,
applies the supplied function to each of them and returns the results in a list.
:param arr: a 2D numpy array.
:param func: a callable function that takes a 1D array as input and returns a result.
:return: A list of results.
Úrowr Úcolc ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS ©)Úk©r# Údiag©Ú.0Úir8 s €r ú
<listcomp>z;TicTacToeGameManager.full_iterate_array.<locals>.<listcomp>M ó( ø€ ĞSĞSĞS q•”˜ Ğ"Ñ"Ô"ĞSĞSĞSr r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> ©r# rA ÚfliplrrB s €r rE z;TicTacToeGameManager.full_iterate_array.<locals>.<listcomp>N ó0 ø€ ĞfĞfĞf¸!�œ¥¤¨3¡ ¤ °1Ğ5Ñ5Ô5ĞfĞfĞfr Ú diag_lefté Ú
diag_right)ÚappendÚreshapeÚTÚrangeÚshapeÚextendÚ enumerater* )
r r8 r9 Úresultsr; r< ÚdiagsÚ
flipped_diagsÚidxrA s
` r Úfull_iterate_arrayz'TicTacToeGameManager.full_iterate_array< s¶ ø€ ğ ˆØ�Š�uÑ Ô Ğ Øğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1à�Š�uÑ Ô Ğ Ø”5ğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1àSĞSĞSĞSE°3´9¸Q´<°-À!Ñ2CÀSÄYÈqÄ\Ñ,RÔ,RĞSÑSÔSˆØfĞfĞfĞf½uÀcÄiĞPQÄlÀ]ĞUVÑEVĞX[ÔXaĞbcÔXdÑ?eÔ?eĞfÑfÔfˆ
Ø
� Š �]Ñ#Ô#Ğ#İ" 5Ñ)Ô)ğ 3ğ 3‰IˆC�Ø�aŠxˆxØ—’˜{Ñ+Ô+Ğ+Ğ+Ø�˜E™
œ
a™Ò'Ğ'Ø—’˜|Ñ,Ô,Ğ,Ø �NŠN˜4˜4 § ¢ ¨RÑ 0Ô 0Ñ1Ô1Ñ 2Ô 2Ğ 2Ğ 2ğ ˆr Tr Ú check_endc óB — | |¦ « rdS d}t | j | j dz d¦ « D ]J}| d| j |¬¦ « }| d| j |¬¦ « }|r|dz
}|r|dz }ŒK|r| d|¦ « €d n|S |S )Nr rL r r r )Ú
is_board_fullrQ r Úcheck_partial_winÚ game_result)r r rZ ÚscorerD Ú current_wonÚopp_wons r Ú
eval_boardz TicTacToeGameManager.eval_board] sÎ € Ø × Ò ˜eÑ $Ô $ğ Ø�1؈İ�t”¨¬¸1Ñ(<¸bÑAÔAğ ğ ˆAØ ×0Ò0°°T´_ÈEĞ0ÑRÔRˆKØ×,Ò,¨Q°´ÀuĞ,ÑMÔMˆGØğ
ؘ‘�Øğ
ؘ‘
�øØ ğ JØ ×+Ò+¨B°Ñ6Ô6Ğ>�4�4ÀEĞ I؈ r c ó\ ‡— g }‰D ]3}| || d¦ « ¦ « ¦ « Œ4‰j D ]3}| || d¦ « ¦ « ¦ « Œ4ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }| |¦ « |D ]3}| || d¦ « ¦ « ¦ « Œ4|S )Nr c ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS r> r@ rB s €r rE zETicTacToeGameManager.full_iterate_array_all_diags.<locals>.<listcomp>u rF r r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> rH rB s €r rE zETicTacToeGameManager.full_iterate_array_all_diags.<locals>.<listcomp>v rJ r )rN rO rP rQ rR rS ) r r8 r9 rU r; r< rV rW rA s ` r Ú full_iterate_array_all_diagsz1TicTacToeGameManager.full_iterate_array_all_diagsl s? ø€ àˆØğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1à”5ğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1àSĞSĞSĞSE°3´9¸Q´<°-À!Ñ2CÀSÄYÈqÄ\Ñ,RÔ,RĞSÑSÔSˆØfĞfĞfĞf½uÀcÄiĞPQÄlÀ]ĞUVÑEVĞX[ÔXaĞbcÔXdÑ?eÔ?eĞfÑfÔfˆ
Ø
� Š �]Ñ#Ô#Ğ#Øğ 3ğ 3ˆDğ
�NŠN˜4˜4 § ¢ ¨RÑ 0Ô 0Ñ1Ô1Ñ 2Ô 2Ğ 2Ğ 2àˆr c óÊ ‡ ‡— ‰ ¦ « ‰j ¦ « z }| ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « ¦ « | ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « ¦ « ˆ fd„|D ¦ « }t j |¦ « S )Nc ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS r> r@ ©rC rD r s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>� s( ø€ Ğ`Ğ`Ğ`°�œ ¨Ğ+Ñ+Ô+Ğ`Ğ`Ğ`r r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> rH ri s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>‚ s2 ø€ ĞkĞkĞk¸1�œ¥¤ ¨%Ñ 0Ô 0°AĞ6Ñ6Ô6ĞkĞkĞkr c
ól •— g | ]0}t j |d ‰j t |¦ « z
fd¬¦ « ‘Œ1S )r éıÿÿÿ)Úconstant_values)r# Úpadr r* )rC Úvectorr s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>„ s> ø€ ĞpĞpĞpĞ^d•2”6˜& 1 d¤o½¸F¹ ¼ Ñ&CĞ"DĞVXĞYÑYÔYĞpĞpĞpr )ÚtolistrP rS rQ rR r# Úarray)r r Úvectorss `` r Úextract_all_vectorsz(TicTacToeGameManager.extract_all_vectors sŞ øø€ Ø—,’,‘.”. 5¤7§>¢>Ñ#3Ô#3Ñ3ˆØ�ŠĞ`Ğ`Ğ`Ğ`µU¸E¼KȼN¸?ÈQÑ;NĞPUÔP[Ğ\]ÔP^Ñ5_Ô5_Ğ`Ñ`Ô`ÑaÔaĞaØ�ŠĞkĞkĞkĞkÅÀuÄ{ĞSTÄ~ÀoĞXYÑFYĞ[`Ô[fĞghÔ[iÑ@jÔ@jĞkÑkÔkÑlÔlĞlàpĞpĞpĞpĞhoĞpÑpÔpˆİŒx˜Ñ Ô Ğ r c ó< — | || j |¬¦ « S )Nr )r] r )r r
r s r Ú check_winz TicTacToeGameManager.check_win‡ s! € ğ ×%Ò% f¨d¬oÀUĞ%ÑKÔKĞKr c ó¤ ‡— |€| ¦ « }| |ˆfd„¦ « }|D ] }t |t ¦ « s|r dS Œ dS )aC
This function checks if the supplied player has won the game with a full win (num_to_win == board_size).
:param player: The player to check for (1 or -1).
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
Nc ó4 •— t j | ‰k ¦ « S r )r# Úall)Úpartr
s €r ú<lambda>z5TicTacToeGameManager.check_full_win.<locals>.<lambda>– s ø€ ½b¼fÀTÈVÂ^Ñ>TÔ>T€ r TF)Ú get_boardrY Ú
isinstanceÚstr)r r
r ÚmatchesÚmatchs ` r Úcheck_full_winz#TicTacToeGameManager.check_full_win� sm ø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×)Ò)¨%Ğ1TĞ1TĞ1TĞ1TÑUÔUˆØ ğ ğ ˆEİ ˜e¥SÑ)Ô)ğ
¨eğ
Ø�t�tøàˆur Únc óª ‡‡— |€| ¦ « }| |ˆˆfd„¦ « }|D ] }t j |‰k ¦ « r dS Œ dS )aˆ
This function checks if the supplied player has won the game with a partial win (num_to_win < board_size).
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
Nc ót •— t j | ‰k t j ‰t j ¬¦ « d¦ « S ©Nr! Úvalid)r# ÚconvolveÚonesr% ©ry r� r
s €€r rz z8TicTacToeGameManager.check_partial_win.<locals>.<lambda>© s: ø€ İ46´KÀÈÂ ÕRTÔRYĞZ[ÕceÔcjĞRkÑRkÔRkØ@Gñ5Iô 5Iğ r TF)r{ rf r# Úany)r r
r� r r~ r s `` r r] z&TicTacToeGameManager.check_partial_win� s‹ øø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×3Ò3°Eğ5Iğ 5Iğ 5Iğ 5Iğ 5IñJô Jˆğ
ğ ğ ˆEİŒv�e˜q’jÑ!Ô!ğ
Ø�t�tğ
ğ ˆur c óâ — |€| ¦ « }t j | |¦ « ¦ « d¦ « }t j ddd|ft j ¬¦ « }t j ||k dd¦ « ¦ « }t j j
||¬¦ « }t j ||k ¦ « ¦ « S )Nr r r! )Úweight)
r{ ÚthÚtensorrs Ú unsqueezer‡ ÚlongÚwhereÚnnÚ
functionalÚconv2dr‰ Úitem)r r
r� r rr r‹ Úvectors_where_playerÚress r Ú check_partial_win_vectorizedz1TicTacToeGameManager.check_partial_win_vectorized³ s € Ø ˆ=Ø—N’NÑ$Ô$ˆEİ”)˜D× 4Ò 4°UÑ ;Ô ;Ñ<Ô<×FÒFÀqÑIÔIˆİ”˜!˜Q 1˜ R¬WĞ5Ñ5Ô5ˆİ !œx¨°6Ò(9¸1¸aÑ @Ô @× EÒ EÑ GÔ GĞ İŒeÔ ×%Ò%Ğ&:À6Ğ%ÑJÔJˆİŒv�c˜Q’hÑ Ô ×$Ò$Ñ&Ô&Ğ&r c óP ‡‡— |€| ¦ « }| |ˆˆfd„¦ « }d„ |D ¦ « }d}|D ]ß}t |t ¦ « r|}Œ|D ]Â}|‰k rº| |¦ « }| |¦ « } || k r|| dz z
n| || ¦ « }
|
| j k rd| j dz
|
z
dfin
dd|
| j z
fi} || d< |xdk r |
df|d œc c S xd
k r d|
f|d œc c S xd k r | c c S d k r| c c S ŒÃŒàddd œS )
aß
This variation of check_partial_win returns the index of the first partial win found.
The index being the index of the first piece in the winning sequence.
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: A dictionary containing the index and the position of the winning sequence.
Nc ój •— t j | ‰k t j ‰t ¬¦ « d¦ « S r„ )r# r† r‡ Úintrˆ s €€r rz zATicTacToeGameManager.check_partial_win_to_index.<locals>.<lambda>È s3 ø€ İ*,¬+°t¸v²~ÍÌĞPQÕY\ĞH]ÑH]ÔH]Ø6=ñ+?ô +?ğ r c ód — g | ]-}t |t ¦ « s| ¦ « n|‘Œ.S © )r| r} rp ©rC Úxs r rE zCTicTacToeGameManager.check_partial_win_to_index.<locals>.<listcomp>Ë s3 € ĞPĞPĞPÀ1¥Z°µ3Ñ%7Ô%7Ğ>�1—8’8‘:”:�:¸QĞPĞPĞPr r; r r r Úpos)r rŸ r< rK rM )r{ rY r| r} r r ) r r
r� r ÚindicesrŸ ro ÚelÚ vector_indexÚ pos_indexÚ
element_indexÚdiag_idxs `` r Úcheck_partial_win_to_indexz/TicTacToeGameManager.check_partial_win_to_index¼ sõ øø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×)Ò)¨%ğ+?ğ +?ğ +?ğ +?ğ +?ñ@ô @ˆğ QĞPÈĞPÑPÔPˆàˆØ ğ ,ğ ,ˆFݘ&¥#Ñ&Ô&ğ
Ø �ØØ ğ
,ğ
,�ؘ’7�7Ø#*§=¢=°Ñ#8Ô#8�LØ '§
¢
¨cÑ 2Ô 2�IàFRĞU^ÒF^ĞF^ L°IÀ±MÑ$BĞ$BĞdk×dqÒdqØ ñe+ô e+�Mğ Q^Ğ`dÔ`oÒPoĞPo˜ 4¤?°QÑ#6¸-Ñ"GÈĞ!Kğ Mğ Mà ! ]°T´_Ñ%DĞ!EğvGğ ğ '*�H˜U‘OØ Ø "˜UšU˜U˜UØ.;¸QĞ-?ÈĞ#LĞ#LĞ LĞ LĞ LĞ LĞ LØ "˜UšU˜U˜UØ./° Ğ-?ÈĞ#LĞ#LĞ LĞ LĞ LĞ LĞ LØ (˜[š[˜[˜[Ø#+˜O˜O˜O˜O˜OØ )š\˜\Ø#+˜O˜O˜O˜O˜Oøğ'
,ğ. dĞ+Ğ+Ğ+r rœ Úreturn_on_failc ó~ — |d dk s.|d | j k s |d dk s|d | j k r|S |S ©Nr r ©r )r r r§ s r Úreturn_index_if_validz*TicTacToeGameManager.return_index_if_validë sI € Ø �Œ8�aŠ<ˆ<˜5 œ8 t¤Ò6Ğ6¸%À¼(ÀQº,¸,È%ĞPQÌ(ĞVZÔVeÒJeĞJeØ!Ğ !؈ r c óD — t | j | j | j ¬¦ « S )N)r )r r r r ©r s r Úmake_fresh_instancez(TicTacToeGameManager.make_fresh_instanceğ s € İ# D¤O°T´]ÈtÌĞ_Ñ_Ô_Ğ_r rŸ c óà — t j |¦ « ¦ « dk r|S |xdk r) | |d |d |z
f|¬¦ « S xdk r) | |d |z
|d f|¬¦ « S xdk r, | |d |z
|d |z
f|¬¦ « S dk r+| |d |z
|d |z f|¬¦ « S d S )Nr r; r ©r§ r< rK rM )r# rq rx r« ©r r rŸ r� s r Ú get_previousz!TicTacToeGameManager.get_previousó s € İ
Œ8�E‰?Œ?× Ò Ñ Ô AÒ %Ğ %؈LØØ�’��Ø×1Ò1°5¸´8¸UÀ1¼Xȹ\Ğ2JĞ[`Ğ1ÑaÔaĞaØ�’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄĞ2JĞ[`Ğ1ÑaÔaĞaØ �’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeØ ’ � Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeğ � r c óğ — t j |¦ « ¦ « | j dz
k r|S |xdk r) | |d |d |z f|¬¦ « S xdk r) | |d |z |d f|¬¦ « S xdk r, | |d |z |d |z f|¬¦ « S dk r+| |d |z |d |z
f|¬¦ « S d S )Nr r; r r° r< rK rM )r# rq rx r r« r± s r Úget_nextz TicTacToeGameManager.get_next s € İ
Œ8�E‰?Œ?× Ò Ñ Ô D¤O°aÑ$7Ò 7Ğ 7؈LØØ�’��Ø×1Ò1°5¸´8¸UÀ1¼Xȹ\Ğ2JĞ[`Ğ1ÑaÔaĞaØ�’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄĞ2JĞ[`Ğ1ÑaÔaĞaØ �’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeØ ’ � Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeğ � r c ó^ — |€| ¦ « }t j |dk ¦ « S ©Nr )r{ r# rx r& s r r\ z"TicTacToeGameManager.is_board_full
s* € Ø ˆ=Ø—N’NÑ$Ô$ˆEİŒv�e˜q’jÑ!Ô!Ğ!r c ó4 — | j ¦ « S r )r Úcopyr s r r{ z TicTacToeGameManager.get_board s € ØŒz�ŠÑ Ô Ğ r c óf — | ¦ « | _ | j ¦ « S r )r r r¸ r s r ÚresetzTicTacToeGameManager.reset s' € Ø×*Ò*Ñ,Ô,ˆŒ
ØŒz�ŠÑ Ô Ğ r c ó — | j S r rª r s r Úget_board_sizez#TicTacToeGameManager.get_board_size s
€ ،Рr c óB — | j rdS | j d¦ « | ¦ « t | j ¦ « D ]–}t | j ¦ « D ]}| j | | | j k r#| |dz dz |dz dz ¦ « ŒA| j | | | j k r"|
|dz dz |dz dz ¦ « Œ€Œ—t j
¦ « t j ¦ « dS )NF)r r r r1 é2 T)r r ÚfillÚ _draw_boardrQ r r r
Ú _draw_circler Ú _draw_crossr2 ÚeventÚpumpr4 Úflip)r r; r< s r ÚrenderzTicTacToeGameManager.render s € Ø Œ=ğ Ø�5à Œ ×Ò˜Ñ#Ô#Ğ#Ø ×ÒÑÔĞݘœ Ñ)Ô)ğ Eğ EˆCİ ˜Tœ_Ñ-Ô-ğ
Eğ
E�Ø”:˜c”? 3Ô'¨4¬;Ò6Ğ6Ø×%Ò% c¨C¡i°"¡n°c¸C±iÀ"±nÑEÔEĞEĞEà”Z ”_ SÔ)¨TÔ->Ò>Ğ>Ø×$Ò$ S¨3¡Y°¡^°S¸3±YÀ±^ÑDÔDĞDøğ
Eõ Œ�
Š
‰Œˆİ
Œ
�ŠÑÔĞ؈tr c ól — | j rd S t j | j d||fdd¦ « d S )NÚgreené( r )r r2 ÚdrawÚcircler ©r r� Úys r rÁ z!TicTacToeGameManager._draw_circle. s9 € Ø Œ=ğ Ø ˆFİ
Œ�Š�t”{ G¨a°¨V°R¸Ñ;Ô;Ğ;Ğ;Ğ;r c óø — | j rd S t j | j d|dz
|dz
f|dz |dz fd¦ « t j | j d|dz |dz
f|dz
|dz fd¦ « d S )NÚredrÉ r )r r2 rÊ Úliner rÌ s r r z TicTacToeGameManager._draw_cross3 s‡ € Ø Œ=ğ Ø ˆFİ
Œ� Š �T”[ %¨!¨b©&°!°b±&Ğ)9¸AÀ¹FÀAÈÁFĞ;KÈQÑOÔOĞOİ
Œ� Š �T”[ %¨!¨b©&°!°b±&Ğ)9¸AÀ¹FÀAÈÁFĞ;KÈQÑOÔOĞOĞOĞOr c óî — t d| j dz d¦ « D ]Z}t d| j dz d¦ « D ]>}t j | j dt j ||dd¦ « d¦ « Œ?Œ[d S )Nr r1 )éÿ rÒ rÒ r )rQ r r2 rÊ Úrectr ÚRectrÌ s r rÀ z TicTacToeGameManager._draw_board9 sŒ € İ�q˜$œ/¨CÑ/°Ñ5Ô5ğ Wğ WˆAݘ1˜dœo°Ñ 3°SÑ9Ô9ğ
Wğ
W�İ”— ’ ˜Tœ[¨/½2¼7À1ÀaÈÈcÑ;RÔ;RĞTUÑVÔVĞVĞVğ
Wğ Wğ Wr c ó$ — | j | dk S r¶ r )r r s r Úis_emptyz TicTacToeGameManager.is_empty> s € ØŒz˜%Ô AÒ%Ğ%r Ú observationc óø — g }| | j | j ¦ « }t | j ¦ « D ]B}t | j ¦ « D ]+}|| | dk r| ||g¦ « Œ,ŒC|S )a
Legal moves are the empty spaces on the board.
:param observation: A 2D numpy array representing the current state of the game.
:param player: The player to check for. Since the game is symmetric, this is ignored.
:return: A list of legal moves.
r )rO r rQ rN )r r× r
Ú legal_movesr; r< s r r) z$TicTacToeGameManager.get_valid_movesA s� € ğ ˆ Ø!×)Ò)¨$¬/¸4¼?ÑKÔKˆ ݘœ Ñ)Ô)ğ 3ğ 3ˆCİ ˜Tœ_Ñ-Ô-ğ
3ğ
3�Ø ˜sÔ# CÔ(¨AÒ-Ğ-Ø ×&Ò&¨¨S zÑ2Ô2Ğ2øğ
3ğ Ğr c ó> — | j rdS t j ¦ « dS )NFT)r r2 Úquitr s r Ú pygame_quitz TicTacToeGameManager.pygame_quitP s! € Ø Œ=ğ Ø�5İ
Œ‰ Œ ˆ ؈tr c ó´ — | j rd S d„ t j ¦ « D ¦ « }t j ¦ « d r|S d S )Nc 3 ó K — | ] }|d z V — Œ
dS )r1 Nrœ r� s r ú <genexpr>z8TicTacToeGameManager.get_click_coords.<locals>.<genexpr>Y s& è è € Ğ:Ğ: !�Q˜#‘XĞ:Ğ:Ğ:Ğ:Ğ:Ğ:r r )r r2 ÚmouseÚget_posÚ get_pressed)r Ú mouse_poss r Úget_click_coordsz%TicTacToeGameManager.get_click_coordsV s] € Ø Œ=ğ Ø ˆFØ:Ğ:¥r¤x×'7Ò'7Ñ'9Ô'9Ğ:Ñ:Ô:ˆ İ
Œ8× Ò Ñ !Ô ! !Ô $ğ Ø Ğ ğ ğ r c ó — | j rd S | ¦ « | ¦ « �-| ¦ « \ }}|| | dk r||fS ŒV)NTr )r Úcheck_pg_eventsrä )r r r� rÍ s r Úget_human_inputz$TicTacToeGameManager.get_human_input] sp € Ø Œ=ğ Ø ˆFğ Ø × Ò Ñ "Ô "Ğ "Ø×$Ò$Ñ&Ô&Ğ2Ø×,Ò,Ñ.Ô.‘��1ؘ”8˜A”; !Ò#Ğ#Ø ˜a˜4�Kğ r c óÔ — | j rd S t j ¦ « D ]?}|j t j k r(| ¦ « t j d¦ « Œ@d S r¶ ) r r2 rà ÚgetÚtypeÚQUITrÜ ÚsysÚexit)r rà s r ræ z$TicTacToeGameManager.check_pg_eventsi sc € Ø Œ=ğ Ø ˆFİ”X—\’\‘^”^ğ ğ ˆEØŒz�RœWÒ$Ğ$Ø× Ò Ñ"Ô"Ğ"İ”˜‘ ” � øğ ğ r c ó@ — t j || j j ¦ « S )zÜ
Converts an integer move from the network to a board index.
:param move: An integer move selected from the network probabilities.
:return: A tuple representing the board index (int,int).
©r# Ú
unravel_indexr rR )r Úmoves r Únetwork_to_boardz%TicTacToeGameManager.network_to_boardq s € õ Ô d¤jÔ&6Ñ7Ô7Ğ7r c óŞ ‡ — ‰ j rd S t j d¬¦ « t | ¦ « � \ }}t |¦ « ˆ fd„|D ¦ « }t
j ||¬¦ « t j d¬¦ « t j d¦ « t j
d¦ « t j d ¦ « t j
¦ « }t j |d
d ¬ ¦ « | d
¦ « t! j |¦ « }t$ j ‰ j d¦ « }t! j d‰ j ¦ « |¦ « }|j |j z } t! j dt5 |j |j ¦ « | f¦ « }
|
|d¦ « |
|d
|j f¦ « |
|¦ « d S )N)é é
)Úfigsizec óª •— g | ]O}t j |‰j j ¦ « d › dt j |‰j j ¦ « d › �‘ŒPS )r ú;r rï )rC r� r s €r rE zKTicTacToeGameManager.save_screenshot_with_probabilities.<locals>.<listcomp> sf ø€ ğ ğ ğ Ğop•RÔ% a¨¬Ô)9Ñ:Ô:¸1Ô=ĞjĞjÅÔ@PĞQRĞTXÔT^ÔTdÑ@eÔ@eĞfgÔ@hĞjĞjğ ğ ğ r )r� rÍ éZ )ÚrotationÚMoveÚ ProbabilityzAction probabilitiesÚpngÚtight)ÚformatÚ bbox_inchesr ÚRGBAÚRGB)r r ) r ÚpltÚfigureÚzipÚitemsÚprintÚsnsÚbarplotÚxticksÚxlabelÚylabelÚtitleÚioÚBytesIOÚsavefigÚseekr Úopenr2 ÚimageÚtostringr Ú frombytesÚget_sizeÚheightÚnewÚmaxÚwidthÚpasteÚsave) r Ú action_probsÚpathÚlabelsÚ
probabilitiesÚbufÚplot_imgÚsurface_bufferÚ surface_imgÚ total_heightÚ combined_imgs ` r Ú"save_screenshot_with_probabilitiesz7TicTacToeGameManager.save_screenshot_with_probabilitiesy sÍ ø€ Ø Œ=ğ Ø ˆFİ Œ
˜8Ğ$Ñ$Ô$Ğ$İ # \×%7Ò%7Ñ%9Ô%9Ğ :Ñ ˆ�
İ
ˆmÑ Ô Ğ ğğ ğ ğ Øğñ ô ˆå Œ �f
Ğ.Ñ.Ô.Ğ.İ Œ
˜BĞ Ñ Ô Ğ İ Œ
�6ÑÔĞİ Œ
�=Ñ!Ô!Ğ!İ Œ Ğ(Ñ)Ô)Ğ)åŒj‰lŒlˆİ Œ �C °7Ğ;Ñ;Ô;Ğ;Ø �Š�‰ Œ ˆ İ”:˜c‘?”?ˆõ œ×*Ò*¨4¬;¸Ñ?Ô?ˆİ”o f¨d¬k×.BÒ.BÑ.DÔ.DÀnÑUÔUˆ 𠔨Ô);Ñ;ˆ İ ”y ¨X¬^¸[Ô=NÑ)OÔ)OĞQ]Ğ(^Ñ_Ô_ˆ Ø×Ò˜8 VÑ,Ô,Ğ,Ø×Ò˜;¨¨H¬OĞ(<Ñ=Ô=Ğ=Ø×Ò˜$Ñ Ô Ğ Ğ Ğ r c ó — | |z S r rœ )r r
s r Úget_canonical_formz'TicTacToeGameManager.get_canonical_form— s € à�v‰~Ğ r Úactionc óŒ — t |t ¦ « r| |¦ « }| ¦ « }|||< |S r )r| rš rò r¸ )r r r* r
Úboard_s r Úget_next_statez#TicTacToeGameManager.get_next_state› sB € İ �f�cÑ "Ô "ğ 3Ø×*Ò*¨6Ñ2Ô2ˆFØ—’‘ ” ˆØ ˆˆv‰Øˆ
r Úvalc ó — || _ d S r )r )r r. s r Ú set_headlessz!TicTacToeGameManager.set_headless¢ s
€ ؈Œ
ˆ
ˆ
r Ústatec ó: — t j |dk dd¦ « }|S r© )r# r� )r r1 r
Úmasks r Úget_invalid_actionsz(TicTacToeGameManager.get_invalid_actions¥ s € İŒx˜ š
A qÑ)Ô)ˆØˆ r c óz — t | j ¦ « dd¦ « dd¦ « S )NÚ1ÚXz-1ÚO)r} r Úreplacer s r Ú__str__z TicTacToeGameManager.__str__© s0 € İ�4”:‰Œ×&Ò& s¨CÑ0Ô0×8Ò8¸¸sÑCÔCĞCr r )T)rœ )r N)6Ú__name__Ú
__module__Ú __qualname__Ú__doc__rš Úboolr Útupler r r r# ÚndarrayÚlistr/ r ÚcallablerY rb rf rs ru r€ r] r— Údictr} r¦ r« r® r² r´ r\ r{ rº r¼ rÆ rÁ r rÀ rÖ r) rÜ rä rç ræ rò r' Ú staticmethodr) r- r0 r4 r: rœ r r r r s8 € € € € € ğğ ğ3ğ 3 3ğ 3°$ğ 3ÈDğ 3ğ 3ğ 3ğ 3ğ#˜3ğ # uğ #°ğ #ğ #ğ #ğ #ğ¨#°©*ğ ¸ğ ğ ğ ğ ğğ ğ ğ*°B´Jğ *ÈTğ *ğ *ğ *ğ *ğ ğ ğ ğ b¤jğ ¸ğ ÀTğ ğ ğ ğ ğB
ğ
¤
ğ
°tğ
ÀsÈTÁzğ
ğ
ğ
ğ
ğ °´
ğ À(ğ ğ ğ ğ ğ&!¨¬ğ !ğ !ğ !ğ !ğLğ L ğ L°Dğ Lğ Lğ Lğ Lğ ğ Sğ ¸ğ ğ ğ ğ ğ ğ ¨ğ °ğ ÀDğ ğ ğ ğ ğ,'ğ '°3ğ '¸3ğ 'Ètğ 'ğ 'ğ 'ğ 'ğ-,ğ -,°ğ -,¸ğ -,ÈTĞRWĞY\ĞR\ÔM]Ğ`dÑMdğ -,ğ -,ğ -,ğ -,ğ^ğ ¨5ğ À%ğ ĞQVğ ğ ğ ğ ğ
`ğ `ğ `ğ f %ğ f¨cğ f°cğ fğ fğ fğ fğ f˜eğ f¨#ğ f°#ğ fğ fğ fğ fğ"ğ "¨4ğ "ğ "ğ "ğ "ğ
!ğ !ğ !ğ!ğ !ğ !ğ ğ ğ ğ˜ğ ğ ğ ğ ğ$<ğ <ğ <ğ <ğ
Pğ Pğ Pğ Pğ Wğ Wğ Wğ
&˜eğ &¨ğ &ğ &ğ &ğ &ğ
ğ
¨2¬:ğ
¸s¸{Àdğ
ĞVZğ
ğ
ğ
ğ
ğ ˜Tğ ğ ğ ğ ğ ğ ğ ğ R¤Zğ ğ ğ ğ ğ ğ ğ ğ8ğ 8ğ 8ğ ğ ğ ğ< ğ ¨R¬Zğ ğ ğ ñ „\ğ ğ B¤J𠸸 ¸uğ Ècğ ĞVXÔV`ğ ğ ğ ğ ğ ğ ğ ğ ğ ğ¨¬ğ ¸Sğ ğ ğ ğ ğDğ Dğ Dğ Dğ Dr r )r r+ rì Úmatplotlib.pyplotÚpyplotr Únumpyr# Úpygamer2 Úseabornr ÚtorchrŒ ÚPILr Ú mu_alpha_zero.General.az_gamer r rœ r r ú<module>rN sÎ ğØ € € € Ø
€
€
€
Ø
€
€
€
à Ğ Ğ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à 7Ğ 7Ğ 7Ğ 7Ğ 7Ğ 7ğ[Dğ [Dğ [Dğ [Dğ [D˜=ñ [Dô [Dğ [Dğ [Dğ [Dr | 31,194 | Python | .pyt | 118 | 260.949153 | 2,695 | 0.344221 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,572 | asteroids.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/__pycache__/asteroids.cpython-311.pyc | §
Ô`�f( ã óJ — d dl Zd dlZd dlmZ d dlmZ G d„ de¦ « ZdS )é N)Úmake)Ú
MuZeroGamec ó6 — e Zd Zdej defd„Zd„ Zdej pej fd„Z
defd„Z defd„Z depd de
pd fd
„Zd „ Zdd
edepd dedej pej ee
ffd„Zdd
edepd dedej pej ee
ffd„Zd„ Zdej fd„Zdej defd„Zd S )Ú AsteroidsÚstateÚplayerc ó — |S ©N© ©Úselfr r s úK/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Game/asteroids.pyÚ get_state_for_passive_playerz&Asteroids.get_state_for_passive_player
s € ؈ ó c ó< — t d¦ « | _ d| _ d S )NzALE/Asteroids-v5F)r ÚenvÚdone©r
s r Ú__init__zAsteroids.__init__
s € İĞ*Ñ+Ô+ˆŒØˆŒ ˆ ˆ r Úreturnc ó> — | j ¦ « \ }}|S r
)r Úreset)r
ÚobsÚ_s r r zAsteroids.reset s € Ø”— ’ Ñ!Ô!‰ˆˆQ؈
r c ó — dS )Nr r r s r Úget_noopzAsteroids.get_noop s € ؈qr c ó$ — | j j j S r
)r Ú action_spaceÚnr s r Úget_num_actionszAsteroids.get_num_actions s € ØŒxÔ$Ô&Ğ&r Nc ó — | j S r
)r )r
r s r Ú game_resultzAsteroids.game_result s
€ ØŒyĞr c ó — t ¦ « S r
)r r s r Úmake_fresh_instancez Asteroids.make_fresh_instance s € ݉{Œ{Ğr é ÚactionÚ
frame_skipc óZ — | j |¦ « \ }}}}}|| _ |||fS r
)r Ústepr )r
r& r r' r Úrewr r s r Úget_next_statezAsteroids.get_next_state! s3 € à #œxŸ}š}¨VÑ 4Ô 4Ñ ˆˆS�$˜˜1؈Œ Ø�C˜ˆ~Ğ r c ó� — | ||¦ « \ }}}t |dz
¦ « D ] }| ||¦ « \ }}}Œ |||fS )Né )r+ Úrange)r
r& r r' r r* r Úis r Úframe_skip_stepzAsteroids.frame_skip_step' sc € à ×,Ò,¨V°VÑ<Ô<‰ˆˆS�$İ�z A‘~Ñ&Ô&ğ Ağ AˆAØ !× 0Ò 0°¸Ñ @Ô @‰NˆC��d�dØ�C˜ˆ~Ğ r c ó — d S r
r r s r ÚrenderzAsteroids.render. s € Ø ˆr c ó> — | j j ¦ « S r
)r r Úsample)r
r Úkwargss r Úget_random_valid_actionz!Asteroids.get_random_valid_action1 s € ØŒxÔ$×+Ò+Ñ-Ô-Ğ-r c óN — t j | ¦ « ¦ « S r
)ÚnpÚonesr r s r Úget_invalid_actionsz Asteroids.get_invalid_actions4 s € İŒw˜×,Ò,Ñ.Ô.Ñ0Ô0Ğ0r )r% )Ú__name__Ú
__module__Ú __qualname__r8 ÚndarrayÚintr r ÚthÚTensorr r r Úboolr" r$ r+ r0 r2 r6 r: r r r r r sÄ € € € € € ğ°"´*ğ Àcğ ğ ğ ğ ğğ ğ ğ�r”zĞ. R¤Yğ ğ ğ ğ ğ˜#ğ ğ ğ ğ ğ' ğ 'ğ 'ğ 'ğ 'ğ # +¨ğ °$°,¸$ğ ğ ğ ğ ğğ ğ ğ ğ Sğ °#°+¸ğ È3ğ Ø ŒJĞ #˜"œ) S¨$ğX0ğ ğ ğ ğ ğ ğ cğ °3°;¸$ğ ÈCğ Ø ŒJĞ #˜"œ) S¨$ğY0ğ ğ ğ ğ ğ
ğ
ğ
ğ.¨R¬Zğ .ğ .ğ .ğ .ğ1¨¬ğ 1¸Sğ 1ğ 1ğ 1ğ 1ğ 1ğ 1r r ) Únumpyr8 Útorchr@ Ú gymnasiumr Ú mu_alpha_zero.General.mz_gamer r r r r ú<module>rG su ğØ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4ğ-1ğ -1ğ -1ğ -1ğ -1�
ñ -1ô -1ğ -1ğ -1ğ -1r | 3,983 | Python | .pyt | 20 | 198.1 | 830 | 0.346367 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,573 | utils.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/__pycache__/utils.cpython-311.pyc | §
‰Ó¢f� ã
ó$ — d dl Z d dlZd dlmZ d dlZd dlZd dlZd dl m
Z
d dl m Z d3dej
dej
fd„Zdej
defd „Zd
ej
d ee fd „Zdej d
eeef dedej fd„Zdej defd„Zdedefd„Zdej
fd„Zd4dej
defd„Zd4dej
defd „Z d efd „Z d ej
d efd!„Z d ej
d efd"„Z d5d%ed&e pdd'e d(e d)ed*e f d+„Z!d,ej d-ej fd.„Z"d/ed0ej
d1ee fd2„Z#dS )6é N)ÚCallable)ÚImage)Ú MuZeroConfigÚ observationsÚactionsc ó2 — t j | |f|¬¦ « S )N©Údim)ÚthÚcat)r r r
s úI/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/utils.pyÚadd_actions_to_obsr
s € İ
Œ6�< Ğ)¨sĞ 3Ñ 3Ô 3Ğ3ó Úactionc ó¢ — t j d| j d | j d f|t j | j ¬¦ « }t | |¦ « S )Né é ©ÚdtypeÚdevice)r ÚfullÚshapeÚfloat32r r )r r s r
Úmatch_action_with_obsr sO € İ
ŒW�a˜ Ô+¨AÔ.° Ô0BÀ1Ô0EĞFÈÕVXÔV`Ø (Ô /ğ1ñ 1ô 1€Få ˜l¨FÑ 3Ô 3Ğ3r Úobservation_batchÚ action_batchc óp ‡ — ˆ fd„|D ¦ « }t j |d¬¦ « }t ‰ |d¬¦ « S )Nc ó’ •— g | ]C}t j d d ‰j d ‰j d f|t j ‰j ¬¦ « ‘ŒDS )r r é r )r r r r r )Ú.0r r s €r
ú
<listcomp>z/match_action_with_obs_batch.<locals>.<listcomp> sm ø€ ğ Tğ Tğ TØ<Bõ Œw˜˜1Ğ /Ô 5°aÔ 8Ğ:KÔ:QĞRSÔ:TĞUĞW]ÕegÔeoØ /Ô 6ğ8ñ 8ô 8ğ Tğ Tğ Tr r r r )r r r )r r Útensorsr s ` r
Úmatch_action_with_obs_batchr# s\ ø€ ğTğ Tğ Tğ TØFRğTñ Tô T€GåŒf�W !Ğ$Ñ$Ô$€Gİ Ğ /°¸aĞ @Ñ @Ô @Ğ@r ÚsizeÚresizeÚreturnc ó„ — |s| S t j | ¦ « }| |¦ « }t j |¦ « S ©N)r Ú fromarrayr% ÚnpÚarray)r r$ r% Úobss r
Ú
resize_obsr- s> € Ø ğ ØĞİ
Œ/˜,Ñ
'Ô
'€CØ
�*Š*�TÑ
Ô
€Cİ
Œ8�C‰=Œ=Ğr ÚstateÚscalec ó — |s| S | dz S )Néÿ © )r. r/ s r
Ú scale_stater3 & s € Ø ğ ؈ à �3‰;Ğr Ú num_actionsc ó — | dz |z S ©Nr r2 )r r4 s r
Ú scale_actionr7 - s € à �Q‰J˜+Ñ %Ğ%r Ú hidden_statec ón — d}t | j ¦ « dk r| d¦ « } d}| | d¦ « | d¦ « d¦ « dd¬¦ « d d¦ « }| | d¦ « | d¦ « d¦ « dd¬¦ « d d¦ « }||z
}d ||dk < | |z
|z } |r| d¦ « n| S )
NFr r Tr éÿÿÿÿr )r
Úkeepdimçñh㈵øä>)Úlenr Ú unsqueezeÚviewr$ ÚmaxÚminÚsqueeze)r8 Ú was_reshapedÚmax_Úmin_Ú max_min_difs r
Úscale_hidden_staterG 2 s4 € Ø€Lİ
ˆ<Ô Ñ Ô !Ò#Ğ#Ø#×-Ò-¨aÑ0Ô0ˆ ؈ Ø × Ò ˜\× .Ò .¨qÑ 1Ô 1°,×2CÒ2CÀAÑ2FÔ2FÀrÑ JÔ J× NÒ NĞSTĞ]aĞ NÑ bÔ bĞcdÔ e× oÒ oĞprÑ sÔ s€DØ × Ò ˜\× .Ò .¨qÑ 1Ô 1°,×2CÒ2CÀAÑ2FÔ2FÀrÑ JÔ J× NÒ NĞSTĞ]aĞ NÑ bÔ bĞcdÔ e× oÒ oĞprÑ sÔ s€Dؘ‘+€KØ$(€K� ˜qÒ Ñ!Ø 4Ñ'¨;Ñ6€LØ&2Ğ Dˆ<× Ò Ñ "Ô "Ğ "¸ ĞDr çü©ñÒMbP?ÚvalueÚec ó� — t j | ¦ « t j t j | ¦ « dz ¦ « dz
z || z z S r6 ©r ÚsignÚsqrtÚabs©rI rJ s r
Úscale_reward_valuerQ ? s: € İ
Œ7�5‰>Œ>�RœW¥R¤V¨E¡]¤]°QÑ%6Ñ 7Ô 7¸!Ñ ;Ñ <¸qÀ5¹yÑ HĞHr c ó¨ — t j | ¦ « t j ddt j | ¦ « dz dz z z ¦ « dz
dz dz dz
z S )Nr gü©ñÒMbp?rH gü©ñÒMb`?r rL rP s r
Úinvert_scale_reward_valuerS C sZ € İ
Œ7�5‰>Œ>İŒg�a˜)¥r¤v¨e¡}¤}°qÑ'8¸5Ñ'@ÑAÑAÑBÔBÀQÑFÈ9Ñ
UØñ
àñ
ñ ğ r Úrewardc ó2 — t j | dz d¦ « S )Nr é )ÚmathÚlog)rT s r
Ú scale_rewardrY L s € İ Œ8�F˜Q‘J Ñ "Ô "Ğ"r ÚxÚ support_sizec ón — t | ¦ « } t j | | |¦ « } d| | ¦ « z
z
}| | ¦ « z
}t j | d¦ « d|z dz f| j ¬¦ « }| d|| ¦ « z t j ¦ « |¦ « | d|| ¦ « z dz t j ¦ « |¦ « n# t $ r Y nw xY w|S )Nr r r )r ) rQ r ÚclampÚfloorÚzerosr$ r Úscatter_ÚtypeÚint64Ú RuntimeError)rZ r[ Úlower_pÚupper_pÚsupports r
Úscalar_to_supportrg P s! € å˜1Ñ Ô €Aİ
Œ��\�M <Ñ0Ô0€AØ�1�q—w’w‘y”y‘=Ñ!€GØ�!—'’'‘)”)‰m€GİŒh˜Ÿš˜q™ œ 1 |Ñ#3°aÑ#7Ğ8ÀÄĞJÑJÔJ€GØ ×Ò�Q˜ ¨¯ª© ¬ Ñ1×7Ò7½¼ÑAÔAÀ7ÑKÔKĞKğ
Ø×Ò˜Ø ˜qŸwšw™yœyÑ(¨1Ñ,¯dªdµ2´8©n¬n¸gñ Gô Gğ Gğ Gøå ğ
ğ
ğ
à ˆğ
øøøğ €Ns ÃA D% Ä%
D2Ä1D2c óğ — t j | |dz d| j | j ¬¦ « d¦ « }t j | |z d¬¦ « }t
|¦ « }| d¦ « S )Nr r r r )r Úaranger r r> ÚsumrS )rZ r[ rf Úoutputs r
Úsupport_to_scalarrl a sq € İŒi˜ ˜
|°aÑ'7¸À!Ä'ĞRSÔRZĞ[Ñ[Ô[×eÒeĞfgÑhÔh€Gİ
ŒV�A˜‘K QĞ
'Ñ
'Ô
'€Fİ
& vÑ
.Ô
.€FØ × Ò ˜AÑ Ô Ğ r FÚmaximizeÚn_trialsÚstorageÚ
study_nameÚ
muzero_configÚ in_memoryÚ directionc óê ‡‡‡‡‡ ‡
‡‡‡‡— dt j fˆ ˆ
ˆˆˆˆˆˆˆˆf
d„ } ddlmŠ ddlmŠ ddlmŠ
ddlm Š ddl
m Š dd l m
Š d
‰_ |rt j ||¬ ¦ « }
n'|€t! d ¦ « ‚t j ||¬
¦ « }
|
| | ¬¦ « t' ‰j › d�d¦ « 5 } t+ j |
j | ¦ « d d d ¦ « d S # 1 swxY w Y d S )NÚtrialc óî •
— | ddd¦ « ‰_ | dddd¬¦ « ‰_ | d d
d ¦ « ‰_ | d d
d ¦ « ‰_ | ddd ¦ « ‰_ | ddd¦ « ‰_ | ddd ¦ « ‰_ | ddd¦ « ‰_ |
ddd
g¦ « ‰_ | ddd¦ « ‰_ | d d d ¦ « ‰_
| d d d ¦ « ‰_ d!‰_ d‰_ t# j t" j ¦ « rd"nd#¦ « }t+ ‰ ¦ « ¦ « | _ ‰
‰¦ « |¦ « } ‰ ‰ ¦ « ‰¦ « } ‰
‰ ¦ « fi ||d$œ¤�}‰€ ‰ ‰ ¦ « ‰|¦ « }n‰}‰€ ‰ ‰j dd%‰j ¬&¦ « }n‰}‰ ‰‰ ¦ « |||d|d%|¬'¦ « }| ¦ « | ¦ « }| |‰j ¦ « tC d(| j" › d)|› d*�¦ « ~~~~|S )+NÚnum_mc_simulationsé< i Úlrr< gš™™™™™©?T)rX Útempg à?r Ú
arena_tempr Úcpuctgffffffæ?Úl2_normg�íµ ÷Æ°>g¸ …ëQ¸®?Úalphagš™™™™™Ù?Úbetagš™™™™™¹?r Ú
loss_scaleÚ
num_blocksrV é Úké ÚepochséÈ iô i, ÚcudaÚcpu)ÚnetworkÚmonte_carlo_tree_searchF)ÚdiskÚ full_diskÚdir_path)ÚheadlessÚarena_overrideÚcheckpointer_verboseÚmemory_overridezTrial z finished with win freq ú.)#Ú suggest_intÚnum_simulationsÚ
suggest_floatry ÚtauÚ arena_tauÚcÚl2r~ r Úsuggest_categoricalÚ balance_termr� ÚKr… Úself_play_gamesÚ num_itersr r r‡ Ú is_availableÚintÚget_num_actionsÚnet_action_sizeÚmake_from_configÚtoÚmake_fresh_instanceÚmax_buffer_sizeÚ
pickle_dirÚcreateÚtrainÚ get_arena_win_frequencies_meanÚreportÚprintÚnumber)ru r r‰ ÚtreeÚ
net_playerÚarenaÚmemÚtrainerÚmeanÚ MemBufferÚ MuZeroNetÚMuZeroSearchTreeÚMzArenaÚ NetPlayerÚTrainerr� Úgamer‘ rq s €€€€€€€€€€r
Ú objectivez-mz_optuna_parameter_search.<locals>.objectivek s ø€ Ø(-×(9Ò(9Ğ:NĞPRĞTWÑ(XÔ(Xˆ
Ô%Ø ×.Ò.¨t°T¸4ÀTĞ.ÑJÔJˆ
ÔØ !× /Ò /°¸¸QÑ ?Ô ?ˆ
ÔØ"'×"5Ò"5°lÀAÀqÑ"IÔ"Iˆ
Ô Ø ×-Ò-¨g°s¸AÑ>Ô>ˆ
ŒØ ×.Ò.¨y¸$ÀÑEÔEˆ
Ôà #× 1Ò 1°'¸3ÀÑ BÔ Bˆ
ÔØ "× 0Ò 0°¸¸aÑ @Ô @ˆ
ÔØ%*×%>Ò%>¸|ÈaĞQTÈXÑ%VÔ%Vˆ
Ô"Ø#(×#4Ò#4°\À1ÀbÑ#IÔ#Iˆ
Ô Ø ×+Ò+¨C°°BÑ7Ô7ˆ
ŒØ $× 0Ò 0°¸3ÀÑ DÔ Dˆ
Ô à(+ˆ
Ô%Ø"#ˆ
Ô å”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆİ # D×$8Ò$8Ñ$:Ô$:Ñ ;Ô ;ˆÔ Ø×,Ò,¨]Ñ;Ô;×>Ò>¸vÑFÔFˆØ Ğ × 8Ò 8Ñ :Ô :¸MÑJÔJˆØ �Y˜t× 7Ò 7Ñ 9Ô 9ĞsĞsÈĞmqĞ=rĞ=rĞsĞsˆ
Ø Ğ !Ø�G˜D× 4Ò 4Ñ 6Ô 6¸
ÀvÑNÔNˆEˆEà"ˆEØ Ğ "Ø�)˜MÔ 9ÀĞPUØ%2Ô%=ğ?ñ ?ô ?ˆCˆCğ "ˆCØ—.’. °×0HÒ0HÑ0JÔ0JÈGĞUYĞ[eĞptØ05ÈEĞcfğ !ñ hô hˆà�
Š
‰ŒˆØ×5Ò5Ñ7Ô7ˆØ
� Š �T˜=Ô2Ñ3Ô3Ğ3İ
ĞD�u”|ĞDĞD¸TĞDĞDĞDÑEÔEĞEØ Ø Ø Ø Øˆ r r )r· )r¶ )rµ )r¹ )r¸ )r´ F)rp rs z,Storage can't be None if in_memory is False.)rp ro )rn z/study_params.jsonÚw)ÚoptunaÚTrialÚ#mu_alpha_zero.MuZero.MZ_Arena.arenar· Ú+mu_alpha_zero.MuZero.MZ_MCTS.mz_search_treer¶ Ú%mu_alpha_zero.MuZero.Network.networksrµ Úmu_alpha_zero.trainerr¹ Ú%mu_alpha_zero.AlphaZero.Arena.playersr¸ Úmu_alpha_zero.mem_bufferr´ Ú show_tqdmÚ create_studyÚ
ValueErrorÚ
load_studyÚoptimizeÚopenÚcheckpoint_dirÚjsonÚdumpÚ best_params)rn ro rp rº rq rr rs r� r‘ r» ÚstudyÚfiler´ rµ r¶ r· r¸ r¹ s `` `` @@@@@@r
Úmz_optuna_parameter_searchrÑ h sÈ øøøøøøøøøø€ ğ*�œ ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğ *ğX <Ğ;Ğ;Ğ;Ğ;Ğ;ØLĞLĞLĞLĞLĞLØ?Ğ?Ğ?Ğ?Ğ?Ğ?Ø-Ğ-Ğ-Ğ-Ğ-Ğ-Ø?Ğ?Ğ?Ğ?Ğ?Ğ?Ø2Ğ2Ğ2Ğ2Ğ2Ğ2à #€MÔØğ JİÔ#¨zÀYĞOÑOÔOˆˆà ˆ?İ Ğ KÑLÔLĞ LİÔ!¨ZÀĞIÑIÔIˆØ ‡N‚N�9 x€NÑ0Ô0Ğ0İ
� Ô-ĞAĞAĞAÀ3Ñ GÔ Gğ +È4İ Œ �%Ô# TÑ*Ô*Ğ*ğ+ğ +ğ +ñ +ô +ğ +ğ +ğ +ğ +ğ +ğ +ğ +øøøğ +ğ +ğ +ğ +ğ +ğ +s à C(Ã(C,Ã/C,Úinvalid_actionsÚpic óÔ — t j | ¦ « dk rt d¦ « | d¦ « | d¦ « z }|| ¦ « z S )Nr zNo valid actions left.r: )r* rj r¬ Úreshape)rÒ rÓ s r
Úmask_invalid_actionsrÖ ª sZ € İ „vˆoÑ Ô !Ò#Ğ#İ
Ğ&Ñ'Ô'Ğ'Ø �Š�B‰ Œ ˜/×1Ò1°"Ñ5Ô5Ñ 5€BØ
�—’‘”‰=Ğr Úget_invalid_actionsÚpisÚplayersc ó� — t j |j ¦ « }t |¦ « D ] \ }} | || |¦ « }|||< Œ |S r( )r Úemptyr Ú enumerate)r× rØ rÙ Úinvalid_actions_tsÚiÚplayerÚinvaid_actionss r
Úmask_invalid_actions_batchrá ± sY € İœ #¤)Ñ,Ô,Ğİ ˜wÑ'Ô'ğ /ğ /‰ ˆˆ6Ø,Ğ,¨S°¬V°VÑ<Ô<ˆØ .И1Ñ Ğ Ø Ğ r )r )rH )Frm NN)$rÌ rW Útypingr Únumpyr* r½ Útorchr ÚPILr Úmu_alpha_zero.configr ÚTensorr r r Úlistr# ÚndarrayÚtupleÚboolr- r3 r7 rG ÚfloatrQ rS rY rg rl ÚstrrÑ rÖ rá r2 r r
ú<module>rî s ğØ € € € Ø € € € Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ Ø
€
€
€
Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à -Ğ -Ğ -Ğ -Ğ -Ğ -ğ4ğ 4 R¤Yğ 4¸¼ğ 4ğ 4ğ 4ğ 4ğ4¨¬ ğ 4¸3ğ 4ğ 4ğ 4ğ 4ğ A°2´9ğ AÈDĞQTÌIğ Ağ Ağ Ağ Ağ˜RœZğ ¨u°S¸#°X¬ğ Èğ ĞQSÔQ[ğ ğ ğ ğ ğ�r”zğ ¨$ğ ğ ğ ğ ğ&˜ğ &¨3ğ &ğ &ğ &ğ &ğ
E R¤Yğ
Eğ
Eğ
Eğ
EğIğ I˜bœiğ I¨Eğ Iğ Iğ Iğ Iğğ R¤Yğ °5ğ ğ ğ ğ ğ#˜ğ #ğ #ğ #ğ #ğ˜œğ °#ğ ğ ğ ğ ğ" ˜œğ °#ğ ğ ğ ğ ğ gqØDHğ?+ğ ?+¨ğ ?+°s°{¸dğ ?+ĞPSğ ?+Ø.:ğ?+ØGKğ?+Ø`cğ?+ğ ?+ğ ?+ğ ?+ğD¨"¬*ğ ¸"¼*ğ ğ ğ ğ ğ °Hğ À2Ä9ğ ĞW[Ğ\_ÔW`ğ ğ ğ ğ ğ ğ r | 13,692 | Python | .pyt | 53 | 257.226415 | 2,296 | 0.333675 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,574 | pickler.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/__pycache__/pickler.cpython-311.pyc | §
føf,
ã ó< — d dl Z d dlZd dlmZ G d„ d¦ « ZdS )é N)ÚLockc ób — e Zd ZdZdefd„Zdedefd„Zdefd„Zde fd „Z
dd e d ed
e fd„Z d„ Z dS )Ú DataPicklerzQ
Class for efficiently storing and retrieving data from the file system.
Ú
pickle_dirc óH — d| _ | |¦ « | _ d S )Nr )Úprocessed_countÚ _DataPickler__init_pickle_dirr ©Úselfr s úK/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/pickler.pyÚ__init__zDataPickler.__init__ s# € Ø ˆÔ Ø ×0Ò0°Ñ<Ô<ˆŒˆˆó Úreturnc ó2 — t j |d¬¦ « |S )NT)Úexist_ok)ÚosÚmakedirsr
s r Ú__init_pickle_dirz DataPickler.__init_pickle_dir s € İ
Œ �J¨Ğ.Ñ.Ô.Ğ.ØĞr Úbufferc ó– ‡— t t |d ¦ « ¦ « D ]”Šˆfd„|D ¦ « }t | j › d‰› d�dd¬¦ « 5 }t j ||¦ « | ¦ « t j | ¦ « ¦ « d d d ¦ « n # 1 swxY w Y Œ•| xj
dz
c_
d S ) Nr c ó •— g | ]
}|‰ ‘Œ S © r )Ú.0ÚitemÚindexs €r ú
<listcomp>z-DataPickler.pickle_buffer.<locals>.<listcomp> s ø€ Ğ3Ğ3Ğ3 D�D˜”KĞ3Ğ3Ğ3r ú/item_ú.pklÚabéZ ©Útimeouté ) ÚrangeÚlenr r ÚpickleÚdumpÚflushr ÚfsyncÚfilenor )r r ÚdataÚfr s @r Ú
pickle_bufferzDataPickler.pickle_buffer s ø€ İ�3˜v aœy™>œ>Ñ*Ô*ğ %ğ %ˆEØ3Ğ3Ğ3Ğ3¨FĞ3Ñ3Ô3ˆDݘœ Ğ;Ğ;°Ğ;Ğ;Ğ;¸TÈ2ĞNÑNÔNğ
%ĞRSİ” ˜D !Ñ$Ô$Ğ$Ø—’‘ ” � İ”˜Ÿš™œÑ$Ô$Ğ$ğ
%ğ
%ğ
%ñ
%ô
%ğ
%ğ
%ğ
%ğ
%ğ
%ğ
%øøøğ
%ğ
%ğ
%ğ
%øğ
Ğ Ô Ñ!Ğ Ô Ğ Ğ s ÁAB-Â-B1 Â4B1 r c óğ — | j › d|› d�}g }t |dd¬¦ « 5 } | t j |¦ « ¦ « n# t
$ r Y nw xY wŒ: d d d ¦ « n # 1 swxY w Y |S )Nr r Úrbr r! )r r Úextendr& ÚloadÚEOFError)r r Úfiler+ r, s r Ú
load_indexzDataPickler.load_index sŞ € Ø”/Ğ4Ğ4¨Ğ4Ğ4Ğ4ˆØˆİ
�$˜ bĞ
)Ñ
)Ô
)ğ ¨Qğ
ğØ—K’K¥¤ ¨A¡¤Ñ/Ô/Ğ/Ğ/øİ ğ ğ ğ Ø�Eğøøøğ
ğ ğ ğ ğ ñ ô ğ ğ ğ ğ ğ ğ øøøğ ğ ğ ğ ğ ˆ s4 ¢A+¥'A
Á A+Á
AÁA+ÁAÁA+Á+A/Á2A/r# Ú
batch_sizeÚindexesÚKc ó6 — d„ t j | j ¦ « D ¦ « }| d„ ¬¦ « d„ t t |¦ « ¦ « D ¦ « }t
|¦ « D �]\ }}t | j › d|› �dd¬¦ « 5 }d } t || ¦ « |k rn¿ t j |¦ « }
|D ]ƒ} | t |
¦ « z | cxk r| k rdn Œ"| t |
¦ « z |z
| k r| || t |
¦ « z | z
z
z } | | z } |
| | |z … } ||
| ¦ « Œ„| t |
¦ « z
} n# t $ r Y nw xY wŒÙd d d ¦ « n # 1 swxY w Y �Œd „ t |� D ¦ « S ) Nc ó< — g | ]}| d ¦ « ¯|‘ŒS ©r ©Úendswith©r Úxs r r z(DataPickler.load_all.<locals>.<listcomp>* ó) € ĞNĞNĞN�q¸1¿:º:ÀfÑ;MÔ;MĞN�ĞNĞNĞNr c ó„ — t | d¦ « d d¦ « d ¦ « S )NÚ_r# ú.r )ÚintÚsplit)r> s r ú<lambda>z&DataPickler.load_all.<locals>.<lambda>+ s/ € ¥ Q§W¢W¨S¡\¤\°!¤_×%:Ò%:¸3Ñ%?Ô%?ÀÔ%BÑ!CÔ!C€ r )Úkeyc ó — g | ]}g ‘ŒS r r )r rA s r r z(DataPickler.load_all.<locals>.<listcomp>, s € Ğ.Ğ.Ğ.�q�Ğ.Ğ.Ğ.r ú/r/ r r! r Tc óŒ — g | ]A}|d |d |d d |d d |d d f|d f‘ŒBS )r r# é é r r= s r r z(DataPickler.load_all.<locals>.<listcomp>@ sL € ĞTĞTĞTÀA��1”�q˜”t˜a œd 1œg q¨¤t¨A¤w°°!´°Q´Ğ 8¸!¸A¼$Ğ?ĞTĞTĞTr )
r Úlistdirr Úsortr$ r% Ú enumerater r& r1 r0 r2 Úzip)
r r5 r6 r7 Úfilesr+ Úir3 r, Úlast_lenÚ file_datar Ú
data_points
r Úload_allzDataPickler.load_all) s € ØNĞN�BœJ t¤Ñ 7Ô 7ĞNÑNÔNˆØ
�
Š
ĞCĞCˆ
ÑDÔDĞDØ.Ğ.�E¥# e¡*¤*Ñ -Ô -Ğ.Ñ.Ô.ˆİ Ñ'Ô'ğ ñ ‰GˆAˆtݘœ Ğ1Ğ1¨4Ğ1Ğ1°4ÀĞDÑDÔDğ
ÈØ �ğ ݘ4 œ7‘|”| zÒ1Ğ1Ø ğ İ$*¤K°¡N¤N˜ Ø%,ğ ;ğ ;˜EØ '#¨i©.¬.Ñ 8¸5Ğ LĞ LÒ LĞ LÀHÒ LĞ LĞ LĞ LĞ LØ#+c°)©n¬nÑ#<¸qÑ#@À5Ò#HĞ#HØ$)¨Q°8½cÀ)¹n¼nÑ3LĞPUÑ2UÑ-VÑ$V EØ %¨Ñ 1 Ø-6°u¸UÀQ¹Y°Ô-G
Ø $ Q¤§¢¨zÑ :Ô :Ğ :øà ¥C¨ ¡N¤NÑ2˜˜øİ#ğ ğ ğ Ø ˜ğ øøøğ ğ
ğ
ğ
ñ
ô
ğ
ğ
ğ
ğ
ğ
ğ
øøøğ
ğ
ğ
ğ
ùğ$ UĞTÍÈdÈĞTÑTÔTĞTs7 Â E;Â1B,E Å E;Å
E+Å(E;Å*E+Å+E;Å;E? ÆE? c ó — d„ t j | j ¦ « D ¦ « }|D ] }t j | j › d|› �¦ « Œ!d| _ d S )Nc ó< — g | ]}| d ¦ « ¯|‘ŒS r: r; r= s r r z)DataPickler.clear_dir.<locals>.<listcomp>C r? r rH r )r rL r Úremover )r rP r3 s r Ú clear_dirzDataPickler.clear_dirB sa € ØNĞN�BœJ t¤Ñ 7Ô 7ĞNÑNÔNˆØğ 3ğ 3ˆDİ ŒI˜œ Ğ1Ğ1¨4Ğ1Ğ1Ñ 2Ô 2Ğ 2Ğ 2Ø ˆÔ Ğ Ğ r N)r# )
Ú__name__Ú
__module__Ú __qualname__Ú__doc__Ústrr
r Úlistr- rC r4 rU rY r r r r r sÖ € € € € € ğğ ğ= 3ğ =ğ =ğ =ğ =ğ¨Cğ °Cğ ğ ğ ğ ğ" Dğ "ğ "ğ "ğ "ğ
ğ
ğ
ğ
ğ
ğUğ U 3ğ U°ğ U¸#ğ Uğ Uğ Uğ Uğ2!ğ !ğ !ğ !ğ !r r )r r& Ú portalockerr r r r r ú<module>ra s[ ğØ € € € Ø
€
€
€
à Ğ Ğ Ğ Ğ Ğ ğ?!ğ ?!ğ ?!ğ ?!ğ ?!ñ ?!ô ?!ğ ?!ğ ?!ğ ?!r | 6,669 | Python | .pyt | 39 | 169.666667 | 1,389 | 0.31398 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,575 | mz_search_tree.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_MCTS/__pycache__/mz_search_tree.cpython-311.pyc | §
|Ú£fX3 ã óf — d dl Z d dlZd dlZd dlZd dlmZ edd¬¦ « d dlmZ d dlZ d dl
Z d dl m
Z
d dlmZ d d lmZ d d
lmZ d d lmZ d d lmZ d d
lmZ d dlmZ d dl m Z m Z m Z m Z m!Z!m"Z" d dl#m$Z$ d dl%m&Z&m'Z'm(Z( d dl)m*Z* G d„ de¦ « Z+d de,pdfd„Z- d de$de.de*de.de,pdf
d„Z/dS ) é N)Úset_start_methodÚspawnT)Úforce)ÚPool)ÚGeneralMemoryBuffer)Ú
MuZeroGame)Ú
SearchTree)Ú HookManager)ÚHookAt)ÚMzAlphaZeroNode)Ú MuZeroNet)Ú LazyArray)Úmatch_action_with_obsÚ
resize_obsÚ scale_actionÚ scale_stateÚscale_hidden_stateÚmask_invalid_actions)Ú MuZeroConfig)ÚMuZeroFrameBufferÚSingleGameDataÚ DataPoint)Ú
SharedStoragec óª — e Zd Zd'dededepdfd„Zd„ Z d(ded e j
d
e pdd e d e
e f
d
„Z d(dej depdd e j
depdd e f
d„Zd„ Zded e j
deded eeeef f
d„Zd„ Zdepdfd„Zd„ Zde fd „Z e d e
d e
ded e j
ded ef d „¦ « Z e d e
d e
d!e d e j
d"ed ed#efd$„¦ « Z!e d!e d"efd%„¦ « Z"d&„ Z#dS ))ÚMuZeroSearchTreeNÚ game_managerÚ
muzero_configÚ hook_managerc óÊ — || _ || _ |�|n
t ¦ « | _ | ¦ « | _ t
d¦ « t
d¦ « g| _ d S )NÚinf)r r r
r Úinit_frame_bufferÚbufferÚfloatÚ min_max_q)Úselfr r r s úZ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/MZ_MCTS/mz_search_tree.pyÚ__init__zMuZeroSearchTree.__init__ sX € Ø (ˆÔØ *ˆÔØ,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ×,Ò,Ñ.Ô.ˆŒ İ ™,œ,¨u© ¬ ¨
Ğ6ˆŒˆˆó c óö — | j j r<t | j j | j ¦ « | j j ¦ « S t d| j ¦ « | j j ¦ « S ©Né )r Úenable_frame_bufferr Úframe_buffer_sizer Úget_noopÚnet_action_size©r% s r& r! z"MuZeroSearchTree.init_frame_buffer% so € Ø Ô Ô 1ğ Iİ$ TÔ%7Ô%IÈ4ÔK\×KeÒKeÑKgÔKgØ%)Ô%7Ô%GñIô Iğ
Iå DÔ$5×$>Ò$>Ñ$@Ô$@À$ÔBTÔBdÑeÔeĞer( FÚnetwork_wrapperÚdeviceÚdir_pathÚcalculate_avg_num_childrenÚreturnc ó — | j j }| j j }| j ¦ « }t || j j | j j ¦ « }t || j j ¦ « }d}| j
||¦ « | j j r6| j
| j || ¦ « | ¦ « t ¦ « } d}
t |¦ « D �]Ì} |
dz
}
| |||||o| dk ¬¦ « \ } \ }
}| j | | j j ¬¦ « }| j |||¬¦ « \ }}}t || j j | j j ¦ « }t || j j ¦ « }t' || j ¦ « ¦ « }| j |¦ « ¦ « ¦ « ¦ « }t3 | |
||||€|nt5 ||¦ « ¦ « }| |¦ « |r nd| j j r| }| j |||¦ « | j | j || ¦ « || ¦ « �ŒÎ t; j d|
i¦ « n# t> $ r Y nw xY w| | j ¦ « | gS )Nr+ r ©r4 ©Útau)Ú
frame_skipz Game length)!r Ú num_stepsr: r Úresetr Útarget_resolutionÚ
resize_imagesr r" Ú init_bufferÚmultiple_playersÚ get_state_for_passive_playerr ÚrangeÚsearchÚ select_mover9 Úframe_skip_stepr Úget_num_actionsÚ
concat_framesÚdetachÚcpuÚnumpyr r Úadd_data_pointÚ add_frameÚwandbÚlogÚ ExceptionÚcompute_initial_priorities)r% r1 r2 r3 r4 r; r: ÚstateÚplayerÚdataÚ game_lengthÚstepÚpiÚvÚlatentÚmoveÚrewÚdoneÚframeÚ
data_points r& Ú
play_one_gamez MuZeroSearchTree.play_one_game+ s € àÔ&Ô0ˆ ØÔ'Ô2ˆ
ØÔ!×'Ò'Ñ)Ô)ˆİ˜5 $Ô"4Ô"FÈÔHZÔHhÑiÔiˆİ˜E 4Ô#5Ô#AÑBÔBˆØˆØ Œ × Ò vÑ.Ô.Ğ.Ø Ô Ô .ğ mØ ŒK× #Ò # DÔ$5×$RÒ$RĞSXĞ[aĞZaÑ$bÔ$bĞekĞdkÑ lÔ lĞ lİ Ñ Ô ˆØˆ ݘ)Ñ$Ô$ğ qñ qˆDØ ˜1Ñ ˆKØ "Ÿkšk¨/¸5À&È&Ø.Ğ<°4¸1²9ğ *ñ ?ô ?‰OˆB‘ ��FàÔ$×0Ò0°¸Ô9KÔ9OĞ0ÑPÔPˆDà #Ô 0× @Ò @ÀÀvĞZdĞ @Ñ eÔ eÑ ˆE�3˜İ ˜u dÔ&8Ô&JÈDÔL^ÔLlÑmÔmˆEİ tÔ'9Ô'EÑFÔFˆEİ dÔ&7×&GÒ&GÑ&IÔ&IÑJÔJˆDØ”K×-Ò-¨fÑ5Ô5×<Ò<Ñ>Ô>×BÒBÑDÔD×JÒJÑLÔLˆEİ" 2 q¨#¨t°VÀhĞFV¸U¸UÕ\eĞfkĞmuÑ\vÔ\vÑwÔwˆJğ
× Ò
Ñ +Ô +Ğ +Øğ
Ø�ØÔ!Ô2ğ
!Ø ˜�Ø ŒK× !Ò ! %¨¨vÑ 6Ô 6Ğ 6Ø ŒK× !Ò ! $Ô"3×"PÒ"PĞQVĞY_ĞX_Ñ"`Ô"`ĞbfĞioĞhoÑ pÔ pĞ pÑ pğ İ ŒI�} kĞ2Ñ 3Ô 3Ğ 3Ğ 3øİğ ğ ğ Ø ˆDğ øøøà ×'Ò'¨Ô(:Ñ;Ô;Ğ;؈vˆ
s Ê>K Ë
K"Ë!K"rQ Úcurrent_playerr9 c óZ — | j |¦ « dk r| j ||¦ « | j j }|€ | j j }t
|¬¦ « }| | j |¦ « ddd¦ «
d¦ « ¦ « d¦ « } t | ¦ « } |
|
d¦ « d¬¦ « \ }
} | j j dk r:|
t j | j j g| j j z ¦ « z }
t' | j ||¦ « |
¦ « }
|
¦ « ¦ « }
| | |
d¦ « t3 |¦ « D �]ç} |}
|
g}d }|
¦ « r…|
| j d | j d | j j | j j | j j | j j ¬¦ « \ }
}| ! |
¦ « |
¦ « °…tE || j # ¦ « ¦ « }tI |
% ¦ « j& |¦ « }| ' |
d¦ « d¬¦ « \ }}t |¦ « }|d d }|
|
d¦ « d¬¦ « \ }
} |
¦ « ¦ « }
| ¦ « ¦ « d } |
||
|¦ « | ( | |¦ « �Œé| ) ¦ « }| * ¦ « | + ¦ « f}| j, - | | j. j/ t` tb j2 |||f¬¦ « d }||fS ) Nr )r_ é r+ T)Úpredict)ÚcÚc2©Úargs)3r" Ú__len__r? r Únum_simulationsr9 r Úrepresentation_forwardrG ÚpermuteÚ unsqueezeÚsqueezer Úprediction_forwardÚdirichlet_alphaÚnpÚrandomÚ dirichletr/ r r Úget_invalid_actionsÚflattenÚtolistÚ expand_noderB Ú was_visitedÚget_best_childr$ Úgammar@ rc rd Úappendr rF r ÚparentrQ Údynamics_forwardÚbackpropÚ get_self_action_probabilitiesÚget_self_valueÚ
get_latentr Úprocess_hook_executesrC Ú__name__Ú__file__r ÚTAIL)r% r1 rQ r_ r2 r9 r4 rh Ú root_nodeÚstate_rV rW Ú
simulationÚ current_nodeÚpathÚactionÚ current_node_state_with_actionÚ
next_stateÚrewardÚ action_probsÚroot_val_latents r& rC zMuZeroSearchTree.searchV sñ € à Œ;× Ò ˜~Ñ .Ô .°!Ò 3Ğ 3Ø ŒK× #Ò # E¨>Ñ :Ô :Ğ :Ø Ô,Ô<ˆØ ˆ;ØÔ$Ô(ˆCå#°>ĞBÑBÔBˆ à ×7Ò7Ø ŒK× %Ò % nÑ 5Ô 5× =Ò =¸aÀÀAÑ FÔ F× PÒ PĞQRÑ SÔ SñUô UßU\ÒU\Ğ]^ÑU_ÔU_ğ å# FÑ+Ô+ˆØ ×2Ò2°6×3CÒ3CÀAÑ3FÔ3FĞPTĞ2ÑUÔU‰ˆˆAØ Ô Ô -°Ò 1Ğ 1Ø•b”i×)Ò)¨4Ô+=Ô+MĞ*NĞQUÔQcÔQsÑ*sÑtÔtÑtˆBİ
! $Ô"3×"GÒ"GÈÈ~Ñ"^Ô"^Ğ`bÑ
cÔ
cˆØ
�ZŠZ‰\Œ\×
Ò
Ñ
"Ô
"ˆØ× Ò ˜f b¨!Ñ,Ô,Ğ,İ Ñ0Ô0ğ #ñ #ˆJØ$ˆLØ �>ˆD؈FØ ×*Ò*Ñ,Ô,ğ
*Ø'3×'BÒ'BÀ4Ä>ĞRSÔCTĞVZÔVdĞefÔVgØCGÔCUÔC[ØCGÔCUÔCfØEIÔEWÔEYĞ^bÔ^pÔ^sğ (Cñ (uô (uÑ$� ˜fğ — ’ ˜LÑ)Ô)Ğ)ğ ×*Ò*Ñ,Ô,ğ
*õ " &¨$Ô*;×*KÒ*KÑ*MÔ*MÑNÔNˆFå-BÀ<×CVÒCVÑCXÔCXÔC^Ğ`fÑ-gÔ-gĞ *Ø!0×!AÒ!AĞB`×BjÒBjĞklÑBmÔBmØJNğ "Bñ "Pô "PÑ ˆJ˜å+¨JÑ7Ô7ˆJؘA”Y˜q”\ˆFØ#×6Ò6°z×7KÒ7KÈAÑ7NÔ7NĞX\Ğ6Ñ]Ô]‰EˆB�Ø—’‘ ” ×$Ò$Ñ&Ô&ˆBØ— ’ ‘ ” ×"Ò"Ñ$Ô$ QÔ'ˆAØ × $Ò $ Z°°VÑ <Ô <Ğ <Ø �MŠM˜!˜TÑ "Ô "Ğ "Ñ "à ×>Ò>Ñ@Ô@ˆ Ø$×3Ò3Ñ5Ô5°y×7KÒ7KÑ7MÔ7MĞNˆØ Ô×/Ò/°°d´kÔ6JÍHÕV\ÔVaØ6BÀOĞU^Ğ5_ğ 0ñ aô ağ ağ
ˆ ؘ_Ğ,Ğ,r( c ó² — |}| j j }t |¦ « D ]¸}| j j rN|xj |z
c_ | |j | ¦ « z
¦ « |j || z z }nL|xj |z
c_ | |j | ¦ « z ¦ « |j ||z z }|xj dz
c_ Œ¹d S r* ) r rx Úreversedr@ Ú total_valueÚupdate_min_max_qrŒ r~ Ú
times_visited)r% rW rˆ ÚG_noderx Únodes r& r| zMuZeroSearchTree.backprop‹ só € àˆØÔ"Ô(ˆİ ˜T‘N”Nğ
$ğ
$ˆDàÔ!Ô2ğ
6ğ Ğ Ô FÑ*Ğ Ô Ø×%Ò% d¤k°D×4GÒ4GÑ4IÔ4IÑ&IÑJÔJĞJØ œ u°°Ñ'8Ñ8��àĞ Ô FÑ*Ğ Ô Ø×%Ò% d¤k°D×4GÒ4GÑ4IÔ4IÑ&IÑJÔJĞJØ œ u¨v¡~Ñ5�à Ğ Ô !Ñ #Ğ Ô Ğ ğ
$ğ
$r( ÚnetÚ num_gamesÚmemoryc ó’ — t |¦ « D ]6}| ||||dz
k ¬¦ « }| |¦ « Œ7dS )Nr+ r7 ©NNN)rB r^ Úadd_list)r% r– r2 r— r˜ ÚgameÚ game_resultss r& Ú self_playzMuZeroSearchTree.self_playŸ s[ € å˜)Ñ$Ô$ğ *ğ *ˆDØ ×-Ò-¨c°6ĞVZĞ^gĞjkÑ^kÒVkĞ-ÑlÔlˆLØ �OŠO˜LÑ )Ô )Ğ )Ğ )à Ğ r( c ó° — t | j ¦ « t j | j ¦ « t j | j ¦ « ¬¦ « S )N)r )r r Úmake_fresh_instanceÚcopyÚdeepcopyr r r0 s r& r z$MuZeroSearchTree.make_fresh_instance§ sN € İ Ô 1× EÒ EÑ GÔ GÍÌ ĞW[ÔWiÑIjÔIjİ-1¬]¸4Ô;LÑ-MÔ-MğOñ Oô Oğ Or( r‰ c ó — d S ©N© )r% r‰ s r& Ú step_rootzMuZeroSearchTree.step_root« s € à ˆr( c ó’ — t | j d |¦ « | j d< t | j d |¦ « | j d< d S ©Nr r+ )Úminr$ Úmax)r% Úqs r& r’ z!MuZeroSearchTree.update_min_max_q¯ sB € İ ¤¨qÔ 1°1Ñ 5Ô 5ˆŒ�qÑİ ¤¨qÔ 1°1Ñ 5Ô 5ˆŒ�qÑĞĞr( r„ c óª — d}t |j ¦ « dk rd}|j ¦ « D ] }| |¦ « }||z
}Œ |S r¨ )ÚlenÚchildrenÚvaluesÚ
get_num_nodes)r% r„ Ú num_nodesÚchildÚnns r& r° z MuZeroSearchTree.get_num_nodes³ sd € ؈ İ ˆyÔ!Ñ "Ô " QÒ &Ğ &؈IØ Ô'×.Ò.Ñ0Ô0ğ ğ ˆEØ×#Ò# EÑ*Ô*ˆBØ ˜‰OˆIˆIàĞr( ÚnetsÚtreesÚnum_jobsc
óÊ ‡ ‡‡‡‡‡— t ‰¦ « 5 }‰j rN‰j rG| t ˆˆˆ ˆˆˆfd„t t
‰ ¦ « ¦ « D ¦ « ¦ « }nF| t ˆˆˆ ˆˆˆfd„t t
‰ ¦ « ¦ « D ¦ « ¦ « }d d d ¦ « n # 1 swxY w Y |D ]}‰ |¦ « ŒdS )Nc ó† •— g | ]=}‰| ‰| t j ‰¦ « ‰‰z t j ‰¦ « d f‘Œ>S r¤ )r¡ r¢ ©Ú.0Úir2 r˜ r´ r— r¶ rµ s €€€€€€r& ú
<listcomp>z7MuZeroSearchTree.parallel_self_play.<locals>.<listcomp>Â sX ø€ ğ 2)ğ 2)ğ 2)Øwx�T˜!”W˜e Aœh¬
°fÑ(=Ô(=¸yÈHÑ?TÕVZÔVcĞdjÑVkÔVkĞmqĞrğ2)ğ 2)ğ 2)r( c ól •— g | ]0}‰| ‰| t j ‰¦ « ‰‰z d ‰j f‘Œ1S r¤ )r¡ r¢ r3 r¹ s €€€€€€r& r¼ z7MuZeroSearchTree.parallel_self_play.<locals>.<listcomp>Æ sO ø€ ğ 2&ğ 2&ğ 2&Øqr�T˜!”W˜e Aœh¬
°fÑ(=Ô(=¸yÈHÑ?TĞVZĞ\bÔ\kĞlğ2&ğ 2&ğ 2&r( rš )r Úis_diskÚ full_diskÚstarmapÚ p_self_playrB r r› ) r´ rµ r˜ r2 r— r¶ ÚpÚresultsÚresults `````` r& Úparallel_self_playz#MuZeroSearchTree.parallel_self_play½ sp øøøøøø€ õ �(‰^Œ^ğ '˜qØŒ~ğ
' &Ô"2ğ
'ØŸ)š)¥Kğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)ğ 2)å �S ™YœYÑ'Ô'ğ2)ñ 2)ô 2)ñ *ô *��ğ Ÿ)š)¥Kğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&ğ 2&å�#˜d™)œ)Ñ$Ô$ğ2&ñ 2&ô 2&ñ 'ô '�ğ
'ğ 'ğ 'ñ 'ô 'ğ 'ğ 'ğ 'ğ 'ğ 'ğ 'øøøğ 'ğ 'ğ 'ğ 'ğ ğ $ğ $ˆFØ �OŠO˜FÑ #Ô #Ğ #Ğ #à Ğ s –B B>Â>CÃCÚshared_storageÚconfigÚnum_worker_itersc
óğ — t |¦ « }t |¦ « D ]V}| t | | || t j |¦ « ||||| ¦ « f¬¦ « ŒW|S )Nre )r rB Ú apply_asyncÚ
c_p_self_playr¡ r¢ Ú get_dir_path) r´ rµ rÆ r2 rÇ r¶ rÈ Úpoolr» s r& Ústart_continuous_self_playz+MuZeroSearchTree.start_continuous_self_playÎ sƒ € õ �H‰~Œ~ˆİ�x‘ ” ğ ğ ˆAØ × Ò �]Ø�Q”˜˜qœ¥4¤=°Ñ#8Ô#8¸&À!À^ĞUeØ ×+Ò+Ñ-Ô-ğ2/Ğ ñ
ô
ğ
ğ
ğ
ˆ r( c óì — dt dt fd„}| |¦ « } | ¦ « t | ¦ « ¦ « |j dz k rAt j d¦ « t | ¦ « ¦ « |j dz k °At |j
¦ « D ]µ} ||j dz |¦ « }|D ]œ\ }} \ }
} } }
}t |t ¦ « r|
¦ « }t j ||¬¦ « ¦ « }| | ||
|¦ « \ }\ } }|j ||j ¬¦ « } Œ�Œ¶d S ) NÚnÚmemc ó� — | ¦ « }g }t | ¦ « D ] }| || |f¦ « Œ |S r¤ )Ú
get_bufferrB ry )rĞ rÑ r" rS r» s r& Ú get_first_nz/MuZeroSearchTree.reanalyze.<locals>.get_first_nÜ sM € Ø—^’^Ñ%Ô%ˆF؈Dݘ1‘X”Xğ
,ğ
,�Ø— ’ ˜V AœY¨˜NÑ+Ô+Ğ+Ğ+؈Kr( é é ra )r2 r8 )Úintr ÚtoÚevalr rÓ Ú
batch_sizeÚtimeÚsleeprB rÈ Ú
isinstancer Ú
load_arrayÚthÚtensorr# rC r rD r9 )r– Útreer2 rÆ rÇ rÔ Úiter_rS rV rW rZ rY Úpred_vrR r\ rQ Ú_s r& Ú reanalyzezMuZeroSearchTree.reanalyzeÚ sŠ € ğ �3ğ ¥]ğ ğ ğ ğ ğ �fŠf�V‰nŒnˆØ �Š‰
Œ
ˆ
İ�.×+Ò+Ñ-Ô-Ñ.Ô.°Ô1BÀQÑ1FÒFĞFİ ŒJ�q‰MŒMˆMõ �.×+Ò+Ñ-Ô-Ñ.Ô.°Ô1BÀQÑ1FÒFĞFå˜6Ô2Ñ3Ô3ğ Iğ IˆEØ �;˜vÔ 0°1Ñ 4°nÑEÔEˆDØ=Ağ
Iğ
IÑ9��AÑ2˜˜T 6¨6°Eİ ˜e¥YÑ/Ô/ğ /Ø !× ,Ò ,Ñ .Ô .�Eİœ %°Ğ7Ñ7Ô7×=Ò=Ñ?Ô?�Ø !Ÿ[š[¨¨e°V¸VÑ DÔ D‘
�‘F�Q˜ØÔ(×4Ò4°R¸V¼ZĞ4ÑHÔH��ğ
Iğ Iğ Ir( c ór — | j | | j j t t
j ¦ « d S r¤ )r r€ Úrun_on_training_endr� r‚ r ÚALLr0 s r& rç z$MuZeroSearchTree.run_on_training_endğ s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr( r¤ )NF)$r� Ú
__module__Ú __qualname__r r r
r' r! r
rß r2 ÚstrÚboolÚlistr r^ ro Úndarrayr× r# rC r| r Útupler� r r¦ r’ r r° Ú staticmethodrÅ r rÎ rå rç r¥ r( r& r r s« € € € € € ğ7ğ 7 Zğ 7À ğ 7Ğ\gĞ\oĞkoğ 7ğ 7ğ 7ğ 7ğfğ fğ fğ dhØ9>ğ)ğ )¨Yğ )ÀÄ ğ )ĞUXĞU`Ğ\`ğ )Ø26ğ)ØCGÈÔCWğ)ğ )ğ )ğ )ğX NSğ3-ğ 3-¨R¬Zğ 3-ÈÈĞPTğ 3-Ğ^`Ô^gğ 3-Ø�M˜Tğ3-ØFJğ3-ğ 3-ğ 3-ğ 3-ğj$ğ $ğ $ğ( ˜Yğ °´ ğ Àcğ ĞSfğ ĞkpØ ˆS�#ˆ
ôlğ ğ ğ ğ ğOğ Oğ Oğ
tğ
ğ
ğ
ğ
ğ6ğ 6ğ 6ğ ğ ğ ğ ğ ğ ğ ğ ¨dğ Ğ<Oğ ĞY[ÔYbğ Ğorğ Ø%(ğ ğ ğ ñ „\ğ ğ ğ ¨ğ °dğ ÈMğ Ø+-¬9ğ Ø>Jğ ØVYğ Ømpğ ğ ğ ñ „\ğ ğ ğI°]ğ IÈLğ Iğ Iğ Iñ „\ğIğ*oğ oğ oğ oğ or( r r3 c óÈ — g }t |¦ « D ]O}| | ||||dz
k ¬¦ « }|�| |¦ « Œ:| |¦ « ŒP|S )Nr+ )r3 r4 )rB r^ r› Úextend) r– rá ÚdevÚnum_grÑ r3 rS rœ r� s r& rÁ rÁ ô sy € Ø
€Dİ�e‘ ” ğ &ğ &ˆØ×)Ò)¨#¨s¸XĞbfĞjoĞrsÑjsÒbsĞ)ÑtÔtˆ Ø ˆ?Ø �LŠL˜ Ñ &Ô &Ğ &Ğ &à �KŠK˜ Ñ %Ô %Ğ %Ğ %Ø €Kr( rÇ Úp_numrÆ rÈ c ó¼ — |dk rt j |j d¬¦ « | |¦ « } | ¦ « t |¦ « D ]�}| ¦ « €| ¦ « } n| ¦ « } | | ¦ « | | ||¬¦ « }
|
|
¦ « Œ‚d S )Nr z Self play)ÚprojectÚname)r3 ) rM ÚinitÚwandbd_project_namerØ rÙ rB Ú get_experimental_network_paramsÚget_stable_network_paramsÚload_state_dictr^ r› ) r– rá r2 rÇ rõ rÆ rÈ r3 râ Úparamsr� s r& rË rË ÿ sà € ğ
�‚z€zİ
Œ
˜6Ô5¸KĞHÑHÔHĞHØ
�&Š&�‰.Œ.€C؇H‚H�J„J€JİĞ'Ñ(Ô(ğ .ğ .ˆğ
× 9Ò 9Ñ ;Ô ;Ğ CØ#×=Ò=Ñ?Ô?ˆFˆFà#×CÒCÑEÔEˆFØ ×Ò˜FÑ#Ô#Ğ#Ø×)Ò)¨#¨vÀĞ)ÑIÔIˆ Ø× Ò Ñ-Ô-Ğ-Ğ-ğ .ğ .r( r¤ )0r¡ ÚgcrÛ rM Ú multiprocessr Úmultiprocess.poolr rJ ro Útorchrß Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.mz_gamer Ú!mu_alpha_zero.General.search_treer Ú mu_alpha_zero.Hooks.hook_managerr
Ú mu_alpha_zero.Hooks.hook_pointr Ú$mu_alpha_zero.MuZero.MZ_MCTS.mz_noder Ú%mu_alpha_zero.MuZero.Network.networksr
Ú mu_alpha_zero.MuZero.lazy_arraysr Úmu_alpha_zero.MuZero.utilsr r r r r r Úmu_alpha_zero.configr Úmu_alpha_zero.mem_bufferr r r Ú$mu_alpha_zero.shared_storage_managerr r rë rÁ r× rË r¥ r( r& ú<module>r sN ğØ € € € Ø € € € Ø € € € à € € € Ø )Ğ )Ğ )Ğ )Ğ )Ğ )à Ğ � Ğ %Ñ %Ô %Ğ %Ø "Ğ "Ğ "Ğ "Ğ "Ğ "à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø @Ğ @Ğ @Ğ @Ğ @Ğ @Ø ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø 6Ğ 6Ğ 6Ğ 6Ğ 6Ğ 6ğ-ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -à -Ğ -Ğ -Ğ -Ğ -Ğ -Ø QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QØ >Ğ >Ğ >Ğ >Ğ >Ğ >ğUoğ Uoğ Uoğ Uoğ Uo�zñ Uoô Uoğ Uoğpğ °c°k¸Tğ ğ ğ ğ ğ +/ğ.ğ .¨\ğ .À#ğ .ĞWdğ .Ø$'ğ.à ˜K 4ğ.ğ .ğ .ğ .ğ .ğ .r( | 20,530 | Python | .pyt | 75 | 272.533333 | 1,754 | 0.326408 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,576 | mz_node.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_MCTS/__pycache__/mz_node.cpython-311.pyc | §
¬Õ£f^ ã ó6 — d dl Z d dlmZ G d„ de¦ « ZdS )é N)Ú
AlphaZeroNodec óŒ ‡ — e Zd Zdˆ fd„ Zddeded ed
efd „Zd efd
„Zdd„Z dede
fd„Z ddeded ed
efd„Z d ed
edefd„Z
ˆ xZS )ÚMzAlphaZeroNoder Né c ó^ •— t ¦ « ||||¦ « d| _ d S )Nr )ÚsuperÚ__init__Úreward)ÚselfÚselect_probabilityÚparentÚtimes_visited_initÚcurrent_playerÚ __class__s €úS/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/MZ_MCTS/mz_node.pyr zMzAlphaZeroNode.__init__ s. ø€ İ
‰Œ×Ò˜ Ğ);¸VĞEWÑXÔXĞX؈Œ ˆ ˆ ó ç ø?éÄL Úmin_qÚmax_qÚgammaÚmultiple_playersc ó¾ — t d¦ « }d }d } | j ¦ « D ],\ }
} | ||||||¬¦ « } | |k r| }| }|
} Œ-|| fS )NÚinf)ÚcÚc2)ÚfloatÚchildrenÚitemsÚcalculate_utc_score)
r r r r r r r Úbest_utcÚ
best_childÚ best_actionÚactionÚchildÚ child_utcs
r Úget_best_childz MzAlphaZeroNode.get_best_child
sƒ € ݘ%‘L”L�=ˆØˆ
؈ Ø !œ]× 0Ò 0Ñ 2Ô 2ğ %ğ %‰MˆF�EØ ×1Ò1°%¸¸uĞEUĞYZĞ_aĞ1ÑbÔbˆIؘ8Ò#Ğ#Ø$�Ø "�
Ø $� øؘ;Ğ&Ğ&r Úprediction_forwardc ó" — || j ¦ « S ©N©Ústate)r r( s r Úget_value_predz MzAlphaZeroNode.get_value_pred s € Ø!Ğ! $¤*Ñ-Ô-Ğ-r Úreturnc ó¸ — | ¦ « | _ || _ t |¦ « D ])\ }}t || | j dz ¬¦ « }|| j |< Œ*d S )Néÿÿÿÿ)r r
r )Úcloner, r
Ú enumerater r r )r r, Úaction_probabilitiesÚ im_rewardr$ Ú probabilityÚnodes r Ú expand_nodezMzAlphaZeroNode.expand_node st € à—[’[‘]”]ˆŒ
Ø ˆŒ İ#,Ğ-AÑ#BÔ#Bğ )ğ )Ñ ˆF�Kİ"°kÈ$Ø26Ô2EÈÑ2LğNñ Nô NˆDà$(ˆDŒM˜&Ñ !Ğ !ğ )ğ )r Údynamics_forwardr$ c ó$ — || j |¦ « S r* r+ )r r8 r$ s r Úget_immediate_rewardz$MzAlphaZeroNode.get_immediate_reward" s € Ø Ğ ¤
¨FÑ3Ô3Ğ3r c ól — | ¦ « }|j dk r'|| j z t j |j dz ¦ « z S | ||||¦ « }|| j t j |j ¦ « d| j z z z |t j |j |z dz |z ¦ « z z z } | S )Nr g:Œ0â�yE>r )r
Ú
times_visitedr ÚmathÚsqrtÚscale_qÚlog)
r r r r r r r r
ÚqÚutcs
r r z#MzAlphaZeroNode.calculate_utc_score% s¿ € Ø—’‘ ” ˆØ Ô 1Ò $Ğ $Ø�tÔ.Ñ.µ´¸6Ô;OĞRVÑ;VÑ1WÔ1WÑWĞ WØ �LŠL˜ eĞ,<Ñ =Ô =ˆØ�$Ô)İ”˜6Ô/Ñ0Ô0°Q¸Ô9KÑ5KÑLñNà�$œ( FÔ$8¸2Ñ$=ÀÑ$AÀRÑ#GÑHÔHÑHñJñ Jˆğ ˆ
r c óÚ — | j |r| ¦ « n| ¦ « z }||k s&|t d¦ « k s|t d¦ « k r|S ||z
||z
z S )Nr z-inf)r
Úget_self_valuer )r r r r r rA s r r? zMzAlphaZeroNode.scale_q0 su € Ø ŒKĞ5EĞ`˜D× /Ò /Ñ 1Ô 1Ğ1Ğ1È4×K^ÒK^ÑK`ÔK`Ñ aˆØ �EŠ>ˆ>˜e¥u¨U¡|¤|Ò 3Ğ 3°uÅÀfÁ
Ä
Ò7MĞ7M؈HØ�E‘ ˜e e™mÑ,Ğ,r )r Nr r )r r )r. N)Ú__name__Ú
__module__Ú __qualname__r r Úboolr' Úcallabler- r7 Úintr: r r? Ú
__classcell__)r s @r r r s# ø€ € € € € ğğ ğ ğ ğ ğ ğ
'ğ
' Eğ
'°%ğ
'Àğ
'ĞX\ğ
'ğ
'ğ
'ğ
'ğ.°ğ .ğ .ğ .ğ .ğ)ğ )ğ )ğ )ğ4°Xğ 4Àsğ 4ğ 4ğ 4ğ 4ğ ğ ¨ğ °uğ ÀUğ Ğ]ağ ğ ğ ğ ğ-¨5ğ -ÀDğ -ÈUğ -ğ -ğ -ğ -ğ -ğ -ğ -ğ -r r )r= Ú$mu_alpha_zero.AlphaZero.MCTS.az_noder r © r r ú<module>rN sT ğØ € € € Ø >Ğ >Ğ >Ğ >Ğ >Ğ >ğ/-ğ /-ğ /-ğ /-ğ /-�mñ /-ô /-ğ /-ğ /-ğ /-r | 4,303 | Python | .pyt | 31 | 137.774194 | 655 | 0.357126 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,577 | java_manager.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/JavaGateway/__pycache__/java_manager.cpython-311.pyc | §
Ùxf¼ ã óˆ — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dlm Z m
Z
m Z d dl m
Z
d dlmZ d dlZ G d„ d¦ « ZdS )é N)Ú JavaGateway)Úget_python_homeÚfind_project_rootÚget_site_packages_path)Ú JavaNetworks)Ú MuZeroNetc óR — e Zd Zdedefd„Zdefd„Zd„ Zdedefd„Z d efd
„Z
d „ Z d S )
Ú JavaManagerÚnetworkÚargsc óH — t ¦ « t j d< t ¦ « | _ || _ |d | _ | |¦ « | _ | ¦ « | _
t ¦ « | _ t j | j ¦ « d S )NÚ
PYTHONHOMEÚenv_id)r ÚosÚenvironr Ú java_netsÚnetr Ú prepare_argsr Ú spawn_java_pÚ java_processr Ú java_gatewayÚatexitÚregisterÚkill_java_process)Úselfr r s ú\/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/JavaGateway/java_manager.pyÚ__init__zJavaManager.__init__ s� € õ $3Ñ#4Ô#4�Œ
�<Ñ å%™ œ ˆŒØˆŒØ˜8”nˆŒ Ø×%Ò% dÑ+Ô+ˆŒ Ø × -Ò -Ñ /Ô /ˆÔİ '™MœMˆÔİŒ˜Ô.Ñ/Ô/Ğ/Ğ/Ğ/ó c ó> — d„ | ¦ « D ¦ « S )Nc óÄ — i | ]]\ }}t |t j ¦ « r>t |t ¦ « s)|t d ¦ « k ¯G|t d¦ « k ¯Z||“Œ^S )Úinfz-inf)Ú
isinstanceÚnumbersÚNumberÚboolÚfloat)Ú.0ÚkÚvs r ú
<dictcomp>z,JavaManager.prepare_args.<locals>.<dictcomp> s| € ğ xğ xğ x™˜˜Aݘ1�gœnÑ-Ô-ğxİ6@ÀÅDÑ6IÔ6IğxØNOÕSXĞY^ÑS_ÔS_ÒN_ĞN_ĞdeÕinĞouÑivÔivÒdvĞdvğ �1ØdvĞdvĞdvr )Úitems)r r s r r zJavaManager.prepare_args s0 € ğxğ x §¢¡ ¤ ğ xñ xô xğ xr c óR — t j ddddt ¦ « › d�g¦ « S )NÚjavaz-jarzH/home/skyr/IdeaProjects/JSelfPlay/target/JSeflplayLinux-1.0-SNAPSHOT.jarz-Djava.library.path=z/jep/)Ú
subprocessÚPopenr ©r s r r zJavaManager.spawn_java_p s= € İÔ Ø
�VĞ gØ
CÕ$:Ñ$<Ô$<Ğ
CĞ
CĞ
Cğ
EñFô Fğ Fr Ún_jobsÚn_gamesc óT — g }| j | j ¦ « }t j t ¦ « › d�d¬¦ « | j j |||| | j
¦ « | j ¦ « }t d„ |D ¦ « ¦ « }g }|D ])}|
t j |¦ « ¦ « Œ*t j |d¬¦ « }t# |¦ « D �]\ } }
|
¦ « } d„ | ¦ « D ¦ « } t) |
¦ « ¦ « }
t) |
¦ « ¦ « ¦ « t) |
¦ « ¦ « ¦ « t) |
¦ « ¦ « ¦ « f}|
| |
||| f¦ « �Œt/ j t ¦ « › d�¦ « |S )Nz/Arrays/T)Úexist_okc ó6 — g | ]}| ¦ « ‘ŒS © )Ú getValue3)r' Úxs r ú
<listcomp>z;JavaManager.run_parallel_java_self_play.<locals>.<listcomp>+ s € Ğ4Ğ4Ğ4¨1˜Ÿš™ œ Ğ4Ğ4Ğ4r r )Úaxisc ó� — i | ]C}t | ¦ « ¦ « t | ¦ « ¦ « “ŒDS r6 )ÚintÚgetKeyr& ÚgetValue)r' Úentrys r r* z;JavaManager.run_parallel_java_self_play.<locals>.<dictcomp>2 s< € Ğ\Ğ\Ğ\À5•#�e—l’l‘n”nÑ%Ô%¥u¨U¯^ª^Ñ-=Ô-=Ñ'>Ô'>Ğ\Ğ\Ğ\r )r Úsaver r Úmakedirsr r Ú entry_pointÚrunParallelSelfPlayÚdict_to_java_mapr r ÚsetÚappendÚnpÚloadÚ concatenateÚ enumerateÚ getValue0ÚentrySetr& Ú getValue1Ú getValue2ÚshutilÚrmtree)r r1 r2 ÚresultsÚpathÚresÚ arr_pathsÚarrsÚarrÚiÚquartedÚhmapÚpdr) Úrmpred_rs r Úrun_parallel_java_self_playz'JavaManager.run_parallel_java_self_play% s÷ € ؈،~×"Ò" 4¤8Ñ,Ô,ˆİ
Œ Õ(Ñ*Ô*Ğ4Ğ4Ğ4¸tĞDÑDÔDĞDØÔ Ô+×?Ò?ÀÈĞQUĞW[×WlÒWlĞmqÔmvÑWwÔWwØ@DÄ ñMô MˆåĞ4Ğ4°Ğ4Ñ4Ô4Ñ5Ô5ˆ ØˆØ ğ 'ğ 'ˆDØ �KŠK�œ ™
œ
Ñ &Ô &Ğ &Ğ &İŒn˜T¨Ğ*Ñ*Ô*ˆİ" 3™ œ ğ 6ñ 6‰IˆAˆgØ×$Ò$Ñ&Ô&ˆDØ\Ğ\ÈDÏMÊMÉOÌOĞ\Ñ\Ô\ˆBİ�g×'Ò'Ñ)Ô)Ñ*Ô*ˆAİ ˜g× /Ò /Ñ 1Ô 1× ;Ò ;Ñ =Ô =Ñ>Ô>ÅÀg×FWÒFWÑFYÔFY×FcÒFcÑFeÔFeÑ@fÔ@fİ ˜g× /Ò /Ñ 1Ô 1× ;Ò ;Ñ =Ô =Ñ>Ô>ğ@ˆHà �NŠN˜B 8¨S°¬VĞ4Ñ 5Ô 5Ğ 5Ñ 5İŒ
Õ*Ñ,Ô,Ğ6Ğ6Ğ6Ñ7Ô7Ğ7؈r Údictc óĞ — | j j j j ¦ « }| ¦ « D ](\ }}| |t |¦ « ¦ « Œ)|S )N)r Újvmr- ÚutilÚHashMapr+ ÚputÚstr)r r] Újava_mapr( r) s r rD z JavaManager.dict_to_java_map: s\ € ØÔ$Ô(Ô-Ô2×:Ò:Ñ<Ô<ˆØ—J’J‘L”Lğ $ğ $‰DˆAˆqØ �LŠL˜�C ™FœFÑ #Ô #Ğ #Ğ #؈r c óˆ — | j ¦ « | j ¦ « t d¦ « d S )NzJava process ended.)r ÚkillÚwaitÚprintr0 s r r z JavaManager.kill_java_process@ s@ € Ø Ô× Ò Ñ Ô Ğ Ø Ô× Ò Ñ Ô Ğ İ
Ğ#Ñ$Ô$Ğ$؈r N) Ú__name__Ú
__module__Ú __qualname__r r] r r r r< r\ rD r r6 r r r
r
s´ € € € € € ğ 0 ğ 0°ğ 0ğ 0ğ 0ğ 0ğx ğ xğ xğ xğ xğFğ Fğ Fğ
°#ğ Àğ ğ ğ ğ ğ* Tğ ğ ğ ğ ğ ğ ğ ğ ğ r r
)r# r rO r. ÚnumpyrG Úpy4j.java_gatewayr Úmu_alpha_zero.General.utilsr r r Ú.mu_alpha_zero.MuZero.JavaGateway.java_networksr Ú%mu_alpha_zero.MuZero.Network.networksr r r
r6 r r ú<module>rq sÇ ğØ €€€Ø € € € Ø
€
€
€
Ø Ğ Ğ Ğ à Ğ Ğ Ğ Ø )Ğ )Ğ )Ğ )Ğ )Ğ )Ø bĞ bĞ bĞ bĞ bĞ bĞ bĞ bĞ bĞ bØ GĞ GĞ GĞ GĞ GĞ GØ ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø
€
€
€
ğ6ğ 6ğ 6ğ 6ğ 6ñ 6ô 6ğ 6ğ 6ğ 6r | 6,645 | Python | .pyt | 27 | 244.888889 | 1,309 | 0.34824 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,578 | java_networks.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/JavaGateway/__pycache__/java_networks.cpython-311.pyc | §
ÙxfA ã ó@ — d dl Z d dlmZ d dlmZ G d„ d¦ « ZdS )é N)Ú MuZeroNet)Úfind_project_rootc ó — e Zd Zdedefd„ZdS )Ú JavaNetworksÚnetworks_wrapperÚreturnc óR — t ¦ « › d�}| |¦ « |S )Nz /mz_net.pth)r Ú to_pickle)Úselfr Úpaths ú]/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/JavaGateway/java_networks.pyÚsavezJavaNetworks.save s/ € İ#Ñ%Ô%Ğ2Ğ2Ğ2ˆØ×"Ò" 4Ñ(Ô(Ğ(؈ ó N)Ú__name__Ú
__module__Ú __qualname__r Ústrr © r r
r r s6 € € € € € ğ Yğ °3ğ ğ ğ ğ ğ ğ r r )ÚmathÚ%mu_alpha_zero.MuZero.Network.networksr Úmu_alpha_zero.General.utilsr r r r r
ú<module>r sd ğØ € € € à ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø 9Ğ 9Ğ 9Ğ 9Ğ 9Ğ 9ğğ ğ ğ ğ ñ ô ğ ğ ğ r | 980 | Python | .pyt | 3 | 325.333333 | 606 | 0.431493 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,579 | arena.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/MZ_Arena/__pycache__/arena.cpython-311.pyc | §
‹fß
ã ó¶ — d dl Z d dlmZ d dlZd dlmZmZmZ d dl m
Z
d dl m Z d dl
mZ d dlmZ d dlmZmZmZ d d lmZ d dlZd d
lmZ G d „ d e
¦ « ZdS )
é N)ÚType)ÚPlayerÚ NetPlayerÚ RandomPlayer)Ú GeneralArena)Ú
MuZeroGame)Ú HookManager)ÚHookAt)Ú
resize_obsÚ scale_stateÚ scale_action)Ú MuZeroConfig)Ú
SharedStoragec
óz — e Zd Z ddededej depdfd„Z dd e e
d
e e
d e d e d
e de f d„Z
d„ ZdS )ÚMzArenaNÚ game_managerÚ
muzero_configÚdeviceÚ hook_managerc ó^ — || _ || _ |�|n
t ¦ « | _ || _ d S ©N)r r r r r )Úselfr r r r s úR/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/MZ_Arena/arena.pyÚ__init__zMzArena.__init__ s2 € à (ˆÔØ *ˆÔØ,8Ğ,D˜L˜LÍ+É-Ì-ˆÔØ ˆŒ ˆ ˆ ó Fé Úplayer1Úplayer2Únum_games_to_playÚnum_mc_simulationsÚ
one_playerÚ start_playerc ó$ — | j j }g g dœ}|r|} n|dz } t j dd¦ « }
||dœ} dD �]ô} || | j |ddœ}
t | ¦ « D �]Ó}| j ¦ « | j | j ¦ « | |
¬¦ « \ }}}t || j j | j j ¦ « }t || j j
¦ « }t | j j ¦ « D �]%}| j ¦ « | t! | ¦ « j |fi |
¤�}| j || ¦ « \ }}}t || j j | j j ¦ « }t || j j
¦ « } | t! | ¦ « j j |t+ || j ¦ « ¦ « | ¦ « n# t. $ r Y nw xY w|| |¦ « |r n�Œ'�ŒÕ�Œö| j | | j j t: t< j ||
f¦ « tA |d ¦ « tA |d
¦ « dfS ) N)r éÿÿÿÿé r é )Ú1z-1F)Únum_simulationsÚcurrent_playerr ÚtauÚunravel)Ú
frame_skipr r$ )!r Ú arena_tauÚrandomÚrandintr Úranger ÚresetÚframe_skip_stepÚget_noopr Útarget_resolutionÚ
resize_imagesr Ú num_stepsÚrenderÚstrÚ choose_moveÚmonte_carlo_tree_searchÚbufferÚ add_framer
Úget_num_actionsÚAttributeErrorÚappendr Úprocess_hook_executesÚpitÚ__name__Ú__file__r
ÚTAILÚsum)r r r r r r! r" r* ÚrewardsÚnum_games_per_playerÚnoop_numÚplayersÚplayerÚkwargsÚgameÚstateÚ_ÚstepÚmoveÚrewardÚdones r rA z MzArena.pit sß € àÔ Ô*ˆØ˜b�/�/ˆØ ğ :Ø#4Ğ Ğ à#4¸Ñ#9Ğ İ”> ! RÑ(Ô(ˆØ wĞ/Ğ/ˆØ ğ ñ ˆFØ);ÈvĞaeÔalØ ¨Uğ4ğ 4ˆFå Ğ 2Ñ3Ô3ğ
ñ
�ØÔ!×'Ò'Ñ)Ô)Ğ)à "Ô /× ?Ò ?ÀÔ@Q×@ZÒ@ZÑ@\Ô@\Ğ^dØKSğ @ñ Uô U‘ ��q˜!å" 5¨$Ô*<Ô*NĞPTÔPbÔPpÑqÔq�İ# E¨4Ô+=Ô+IÑJÔJ�İ ! $Ô"4Ô">Ñ ?Ô ?ğ ñ �DØÔ%×,Ò,Ñ.Ô.Ğ.Ø;˜7¥3 v¡;¤;Ô/Ô;¸EĞLĞLÀVĞLĞL�Dà*.Ô*;×*KÒ*KÈDĞRXÑ*YÔ*YÑ'�E˜6 4İ & u¨dÔ.@Ô.RĞTXÔTfÔTtÑ uÔ u�Eİ '¨¨tÔ/AÔ/MÑ NÔ N�Eğ Ø ¥ F¡ ¤ Ô,ÔDÔK×UÒUĞV[Õ]iĞjnØjnÔj{÷ kLò kLñ kNô kNñ^Oô ^OàV\ñ^ô ^ğ ^ğ ^øõ *ğ ğ ğ à ˜ğ øøøğ ˜F”O×*Ò*¨6Ñ2Ô2Ğ2Øğ Ø ˜ñ ùñ-
ğ2
Ô×/Ò/°°d´hÔ6GÍÕSYÔS^ĞahĞjpĞ`qÑrÔrĞrİ�7˜1”:‰Œ¥ G¨B¤KÑ 0Ô 0°!Ğ3Ğ3s Æ AG9Ç9
H ÈH c ór — | j | | j j t t
j ¦ « d S r )r r@ Úrun_on_training_endrB rC r
ÚALL)r s r rT zMzArena.run_on_training_endC s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr r )Fr )rB Ú
__module__Ú __qualname__r r Úthr r r r r ÚintÚboolrA rT © r r r r s¿ € € € € € à59ğ ğ Zğ À ğ ĞVXÔV_ğ Ø *Ğ 2¨dğ ğ ğ ğ ğ ;<ğ'4ğ '4˜4 œ<ğ '4°$°v´,ğ '4ĞSVğ '4Ğloğ '4Ø ğ'4Ø47ğ'4ğ '4ğ '4ğ '4ğRoğ oğ oğ oğ or r )r. Útypingr ÚtorchrX Ú%mu_alpha_zero.AlphaZero.Arena.playersr r r Úmu_alpha_zero.General.arenar Ú mu_alpha_zero.General.mz_gamer Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr
Úmu_alpha_zero.MuZero.utilsr r r
Úmu_alpha_zero.configr ÚwandbÚ$mu_alpha_zero.shared_storage_managerr r r[ r r ú<module>rg s ğØ
€
€
€
Ø Ğ Ğ Ğ Ğ Ğ à Ğ Ğ Ğ à QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QĞ QØ 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø LĞ LĞ LĞ LĞ LĞ LĞ LĞ LĞ LĞ LØ -Ğ -Ğ -Ğ -Ğ -Ğ -Ø € € € à >Ğ >Ğ >Ğ >Ğ >Ğ >ğ2oğ 2oğ 2oğ 2oğ 2oˆlñ 2oô 2oğ 2oğ 2oğ 2or | 5,155 | Python | .pyt | 22 | 233.272727 | 971 | 0.387807 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,580 | networks.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/MuZero/Network/__pycache__/networks.cpython-311.pyc | §
k¤føU ã ó˜ — d dl Z d dlZd dlZd dlmc mZ d dl Z d dlmZ d dlm
Z
d dl m Z
mZ d dlmZ d dlmZ d dlmZ d dlmZ d d lmZ d d
lmZ d d lm Z d d l m Z m Z m Z d d
l!m"Z" d dl#m$Z$ G d„ dej j% e¦ « Z& G d„ dej j% ¦ « Z' G d„ dej% ¦ « Z( G d„ dej j% ¦ « Z)dS )é N)Únn)Úmse_loss)Ú AlphaZeroNetÚOriginalAlphaZerNetwork)Ú CheckPointer)ÚLogger)ÚGeneralMemoryBuffer)Ú
MuZeroGame)ÚGeneralMuZeroNetwork)Ú HookManager)ÚHookAt)Úmatch_action_with_obs_batchÚscalar_to_supportÚsupport_to_scalar)Ú MuZeroConfig)Ú
SharedStoragec # ó ‡ — e Zd Z d8dededededee ded epee d
ed ed ed
edededededepddef"ˆ fd„
Ze d9de
depdfd„¦ « Z d:de j
dededefd „Zd;de j
dedefd „Zde j
fd „Zd<d e j
d edefd!„Zd"„ Zd#ed$e
d%eeee f fd&„Zd#ed$e
d%dfd'„Zd(„ Zd)ed%ee j
e j
e j
ee j
f fd*„Zd+ed,efd-„Zd+ed.efd/„Zd0„ Zd1e d$e
d2e d3e fd4„Z d5e fd6„Z!d7„ Z"ˆ xZ#S )=Ú MuZeroNetNTÚinput_channelsÚdropoutÚ action_sizeÚ num_channelsÚ latent_sizeÚnum_out_channelsÚlinear_input_sizeÚrep_input_channelsÚ use_originalÚ support_sizeÚ
num_blocksÚstate_linear_layersÚpi_linear_layersÚv_linear_layersÚlinear_head_hidden_sizeÚ hook_managerÚ use_poolingc ó •— t t | ¦ « ¦ « || _ || _ || _ | | _ || _ t j t j
¦ « rdnd¦ « | _ || _ || _
|| _ d | _ || _ || _ |
| _ | | _ | | _ |
| _ || _ || _ |�|n
t1 ¦ « | _ t5 ||¬¦ « | _ | rFt9 d||||| |
|||
|| dd¬¦ « | _ t9 d|||| |
||||
|| dd¬ ¦ « | _ d S t? d||||¬
¦ « | _ tA d||||¬ ¦ « | _ d S ) NÚcudaÚcpu)r% é T)Ú in_channelsr r r r r r! r" r# r r r ÚmuzeroÚ is_dynamicsé F)r* r r r r r! r" r# r r r r r+ r, )r* r r r Ú out_channels)r* r r r r )!Úsuperr Ú__init__r r r r r% ÚthÚdevicer' Ú is_availabler r r Ú optimizerr r r r r r! r" r# r r$ ÚRepresentationNetÚrepresentation_networkr Údynamics_networkÚprediction_networkÚ DynamicsNetÚ
PredictionNet)Úselfr r r r r r r r r r r r r! r" r# r$ r% Ú __class__s €úT/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/MuZero/Network/networks.pyr0 zMuZeroNet.__init__ sê ø€ õ
�i˜Ñ Ô ×'Ò'Ñ)Ô)Ğ)Ø ,ˆÔØ"4ˆÔ Ø ˆŒ Ø (ˆÔØ&ˆÔİ”i"¬'×*>Ò*>Ñ*@Ô*@Ğ K ÀeÑLÔLˆŒ Ø&ˆÔØ (ˆÔØ&ˆÔØ ˆŒØ 0ˆÔ Ø!2ˆÔ Ø (ˆÔØ$ˆŒØ#6ˆÔ Ø 0ˆÔ Ø .ˆÔ Ø'>ˆÔ$Ø,8Ğ,D˜L˜LÍ+É-Ì-ˆÔå&7Ğ8JĞXcĞ&dÑ&dÔ&dˆÔ#Ø ğ rİ$;ÈĞZjØDKØHSØN_ØPcØM]ØL[ØTkØIUĞcnØGQĞZ^Ğlpğ %rñ %rô %rˆDÔ !õ '>È#Ğ\lØFMØJUØReØO_ØN]ØVmØPaØKWĞepØISĞ\`Ğnsğ 'uñ 'uô 'uˆDÔ #Ğ #Ğ #õ %0¸CÈlĞdkØ<GĞVfğ%hñ %hô %hˆDÔ !õ '4ÀĞR^ĞhoØ@KĞ_pğ'rñ 'rô 'rˆDÔ #Ğ #Ğ #ó Úconfigc óÚ — | |j |j |j |j |j |j |j |j |j |j |j
|j |j |j
|j ||j ¬¦ « S )N)r$ r% )Únum_net_in_channelsÚ net_dropoutÚnet_action_sizeÚnum_net_channelsÚnet_latent_sizeÚnum_net_out_channelsÚaz_net_linear_input_sizer r r r r r! r" r# r% )Úclsr? r$ s r= Úmake_from_configzMuZeroNet.make_from_configO sx € àˆs�6Ô-¨vÔ/AÀ6ÔCYĞ[aÔ[rØÔ)¨6Ô+FÈÔHgØÔ,¨fÔ.AÀ6ÔCVĞX^ÔXiØÔ-¨vÔ/FÈÔH^ØÔ1Ø ,¸&Ô:Lğ Nñ Nô Nğ Nr> FÚxÚpredictÚreturn_supportÚconvert_to_statec óæ — |rX| j |d¬¦ « \ }}| ¦ « ¦ « ¦ « }n| |d|¬¦ « \ }}|rw | | j | j d | j d ¦ « }nC# t $ r6 | d| j | j d | j d ¦ « }Y nw xY w||fS )NT©r+ ©r+ rL r é éÿÿÿÿ) r7 ÚforwardÚdetachr( ÚnumpyÚviewr r Ú RuntimeError)r; rJ rK rL rM ÚstateÚrÚrewards r= Údynamics_forwardzMuZeroNet.dynamics_forwardX s € à ğ aØÔ,×4Ò4°Q¸tĞ4ÑDÔD‰HˆE�1Ø—X’X‘Z”Z—^’^Ñ%Ô%×+Ò+Ñ-Ô-ˆFˆFà × 1Ò 1°!¸DĞQ_Ğ 1Ñ `Ô `‰MˆE�6Ø ğ hğ
hØ Ÿ
š
4Ô#8¸$Ô:JÈ1Ô:MÈtÔO_Ğ`aÔObÑcÔc��øİ ğ
hğ
hğ
hà Ÿ
š
2 tÔ'<¸dÔ>NÈqÔ>QĞSWÔScĞdeÔSfÑgÔg���ğ
høøøğ �fˆ}Ğ s Á92B, Â,=C,Ã+C,c óŠ — |r#| j |d¬¦ « \ }}||fS | |d|¬¦ « \ }}||fS )NTrO rP )r8 rK )r; rJ rK rL ÚpiÚvs r= Úprediction_forwardz MuZeroNet.prediction_forwardg sY € Ø ğ ØÔ+×3Ò3°A¸dĞ3ÑCÔC‰EˆB�Ø�q�5ˆLØ×'Ò'¨°$À~Ğ'ÑVÔV‰ˆˆAØ�1ˆuˆ r> c ó0 — | |¦ « }|S ©N)r6 )r; rJ s r= Úrepresentation_forwardz MuZeroNet.representation_forwardn s € Ø × 'Ò '¨Ñ *Ô *ˆØˆr> Úhidden_state_with_actionÚ all_predictc óz — | |||¬¦ « \ }}| |||¬¦ « \ }}||||fS )N)rK rL )r[ r_ )r; rc rd rL Ú
next_staterZ r] r^ s r= Úforward_recurrentzMuZeroNet.forward_recurrentr sZ € Ø !× 2Ò 2Ğ3KĞU`ØBPğ 3ñ Rô Rш
�Fà×'Ò'¨
¸KĞXfĞ'ÑgÔg‰ˆˆAؘ6 2 qĞ(Ğ(r> c óì — t | j | j | j | j | j | j | j | j | j | j
| j | j | j
| j | j | j | j ¬¦ « S )N) r$ r r r r% r r! r" r# )r r r r r r r r r r$ r r r r% r r! r" r# ©r; s r= Úmake_fresh_instancez MuZeroNet.make_fresh_instancex sy € ݘÔ,¨d¬l¸DÔ<LÈdÔN_ĞaeÔaqØ Ô.°Ô0FÈÔH_Ø&*Ô&7ÀdÔFWĞfjÔfwØ$(¤OÀÔAQØ-1Ô-EØ*.Ô*?ĞQUÔQeØ15Ô1Mğ
Oñ Oô Oğ Or> Ú
memory_bufferÚ
muzero_configÚreturnc
󠇇— t j t j ¦ « rdnd¦ « }| j €Ct j | ¦ « ‰j ‰j ¬¦ « | _ g }d}ˆˆfd„}t ‰j ¦ « D �]e} |¦ « \ }} }
t |¦ « dk rŒ%|
||
|‰¦ « \ } } }
}t j | ¦ « | ¦ « |
¦ « | ¦ « dœ¦ « | | ¦ « ¦ « | j ¦ « | ¦ « | j ¦ « | j | | j j t2 t4 j || ¦ « | |
||f¬¦ « |dz
}�Œg| j | | j j t2 t4 j ‰|f¬¦ « t; |¦ « t |¦ « z |fS ) Nr' r( ©ÚlrÚ weight_decayr c óF •— ‰ ‰j ‰j ‰¦ « S ra ©Úbatch_with_prioritiesÚ
enable_perÚ
batch_size©rk rl s €€r= ú<lambda>z%MuZeroNet.train_net.<locals>.<lambda>ˆ s' ø€ ˜ ×<Ò<¸]Ô=UØ=JÔ=UĞWdñfô f€ r> rQ )Ú
combined_lossÚloss_vÚloss_piÚloss_r)Úargs) r1 r2 r' r3 r4 ÚoptimÚAdamÚ
parametersrp Úl2ÚrangeÚepochsÚlenÚcalculate_lossesÚwandbÚlogÚitemÚappendÚ zero_gradÚbackwardÚstepr$ Úprocess_hook_executesÚ train_netÚ__name__Ú__file__r
ÚMIDDLEÚTAILÚsum)r; rk rl r2 ÚlossesÚ iterationÚloaderÚepochÚsampled_game_dataÚ
prioritiesÚweightsÚlossrz r{ r| s `` r= r� zMuZeroNet.train_net� sG øø€ İ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆØ Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<À ÔAQØ8EÔ8Hğ +ñ Jô JˆDŒNàˆØˆ ğfğ fğ fğ fğ fˆå˜=Ô/Ñ0Ô0ğ ñ ˆEØ5;°V±X´XÑ 2Ğ ˜z¨7İĞ$Ñ%Ô%¨Ò*Ğ*ØØ,0×,AÒ,AĞBSĞU\Ğ^dĞfsÑ,tÔ,tÑ )ˆD�&˜' 6İ ŒI¨¯ ª © ¬ ¸v¿{º{¹}¼}ĞY`×YeÒYeÑYgÔYgØ!'§¢¡ ¤ ğ0ğ 0ñ
1ô
1ğ
1à �MŠM˜$Ÿ)š)™+œ+Ñ &Ô &Ğ &Ø ŒN× $Ò $Ñ &Ô &Ğ &Ø �MŠM‰OŒOˆOØ ŒN× Ò Ñ !Ô !Ğ !Ø Ô × 3Ò 3°D¸$¼.Ô:QÕS[Õ]cÔ]jØ! 4§9¢9¡;¤;°¸ÀØğrĞ 3ñ
ô
ğ
ğ
˜‰NˆI‰IØ Ô×/Ò/°°d´nÔ6MÍxÕY_ÔYdØ6CÀVĞ5Lğ 0ñ Nô Nğ Nå�6‰{Œ{�S ™[œ[Ñ(¨&Ğ0Ğ0r> c ó´ ‡‡— ‰ ¦ « dk rd S t j t j ¦ « rdnd¦ « }| j €Ct j | ¦ « ‰j ‰j
¬¦ « | _ ˆˆfd„}t ‰j ¦ « D ]¢} |¦ « \ }}}t |¦ « dk rŒ$| |||‰¦ « \ } }
} } t j | ¦ « |
¦ « | ¦ « | ¦ « dœ¦ « Œ£d S )Nr r' r( ro c óJ •— ‰ ‰j ‰j ‰d¬¦ « S )NT)Úis_evalrs rw s €€r= rx z$MuZeroNet.eval_net.<locals>.<lambda>¥ s/ ø€ ˜ ×<Ò<¸]Ô=UØ=JÔ=UØ=JØEIğ =ñ Kô K€ r> rQ )Úeval_combined_lossÚ eval_loss_vÚ eval_loss_piÚ eval_loss_r)Ú eval_lengthr1 r2 r' r3 r4 r~ r r€ rp r� r‚ Ú eval_epochsr„ r… r† r‡ rˆ )
r; rk rl r2 r– r— Úexperience_batchr™ rš r› rz r{ r| s
`` r= Úeval_netzMuZeroNet.eval_net� sj øø€ Ø × $Ò $Ñ &Ô &¨!Ò +Ğ +Ø ˆFİ”¥R¤W×%9Ò%9Ñ%;Ô%;ĞF˜6˜6ÀÑGÔGˆØ Œ>Ğ !İ œXŸ]š]¨4¯?ª?Ñ+<Ô+<À ÔAQØ8EÔ8Hğ +ñ Jô JˆDŒNğKğ Kğ Kğ Kğ Kˆõ ˜=Ô4Ñ5Ô5ğ 6ğ 6ˆEØ4:°F±H´HÑ 1Ğ ˜j¨'İĞ#Ñ$Ô$¨Ò)Ğ)ØØ,0×,AÒ,AĞBRĞT[Ğ]cĞerÑ,sÔ,sÑ )ˆD�&˜' 6İ ŒI¨T¯YªY©[¬[ÈÏÊÉ Ì Ğho×htÒhtÑhvÔhvØ&,§k¢k¡m¤mğ5ğ 5ñ
6ô
6ğ
6ğ
6ğ 6ğ 6r> c ó ‡— | d||¦ « \ }}}}} t |‰j ¦ « }
| |¦ « } | | d¬¦ « \ } }
d\ }}}|| | | ¦ « z
}|| |
|
¦ « z
}d„ t
| d¦ « ¦ « D ¦ « }‰j rk| t j t |
‰j ¦ « |z
¦ « ‰j
z d¦ « ¦ « |¦ « t
d‰j dz ¦ « D �]–}| |||¦ « \ }}}}} t |‰j ¦ « }t |‰j ¦ « }
| t% | |¦ « dd¬¦ « \ } }} }
| d „ ¦ « | | | ¦ « }| |
|
¦ « }| ||¦ « }| ˆfd
„¦ « | ˆfd „¦ « | ˆfd „¦ « ||z
}||z
}||z
}‰j rk| t j t |
‰j ¦ « |z
¦ « ‰j
z d¦ « ¦ « |¦ « �Œ˜|d
z }||z |z }‰j r$|t j ||j |j ¬¦ « z }| ¦ « }‰j r| ||¦ « || ¦ « | ¦ « | ¦ « fS )Nr T)rL )r r r c ó — g | ]}g ‘ŒS © r© )Ú.0rJ s r= ú
<listcomp>z.MuZeroNet.calculate_losses.<locals>.<listcomp>º s € Ğ>Ğ>Ğ> ˜"Ğ>Ğ>Ğ>r> rR rQ Fc ó — | dz S )Ng à?r© )Úgrads r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Æ s
€ °D¸3±J€ r> c ó •— | d‰j z z S ©NrQ ©ÚK©r rl s €r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Ê ó ø€ °d¸aÀ-Ä/Ñ>QÑ6R€ r> c ó •— | d‰j z z S r¯ r° r² s €r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Ë r³ r> c ó •— | d‰j z z S r¯ r° r² s €r= rx z,MuZeroNet.calculate_losses.<locals>.<lambda>Ì s ø€ °t¸qÀ=Ä?Ñ?RÑ7S€ r> g Ğ?©Údtyper2 )Úget_batch_for_unroll_indexr r rb r_ Ú muzero_lossr‚ Úsizeru Úpopulate_prioritiesr1 Úabsr ÚalphaÚreshapeÚtolistr± rg r Ú
register_hookÚtensorr· r2 r“ Úupdate_priorities)r; r¥ rš r2 rl Ú init_statesÚrewardsÚ
scalar_valuesÚmovesÚpisÚvaluesÚ hidden_stateÚpred_pisÚpred_vsÚpi_lossÚv_lossÚr_lossÚnew_prioritiesÚiÚ_Úpred_rsÚcurrent_pi_lossÚcurrent_v_lossÚcurrent_r_lossr› s ` r= r… zMuZeroNet.calculate_losses± s ø€ Ø:>×:YÒ:YĞZ[Ğ]mĞouÑ:vÔ:vÑ7ˆ �W˜m¨U°Cå" =°-Ô2LÑMÔMˆØ×2Ò2°;Ñ?Ô?ˆ Ø × 3Ò 3°LĞQUĞ 3Ñ VÔ Vш�'Ø")Ñ ˆ�˜Ø�4×#Ò# H¨cÑ2Ô2Ñ2ˆØ�$×"Ò" 7¨FÑ3Ô3Ñ3ˆØ>Ğ>¥e¨H¯MªM¸!Ñ,<Ô,<Ñ&=Ô&=Ğ>Ñ>Ô>ˆØ Ô #ğ .Ø × $Ò $¥b¤fÕ->¸wØ?LÔ?Yñ.[ô .[Ø]jñ.kñ 'lô 'lØo|ô pCñ'C÷ ELò ELØñEô Eß’F‘H”H˜nñ
.ô
.ğ
.õ �q˜-œ/¨AÑ-Ñ.Ô.ğ 2ñ 2ˆAØ48×4SÒ4SĞTUĞWgØTZñ5\ô 5\Ñ 1ˆAˆw˜
u¨cå'¨° Ô1KÑLÔLˆGİ& }°mÔ6PÑQÔQˆFØ7;×7MÒ7Mİ+¨L¸%Ñ@Ô@À%ĞX\ğ 8Nñ 8^ô 8^Ñ 4ˆL˜' 8¨Wà × &Ò &Ğ'>Ğ'>Ñ ?Ô ?Ğ ?Ø "× .Ò .¨x¸Ñ =Ô =ˆOØ !× -Ò -¨g°vÑ >Ô >ˆNØ !× -Ò -¨g°wÑ ?Ô ?ˆNØ × (Ò (Ğ)RĞ)RĞ)RĞ)RÑ SÔ SĞ SØ × (Ò (Ğ)RĞ)RĞ)RĞ)RÑ SÔ SĞ SØ × )Ò )Ğ*SĞ*SĞ*SĞ*SÑ TÔ TĞ TØ �Ñ &ˆGØ �nÑ $ˆFØ �nÑ $ˆFØ Ô'ğ
2Ø×(Ò("¬&Õ1BÀ7ØCPÔC]ñ2_ô 2_Øanñ2oñ +pô +pğ tAô tGñ+G÷ IPò IPØñIô Iß š™œ .ñ2ô 2ğ 2ùğ �$‰ˆØ˜Ñ &Ñ(ˆØ Ô #ğ MØ •B”I˜g¨T¬ZÀÄ ĞLÑLÔLÑ LˆDØ�xŠx‰zŒzˆØ Ô #ğ EØ × "Ò " >Ğ3CÑ DÔ DĞ DØ�V—Z’Z‘\”\ 7§;¢;¡=¤=°&·*²*±,´,Ğ>Ğ>r> Úindexc ó ‡‡— ˆfd„}d }‰dk rAˆfd„|D ¦ « } |t j |¦ « ¦ « dddd¦ « }t j ˆfd„|D ¦ « ¦ « } ||¦ « }t j ˆfd„|D ¦ « ¦ « } ||¦ « }ˆfd „|D ¦ « }t j ˆfd
„|D ¦ « ¦ « } || ¦ « } || d¦ « | d¦ « || fS ) Nc óF •— t j | t j ‰¬¦ « S )Nr¶ )r1 rÁ Úfloat32)rJ r2 s €r= rx z6MuZeroNet.get_batch_for_unroll_index.<locals>.<lambda>à s ø€ ¥"¤)¨AµR´ZÈĞ"OÑ"OÔ"O€ r> r c óX •— g | ]&}t j |j ‰ j ¦ « ‘Œ'S r© )ÚnpÚarrayÚ
datapointsÚframe©rª rJ rÖ s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>ã s- ø€ ĞYĞYĞYÀ1�2œ8 A¤L°Ô$7Ô$=Ñ>Ô>ĞYĞYĞYr> é rQ é c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ rZ rß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>å s# ø€ ĞQĞQĞQ¸1˜AœL¨Ô /Ô 6ĞQĞQĞQr> c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ r^ rß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>ç s# ø€ ĞKĞKĞK°Q˜1œ<¨Ô.Ô0ĞKĞKĞKr> c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ Úmoverß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>é s# ø€ ĞDĞDĞD¨a�” ˜eÔ$Ô)ĞDĞDĞDr> c ó4 •— g | ]}|j ‰ j ‘ŒS r© )rİ r] rß s €r= r« z8MuZeroNet.get_batch_for_unroll_index.<locals>.<listcomp>ì s# ø€ ĞIĞIĞI°1˜œ UÔ+Ô.ĞIĞIĞIr> )rÛ rÜ ÚpermuteÚ unsqueeze)
r; rÖ r¥ r2 Ú
tensor_from_xrà rÄ rÈ rÆ rÇ s
` ` r= r¸ z$MuZeroNet.get_batch_for_unroll_indexŞ sM øø€ àOĞOĞOĞOˆ
؈ Ø �AŠ:ˆ:ØYĞYĞYĞYĞHXĞYÑYÔYˆKØ'˜-¬°Ñ(=Ô(=Ñ>Ô>×FÒFÀqÈ!ÈQĞPQÑRÔRˆKİ”(ĞQĞQĞQĞQĞ@PĞQÑQÔQÑRÔRˆØ �- Ñ(Ô(ˆİ”ĞKĞKĞKĞKĞ:JĞKÑKÔKÑLÔLˆØ � ˜vÑ&Ô&ˆØDĞDĞDĞDĞ3CĞDÑDÔDˆõ ŒhĞIĞIĞIĞIĞ8HĞIÑIÔIÑJÔJˆØˆm˜CÑ Ô ˆØ˜G× -Ò -¨aÑ 0Ô 0°&×2BÒ2BÀ1Ñ2EÔ2EÀuÈcĞQĞQr> rÏ Úpriorities_listc óf — t |¦ « D ] \ }}|| |¦ « Œ!d S ra )Ú enumerater‰ )r; rÏ rê ÚidxÚprioritys r= r» z MuZeroNet.populate_prioritiesğ sC € İ & ~Ñ 6Ô 6ğ 2ğ 2‰MˆC�Ø ˜CÔ × 'Ò '¨Ñ 1Ô 1Ğ 1Ğ 1ğ 2ğ 2r> r¥ c ó´ — t |¦ « D ]G\ }}t t |j ¦ « ¦ « D ] }|| | |j | _ Œ!ŒHd S ra )rì r‚ r„ rİ rî )r; rÏ r¥ rí ÚgamerĞ s r= r zMuZeroNet.update_prioritiesô sq € İ"Ğ#3Ñ4Ô4ğ Eğ E‰IˆC�İ�3˜tœÑ/Ô/Ñ0Ô0ğ
Eğ
E�Ø.<¸SÔ.AÀ!Ô.D�” Ô"Ô+Ğ+ğ
Eğ Eğ Er> c ó’ — t j ||z d¬¦ « d¦ « | ¦ « d z S )NrQ )Údimr )r1 r“ rè rº )r; Úy_hatÚys r= r¹ zMuZeroNet.muzero_lossù s= € İ”�q˜5‘y aĞ(Ñ(Ô(×2Ò2°1Ñ5Ô5Ğ5¸¿º¹¼À¼ ÑCĞCr> Úshared_storageÚ checkpointerÚloggerc óP — t j |j d¬¦ « | ¦ « d|_ d|_ t
| ¦ « ¦ « |j dz k rAt j
d¦ « t
| ¦ « ¦ « |j dz k °A| d¦ « t |j
¦ « D ]Ğ}| ||¦ « \ }}| | ¦ « ¦ « | | j ¦ « ¦ « | ||¦ « |dz dk rC|dk r=| d |› d
�¦ « | | | | j |j ||¦ « ŒÑd S ) NzContinuous Weight Update)ÚprojectÚnamerQ é é z:Finished waiting for target buffer size,starting training.iô r z Saving checkpoint at iteration ú.)r† ÚinitÚwandbd_project_nameÚtrainrƒ r¤ r„ Ú
get_bufferrv ÚtimeÚsleepr‡ r‚ Únum_worker_itersr� Ú set_experimental_network_paramsÚ
state_dictÚ
set_optimizerr4 r¦ Úsave_checkpointrp )r; rõ rl rö r÷ Úiter_ÚavgÚ iter_lossess r= Úcontinuous_weight_updatez"MuZeroNet.continuous_weight_updateü s· € å
Œ
˜=Ô<ĞC]Ğ^Ñ^Ô^Ğ^Ø �
Š
‰ Œ ˆ ğ !ˆ
Ô Ø$%ˆ
Ô!İØ ×)Ò)Ñ+Ô+ñ-ô -Ø/<Ô/GÈ1Ñ/LòMğ Må ŒJ�q‰MŒMˆMõ Ø ×)Ò)Ñ+Ô+ñ-ô -Ø/<Ô/GÈ1Ñ/LòMğ Mğ �
Š
ĞOÑPÔPĞPݘ=Ô9Ñ:Ô:ğ qğ qˆEğ $Ÿ~š~¨n¸mÑ LÔ LÑ ˆC�Ø × :Ò :¸4¿?º?Ñ;LÔ;LÑ MÔ MĞ MØ × (Ò (¨¬ ×)BÒ)BÑ)DÔ)DÑ EÔ EĞ Eğ
�MŠM˜.¨-Ñ 8Ô 8Ğ 8Ø�s‰{˜aÒ Ğ E¨Q¢J JØ—
’
ĞE¸UĞEĞEĞEÑFÔFĞFØ ×,Ò,¨T°4¸¼ È ÔIYĞ[`ĞboÑpÔpĞpøğ qğ qr> Úpathc ó0 — t j | |¦ « d S ra )r1 Úsave)r; r
s r= Ú to_picklezMuZeroNet.to_pickle s € İ
Œ��dÑÔĞĞĞr> c ór — | j | | j j t t
j ¦ « d S ra )r$ r� Úrun_on_training_endr� r� r
ÚALLri s r= r z MuZeroNet.run_on_training_end s0 € Ø Ô×/Ò/°°dÔ6NÔ6WÕYaÕciÔcmÑnÔnĞnĞnĞnr> )NTra )FFT)FF©F)$r� Ú
__module__Ú __qualname__ÚintÚfloatÚlistÚboolr r0 Ú classmethodr rI r1 ÚTensorr[ r_ rb rg rj r Útupler� r¦ r… r¸ r» r r¹ r r r r Ústrr r Ú
__classcell__©r< s @r= r r s¨ ø€ € € € € ğ
PTğ 5rğ 5r sğ 5r°Uğ 5rÈğ 5rĞ\_ğ 5rĞnrĞsvÔnwğ 5rØ#&ğ5rØ;>Ğ;KÀ$ÀsÄ)ğ5rØadğ5rà #ğ5rà36ğ5ràDGğ5rğ '*ğ5rğ >Ağ5rğ TWğ5rğ ruğ5rğ +Ğ 2¨dğ 5rğ IMğ 5rğ 5rğ 5rğ 5rğ 5rğ 5rğn ğNğ N lğ NÀ+ĞBUĞQUğ Nğ Nğ Nñ „[ğNğ \aØ26ğ
ğ
"¤)ğ
°dğ
ĞTXğ
Ø+/ğ
ğ
ğ
ğ
ğ ğ B¤Iğ ¸ğ ĞVZğ ğ ğ ğ 𨬠ğ ğ ğ ğ ğ)ğ )¸"¼)ğ )ĞRVğ )Ğhlğ )ğ )ğ )ğ )ğ Oğ Oğ Oğ1Ğ':ğ 1È<ğ 1Ğ\aĞbgĞimĞnsÔitĞbtÔ\uğ 1ğ 1ğ 1ğ 1ğ86Ğ&9ğ 6È,ğ 6Ğ[_ğ 6ğ 6ğ 6ğ 6ğ(+?ğ +?ğ +?ğZR°ğ RĞRWØ
Œ �2”9˜bœi¨¨r¬yĞ8ôS:ğ Rğ Rğ Rğ Rğ$2°$ğ 2Èğ 2ğ 2ğ 2ğ 2ğE°ğ EÈğ Eğ Eğ Eğ Eğ
Dğ Dğ Dğ q°}ğ qĞUağ qØ/;ğ qØEKğ qğ qğ qğ qğ> ˜cğ ğ ğ ğ ğoğ oğ oğ oğ oğ oğ or> r c óF ‡ — e Zd Zddedefˆ fd„
Zdej fd„Zd„ Z ˆ xZ
S ) r5 Tr r% c óB •— t t | ¦ « ¦ « t j t j ¦ « rdnd¦ « | _ t j |dddd¬¦ « | _ t j
d„ t d¦ « D ¦ « ¦ « | _ t j dd ddd¬¦ « | _
t j
d
„ t d¦ « D ¦ « ¦ « | _ t j
d „ t d¦ « D ¦ « ¦ « | _ |rOt j ddd¬ ¦ « | _ t j ddd¬ ¦ « | _ nFt j ¦ « | _ t j ¦ « | _ t j ¦ « | _ d S )
Nr' r( é€ rà rá rQ )r* r. Ú kernel_sizeÚstrideÚpaddingc ó, — g | ]}t d ¦ « ‘ŒS )r# ©Ú
ResidualBlock©rª rÑ s r= r« z.RepresentationNet.__init__.<locals>.<listcomp>' ó € Ğ+QĞ+QĞ+QÀ1M¸#Ñ,>Ô,>Ğ+QĞ+QĞ+Qr> r- c ó, — g | ]}t d ¦ « ‘ŒS ©r- r( r* s r= r« z.RepresentationNet.__init__.<locals>.<listcomp>) r+ r> c ó, — g | ]}t d ¦ « ‘ŒS r- r( r* s r= r« z.RepresentationNet.__init__.<locals>.<listcomp>+ r+ r> )r$ r% r& )r/ r5 r0 r1 r2 r' r3 r ÚConv2dÚconv1Ú
ModuleListr‚ Ú
residuals1Úconv2Ú
residuals2Ú
residuals3Ú AvgPool2dÚpool1Úpool2ÚIdentityÚReLUÚrelu)r; r r% r< s €r= r0 zRepresentationNet.__init__# sŒ ø€ İ
Õ Ñ&Ô&×/Ò/Ñ1Ô1Ğ1İ”i"¬'×*>Ò*>Ñ*@Ô*@Ğ K ÀeÑLÔLˆŒ İ”U—\’\Ğ.@ÈsĞ`aĞjkĞuv�\ÑwÔwˆŒ
İ œ%×*Ò*Ğ+QĞ+QÍÈaÉÌĞ+QÑ+QÔ+QÑRÔRˆŒİ”U—\’\¨cÀĞQRĞ[\Ğfg�\ÑhÔhˆŒ
İ œ%×*Ò*Ğ+QĞ+QÍÈaÉÌĞ+QÑ+QÔ+QÑRÔRˆŒå œ%×*Ò*Ğ+QĞ+QÍÈaÉÌĞ+QÑ+QÔ+QÑRÔRˆŒØ ğ *İœŸ š °Q¸qÈ!˜ ÑLÔLˆDŒJİœŸ š °Q¸qÈ!˜ ÑLÔLˆDŒJˆJ土 š Ñ)Ô)ˆDŒJİœŸ š Ñ)Ô)ˆDŒJİ”E—J’J‘L”LˆŒ ˆ ˆ r> rJ c ó¬ — | | j ¦ « }| | |¦ « ¦ « }| j D ]
} ||¦ « }Œ| | |¦ « ¦ « }| j D ]
} ||¦ « }Œ| |¦ « }| j D ]
} ||¦ « }Œ| |¦ « }|S ra )
Útor2 r; r0 r2 r3 r4 r7 r5 r8 )r; rJ Úresiduals r= rS zRepresentationNet.forward4 sÎ € à
�DŠD�”Ñ Ô ˆØ �IŠI�d—j’j ‘m”mÑ $Ô $ˆØ œğ ğ ˆHØ�˜‘ ” ˆAˆAØ �IŠI�d—j’j ‘m”mÑ $Ô $ˆØ œğ ğ ˆHØ�˜‘ ” ˆAˆAØ �JŠJ�q‰MŒMˆØ œğ ğ ˆHØ�˜‘ ” ˆAˆAØ �JŠJ�q‰MŒMˆØˆr> c ón — t j d¦ « }t j | |¦ « }|S )N)rQ r# é r@ )r1 ÚrandÚjitÚtrace©r; ÚdataÚtraced_script_modules r= rC zRepresentationNet.traceC s- € İŒw�~Ñ&Ô&ˆİ !œvŸ|š|¨D°$Ñ 7Ô 7Ğ Ø#Ğ#r> )T) r� r r r r r0 r1 r rS rC r r s @r= r5 r5 " s| ø€ € € € € ğ!ğ !¨3ğ !¸Tğ !ğ !ğ !ğ !ğ !ğ !ğ"
˜œğ
ğ
ğ
ğ
ğ $ğ $ğ $ğ $ğ $ğ $ğ $r> r5 c óv ‡ — e Zd Zˆ fd„Zdd„Zdej j fd„Z ej ¦ « d„ ¦ « Z
ˆ xZ S )r9 c ó¨ •— t t | ¦ « ¦ « || _ || _ t j ||dd¬¦ « | _ t j |¦ « | _ t j ||dd¬¦ « | _
t j |¦ « | _ t j ||dd¬¦ « | _ t j |¦ « | _
t j ||d¦ « | _ t j |¦ « | _ t j dd¦ « | _ t j d¦ « | _ t j dd¦ « | _ t j d¦ « | _ t j |¦ « | _ t j d|d |d z |z ¦ « | _ t j dd¦ « | _ d S )Nrà rQ )r& i i i r )r/ r9 r0 r. r r r/ r0 Ú BatchNorm2dÚbn1r3 Úbn2Úconv3Úbn3Úconv4Úbn4ÚLinearÚfc1Ú BatchNorm1dÚfc1_bnÚfc2Úfc2_bnÚDropoutr Ú
state_headÚ reward_head)r; r* r r r r. r< s €r= r0 zDynamicsNet.__init__J st ø€ İ
�k˜4Ñ Ô ×)Ò)Ñ+Ô+Ğ+Ø (ˆÔØ&ˆÔõ ”Y˜{¨L¸!ÀQĞGÑGÔGˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1ÀaĞHÑHÔHˆŒ
İ”> ,Ñ/Ô/ˆŒİ”Y˜|¨\¸1Ñ=Ô=ˆŒ
İ”> ,Ñ/Ô/ˆŒõ ”9˜T 4Ñ(Ô(ˆŒİ”n TÑ*Ô*ˆŒ İ”9˜T 3Ñ'Ô'ˆŒİ”n SÑ)Ô)ˆŒ å”z 'Ñ*Ô*ˆŒ õ œ) C¨°Q¬ ¸+Àa¼.Ñ)HÈ<Ñ)WÑXÔXˆŒİ œ9 S¨!Ñ,Ô,ˆÔĞĞr> Fc ó¼ — t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }t j | | |¦ « ¦ « ¦ « }|
| d¦ « d¦ « }t j | |
|¦ « ¦ « ¦ « }| |¦ « }t j | | |¦ « ¦ « ¦ « }| |¦ « }| |¦ « }| |¦ « }||fS )Nr rR )ÚFr; rJ r0 rK r3 rM rL rO rN rV rº rS rQ r rU rT rW rX )r; rJ r+ rX rY s r= rS zDynamicsNet.forwardf sH € å
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆİ
ŒF�4—8’8˜DŸJšJ q™MœMÑ*Ô*Ñ +Ô +ˆà
�FŠF�1—6’6˜!‘9”9˜bÑ !Ô !ˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆå
ŒF�4—;’;˜tŸxšx¨™{œ{Ñ+Ô+Ñ ,Ô ,ˆØ �LŠL˜‰OŒOˆà—’ Ñ"Ô"ˆØ × Ò ˜QÑ Ô ˆà�aˆxˆr> rm c ó” — t j d¦ « d¦ « }t j | |¦ « }|S )N)rQ r) é r\ zcuda:0)r1 rA r= rB rC rD s r= rC zDynamicsNet.tracez s; € İŒw�~Ñ&Ô&×)Ò)¨(Ñ3Ô3ˆİ !œvŸ|š|¨D°$Ñ 7Ô 7Ğ Ø#Ğ#r> c ó
— | |¦ « \ }}| | j | j d | j d ¦ « }|| ¦ « ¦ « ¦ « fS )Nr rQ )rS rV r. r rT r( rU )r; rJ rX rY s r= rK zDynamicsNet.predict sg € à—<’< ‘?”?‰ˆˆqØ—
’
˜4Ô,¨dÔ.>¸qÔ.AÀ4ÔCSĞTUÔCVÑWÔWˆØ�a—h’h‘j”j—n’nÑ&Ô&×,Ò,Ñ.Ô.Ğ.Ğ.r> r ) r� r r r0 rS r1 rB ÚScriptFunctionrC Úno_gradrK r r s @r= r9 r9 I s� ø€ € € € € ğ-ğ -ğ -ğ -ğ -ğ8ğ ğ ğ ğ($�r”vÔ,ğ $ğ $ğ $ğ $ğ
€R„Z�\„\ğ/ğ /ñ „\ğ/ğ /ğ /ğ /ğ /r> r9 c ó* ‡ — e Zd Zdefˆ fd„ Zd„ Zˆ xZS )r) Úchannelsc óÌ •— t t | ¦ « ¦ « t j ||dd¬¦ « | _ t j |¦ « | _ t j ||dd¬¦ « | _ t j |¦ « | _
t j ¦ « | _ d S )Nrà rQ )r* r. r$ r& )
r/ r) r0 r1 r r/ Ú convolution1rI Úbnorm1Ú convolution2Úbnorm2r: r; )r; ra r< s €r= r0 zResidualBlock.__init__‡ s¢ ø€ İ
�m˜TÑ"Ô"×+Ò+Ñ-Ô-Ğ-İ œEŸLšL°XÈHĞbcĞmn˜LÑ oÔ oˆÔİ”e×'Ò'¨Ñ1Ô1ˆŒ İ œEŸLšL°XÈHĞbcĞmn˜LÑ oÔ oˆÔİ”e×'Ò'¨Ñ1Ô1ˆŒ İ”E—J’J‘L”LˆŒ ˆ ˆ r> c ó — |}| |¦ « }| | |¦ « ¦ « }| | |¦ « ¦ « }||z
}| |¦ « }|S ra )rc r; rd rf re )r; rJ Úx_resÚ convolveds r= rS zResidualBlock.forward� ss € àˆØ×%Ò% aÑ(Ô(ˆ Ø �IŠI�d—k’k )Ñ,Ô,Ñ -Ô -ˆØ �KŠK˜×)Ò)¨!Ñ,Ô,Ñ -Ô -ˆØ ˆU‰
ˆØ �IŠI�a‰LŒLˆØˆr> )r� r r r r0 rS r r s @r= r) r) † sS ø€ € € € € ğ! ğ !ğ !ğ !ğ !ğ !ğ !ğğ ğ ğ ğ ğ ğ r> r) )*r rU rÛ Útorchr1 Útorch.nn.functionalr Ú
functionalrZ r† r Ú$mu_alpha_zero.AlphaZero.Network.nnetr r: r Ú$mu_alpha_zero.AlphaZero.checkpointerr Ú mu_alpha_zero.AlphaZero.loggerr Ú mu_alpha_zero.General.memoryr Ú mu_alpha_zero.General.mz_gamer
Ú mu_alpha_zero.General.networkr Ú mu_alpha_zero.Hooks.hook_managerr Ú mu_alpha_zero.Hooks.hook_pointr
Úmu_alpha_zero.MuZero.utilsr r r Úmu_alpha_zero.configr Ú$mu_alpha_zero.shared_storage_managerr ÚModuler r5 r9 r) r© r> r= ú<module>ry s ğØ € € € à Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ğ Ø € € € Ø Ğ Ğ Ğ Ğ Ğ Ø (Ğ (Ğ (Ğ (Ğ (Ğ (à gĞ gĞ gĞ gĞ gĞ gĞ gĞ gØ =Ğ =Ğ =Ğ =Ğ =Ğ =Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ø 4Ğ 4Ğ 4Ğ 4Ğ 4Ğ 4Ø >Ğ >Ğ >Ğ >Ğ >Ğ >Ø 8Ğ 8Ğ 8Ğ 8Ğ 8Ğ 8Ø 1Ğ 1Ğ 1Ğ 1Ğ 1Ğ 1Ø hĞ hĞ hĞ hĞ hĞ hĞ hĞ hĞ hĞ hØ -Ğ -Ğ -Ğ -Ğ -Ğ -Ø >Ğ >Ğ >Ğ >Ğ >Ğ >ğHoğ Hoğ Hoğ Hoğ Ho�”” Ğ 2ñ Hoô Hoğ HoğV$$ğ $$ğ $$ğ $$ğ $$˜œœ ñ $$ô $$ğ $$ğN:/ğ :/ğ :/ğ :/ğ :/�"”)ñ :/ô :/ğ :/ğzğ ğ ğ ğ �B”E”Lñ ô ğ ğ ğ r> | 33,954 | Python | .pyt | 94 | 360.12766 | 2,500 | 0.306695 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,581 | hook_manager.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/__pycache__/hook_manager.cpython-311.pyc | §
føf� ã ó< — d dl mZ d dlmZmZ G d„ d¦ « ZdS )é )Ú HookCallable)Ú HookPointÚHookAtc óB — e Zd Zd„ Zdedefd„Zd
dededed e d
e
f
d „Z d S )Ú HookManagerc ó — i | _ d S ©N©Úhooks)Úselfs úO/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Hooks/hook_manager.pyÚ__init__zHookManager.__init__ s
€ ؈Œ
ˆ
ˆ
ó ÚwhereÚhookc ó — || j |< d S r r
)r r r s r
Úregister_method_hookz HookManager.register_method_hook s € Ø ˆŒ
�5ÑĞĞr © ÚclsÚfn_nameÚfileÚatÚargsc óò — | dd¦ « d¦ « d }| j ¦ « D ]-\ }}| |||¦ « r |j |g|¢R � d S Œ.d S )Nú\ú/éÿÿÿÿ)ÚreplaceÚsplitr ÚitemsÚhereÚexecute) r r r r r r Ú file_nameÚ
hook_pointÚ
hook_callables r
Úprocess_hook_executesz!HookManager.process_hook_executes s“ € Ø—L’L sÑ+Ô+×1Ò1°#Ñ6Ô6°rÔ:ˆ Ø)-¬×)9Ò)9Ñ);Ô);ğ ğ Ñ %ˆJ˜
Ø�Š˜y¨'°2Ñ6Ô6ğ
Ø%�
Ô% cĞ1¨DĞ1Ğ1Ğ1Ğ1Ø��ğ
ğ ğ r N)r ) Ú__name__Ú
__module__Ú __qualname__r r r r ÚobjectÚstrr Útupler& r r r
r r s‚ € € € € € ğğ ğ ğ!¨)ğ !¸<ğ !ğ !ğ !ğ !ğğ ¨ğ ¸#ğ ÀSğ Èfğ Ğ\ağ ğ ğ ğ ğ ğ r r N)Ú"mu_alpha_zero.Hooks.hook_callablesr Ú mu_alpha_zero.Hooks.hook_pointr r r r r r
ú<module>r/ sa ğØ ;Ğ ;Ğ ;Ğ ;Ğ ;Ğ ;Ø <Ğ <Ğ <Ğ <Ğ <Ğ <Ğ <Ğ <ğ ğ ğ ğ ğ ñ ô ğ ğ ğ r | 1,829 | Python | .pyt | 12 | 151.333333 | 480 | 0.388889 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,582 | hook_callables.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/__pycache__/hook_callables.cpython-311.pyc | §
føf ã óZ — d dl mZmZ d dlmZ G d„ de¦ « Z G d„ de¦ « ZdS )é )ÚabstractmethodÚABC)ÚCallablec ó* — e Zd Zedefd„¦ « ZdS )Ú HookCallableÚclsc ó — dS )zJ
Execute the hook
:return: The return of the hook
N© ©Úselfr Úargss úQ/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Hooks/hook_callables.pyÚexecutezHookCallable.execute s € ğ
ˆó N)Ú__name__Ú
__module__Ú __qualname__r Úobjectr r
r r r r s9 € € € € € àğ
˜6ğ
ğ
ğ
ñ „^ğ
ğ
ğ
r r c ó4 ‡ — e Zd Zdedefˆ fd„ Zdefd„Zˆ xZS )ÚHookMethodCallableÚmethodr
c ód •— t ¦ « ¦ « || _ || _ d S ©N)ÚsuperÚ__init__Ú_HookMethodCallable__methodÚ_HookMethodCallable__args)r r r
Ú __class__s €r r zHookMethodCallable.__init__ s+ ø€ İ
‰Œ×ÒÑÔĞØ ˆŒ
؈Œ ˆ ˆ r r c ó, — | j |g| j ¢|¢R � S r )r r r s r r zHookMethodCallable.execute s# € Ø ˆtŒ}˜SĞ6 4¤;Ğ6°Ğ6Ğ6Ğ6Ğ6r ) r r r r Útupler r r Ú
__classcell__)r s @r r r sf ø€ € € € € ğ˜xğ ¨uğ ğ ğ ğ ğ ğ ğ
7˜6ğ 7ğ 7ğ 7ğ 7ğ 7ğ 7ğ 7ğ 7r r N)Úabcr r Útypingr r r r
r r ú<module>r$ s� ğØ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ğ #Ø Ğ Ğ Ğ Ğ Ğ ğ
ğ
ğ
ğ
ğ
�3ñ
ô
ğ
ğ7ğ 7ğ 7ğ 7ğ 7˜ ñ 7ô 7ğ 7ğ 7ğ 7r | 1,774 | Python | .pyt | 9 | 193.333333 | 893 | 0.378256 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,583 | hook_point.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Hooks/__pycache__/hook_point.cpython-311.pyc | §
føf« ã óN — d dl Z G d„ de j ¦ « Z G d„ d¦ « ZdS )é Nc ó — e Zd ZdZdZdZdZdS )ÚHookAtÚheadÚtailÚmiddleÚallN)Ú__name__Ú
__module__Ú __qualname__ÚHEADÚTAILÚMIDDLEÚALL© ó úM/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Hooks/hook_point.pyr r s" € € € € € Ø €DØ €DØ
€FØ
€C€C€Cr r c óŠ — e Zd Zdededefd„Zedefd„¦ « Zedefd„¦ « Zedefd„¦ « Z ded edefd
„Z
d S ) Ú HookPointÚatÚfileÚfn_namec ó0 — || _ || _ || _ d S ©N)Ú_HookPoint__atÚ_HookPoint__fileÚ_HookPoint__function_name)Úselfr r r s r Ú__init__zHookPoint.__init__ s € ؈Œ ؈Œ Ø &ˆÔ Ğ Ğ r Úreturnc ó — | j S r )r ©r s r r z HookPoint.at s
€ àŒyĞr c ó — | j S r )r r! s r r zHookPoint.file s
€ àŒ{Ğr c ó — | j S r )r r! s r Ú
function_namezHookPoint.function_name s € àÔ#Ğ#r r$ c ód — || j k o%|| j k o|| j k p|t j k S r )r r$ r r r )r r r$ r s r ÚherezHookPoint.here s8 € Ø�t”yÒ Ğp ]°dÔ6HÒ%HĞpÈbĞTXÔT[ÊmĞNoĞ_aÕekÔeoÒ_oĞpr N) r r
r r Ústrr Úpropertyr r r$ r& r r r r r sä € € € € € ğ'˜6ğ '¨ğ '°sğ 'ğ 'ğ 'ğ 'ğ
ğ�Fğ ğ ğ ñ „Xğğ ğ�cğ ğ ğ ñ „Xğğ ğ$˜sğ $ğ $ğ $ñ „Xğ$ğq˜ğ q¨Sğ q°fğ qğ qğ qğ qğ qğ qr r )ÚenumÚEnumr r r r r ú<module>r+ sv ğØ € € € ğğ ğ ğ ğ ˆTŒYñ ô ğ ğqğ qğ qğ qğ qñ qô qğ qğ qğ qr | 2,064 | Python | .pyt | 10 | 205.3 | 453 | 0.345985 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,584 | tictactoe_game.py | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/tictactoe_game.py | import io
import random
import sys
import numpy as np
import pygame as pg
import torch as th
from PIL import Image
from mu_alpha_zero.General.az_game import AlphaZeroGame
class TicTacToeGameManager(AlphaZeroGame):
"""
This class is the game manager for the game of Tic Tac Toe and its variants.
"""
def __init__(self, board_size: int, headless: bool, num_to_win=None) -> None:
# TODO: Implement the possibility to play over internet using sockets.
self.player = 1
self.enemy_player = -1
self.board_size = board_size
self.board = self.initialize_board()
self.headless = headless
self.num_to_win = self.init_num_to_win(num_to_win)
self.screen = self.create_screen(headless)
def play(self, player: int, index: tuple) -> None:
self.board[index] = player
def init_num_to_win(self, num_to_win: int | None) -> int:
if num_to_win is None:
num_to_win = self.board_size
if num_to_win > self.board_size:
raise Exception("Num to win can't be greater than board size")
return num_to_win
def initialize_board(self):
board = np.zeros((self.board_size, self.board_size), dtype=np.int8)
return board
def get_random_valid_action(self, observations: np.ndarray, **kwargs) -> list:
valid_moves = self.get_valid_moves(observations)
if len(valid_moves) == 0:
raise Exception("No valid moves")
return random.choice(valid_moves)
def create_screen(self, headless):
if headless:
return
pg.init()
pg.display.set_caption("Tic Tac Toe")
board_rect_size = self.board_size * 100
screen = pg.display.set_mode((board_rect_size, board_rect_size))
return screen
def full_iterate_array(self, arr: np.ndarray, func: callable) -> list:
"""
This function iterates over all rows, columns and the two main diagonals of supplied array,
applies the supplied function to each of them and returns the results in a list.
:param arr: a 2D numpy array.
:param func: a callable function that takes a 1D array as input and returns a result.
:return: A list of results.
"""
results = []
results.append("row")
for row in arr:
results.append(func(row.reshape(-1)))
results.append("col")
for col in arr.T:
results.append(func(col.reshape(-1)))
diags = [np.diag(arr, k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
flipped_diags = [np.diag(np.fliplr(arr), k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
diags.extend(flipped_diags)
for idx, diag in enumerate(diags):
if idx == 0:
results.append("diag_left")
elif idx == len(diags) // 2:
results.append("diag_right")
results.append(func(diag.reshape(-1)))
# results.append("diag_left")
# results.append(func(arr.diagonal().reshape(-1)))
# results.append("diag_right")
# results.append(func(np.fliplr(arr).diagonal().reshape(-1)))
return results
def eval_board(self, board: np.ndarray, check_end: bool = True) -> int | None:
if self.is_board_full(board):
return 0
score = 0
for i in range(self.num_to_win, self.num_to_win // 2, -1):
current_won = self.check_partial_win(-1, self.num_to_win, board=board)
opp_won = self.check_partial_win(1, self.num_to_win, board=board)
if current_won:
score += 1 * 2
if opp_won:
score -= 1
if check_end:
return None if self.game_result(-1, board) is None else score
return score
def full_iterate_array_all_diags(self, arr: np.ndarray, func: callable):
results = []
for row in arr:
results.append(func(row.reshape(-1)))
for col in arr.T:
results.append(func(col.reshape(-1)))
diags = [np.diag(arr, k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
flipped_diags = [np.diag(np.fliplr(arr), k=i) for i in range(-arr.shape[0] + 1, arr.shape[1])]
diags.extend(flipped_diags)
for diag in diags:
# if diag.size < self.num_to_win:
# continue
results.append(func(diag.reshape(-1)))
return results
def extract_all_vectors(self, board: np.ndarray):
vectors = board.tolist() + board.T.tolist()
vectors.extend([np.diag(board, k=i) for i in range(-board.shape[0] + 1, board.shape[1])])
vectors.extend([np.diag(np.fliplr(board), k=i) for i in range(-board.shape[0] + 1, board.shape[1])])
# pad vectors with -3's to have the same length as the board size and stack them
vectors = [np.pad(vector, (0, self.board_size - len(vector)), constant_values=-3) for vector in vectors]
return np.array(vectors)
def check_win(self, player: int, board=None) -> bool:
# if self.num_to_win == self.board_size:
# return self.check_full_win(player, board=board)
# else:
return self.check_partial_win(player, self.num_to_win, board=board)
def check_full_win(self, player: int, board=None) -> bool:
"""
This function checks if the supplied player has won the game with a full win (num_to_win == board_size).
:param player: The player to check for (1 or -1).
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
"""
if board is None:
board = self.get_board()
matches = self.full_iterate_array(board, lambda part: np.all(part == player))
for match in matches:
if not isinstance(match, str) and match:
return True
return False
def check_partial_win(self, player: int, n: int, board=None) -> bool:
"""
This function checks if the supplied player has won the game with a partial win (num_to_win < board_size).
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
"""
if board is None:
board = self.get_board()
matches = self.full_iterate_array_all_diags(board,
lambda part:
np.convolve((part == player), np.ones(n, dtype=np.int8),
"valid"))
for match in matches:
if np.any(match == n):
return True
return False
def check_partial_win_vectorized(self, player: int, n: int, board=None) -> bool:
if board is None:
board = self.get_board()
vectors = th.tensor(self.extract_all_vectors(board)).unsqueeze(0)
weight = th.ones((1, 1, 1, n), dtype=th.long)
vectors_where_player = th.where(vectors == player, 1, 0).long()
res = th.nn.functional.conv2d(vectors_where_player, weight=weight)
return th.any(res == n).item()
def check_partial_win_to_index(self, player: int, n: int, board=None) -> dict[tuple, str] | dict:
"""
This variation of check_partial_win returns the index of the first partial win found.
The index being the index of the first piece in the winning sequence.
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: A dictionary containing the index and the position of the winning sequence.
"""
if board is None:
board = self.get_board()
indices = self.full_iterate_array(board,
lambda part:
np.convolve((part == player), np.ones(n, dtype=int),
"valid"))
indices = [x.tolist() if not isinstance(x, str) else x for x in indices]
pos = "row"
for vector in indices:
if isinstance(vector, str):
pos = vector
continue
for el in vector:
if el == n:
vector_index = indices.index(vector)
pos_index = indices.index(pos)
# num_strings_before = len([x for index,x in enumerate(indices) if isinstance(x,str) and index < pos_index])
element_index = vector_index - (pos_index + 1) if vector_index > pos_index else indices.index(
vector, pos_index)
diag_idx = {
"index": ((self.board_size - 1) - element_index, 0)} if element_index < self.board_size else {
"index": (0, element_index - self.board_size)}
diag_idx["pos"] = pos
match pos:
case "row":
return {"index": (element_index, 0), "pos": pos}
case "col":
return {"index": (0, element_index), "pos": pos}
case "diag_left":
return diag_idx
case "diag_right":
return diag_idx
# case "diag_right":
return {"index": None, "pos": None}
def return_index_if_valid(self, index: tuple, return_on_fail: tuple = ()) -> tuple:
if index[0] < 0 or index[0] >= self.board_size or index[1] < 0 or index[1] >= self.board_size:
return return_on_fail
return index
def make_fresh_instance(self):
return TicTacToeGameManager(self.board_size, self.headless, num_to_win=self.num_to_win)
def get_previous(self, index: tuple, pos: str, n: int):
if np.array(index).all() == 0:
return index
match pos:
case "row":
return self.return_index_if_valid((index[0], index[1] - n), return_on_fail=index)
case "col":
return self.return_index_if_valid((index[0] - n, index[1]), return_on_fail=index)
case "diag_left":
return self.return_index_if_valid((index[0] - n, index[1] - n), return_on_fail=index)
case "diag_right":
return self.return_index_if_valid((index[0] - n, index[1] + n), return_on_fail=index)
def get_next(self, index: tuple, pos: str, n: int):
if np.array(index).all() == self.board_size - 1:
return index
match pos:
case "row":
return self.return_index_if_valid((index[0], index[1] + n), return_on_fail=index)
case "col":
return self.return_index_if_valid((index[0] + n, index[1]), return_on_fail=index)
case "diag_left":
return self.return_index_if_valid((index[0] + n, index[1] + n), return_on_fail=index)
case "diag_right":
return self.return_index_if_valid((index[0] + n, index[1] - n), return_on_fail=index)
def is_board_full(self, board=None) -> bool:
if board is None:
board = self.get_board()
return np.all(board != 0)
def get_board(self):
return self.board.copy()
def reset(self):
self.board = self.initialize_board()
return self.board.copy()
def get_board_size(self):
return self.board_size
def render(self) -> bool:
if self.headless:
return False
self.screen.fill((0, 0, 0))
self._draw_board()
for row in range(self.board_size):
for col in range(self.board_size):
if self.board[row][col] == self.player:
self._draw_circle(col * 100 + 50, row * 100 + 50)
elif self.board[row][col] == self.enemy_player:
self._draw_cross(col * 100 + 50, row * 100 + 50)
pg.event.pump()
pg.display.flip()
return True
def _draw_circle(self, x, y) -> None:
if self.headless:
return
pg.draw.circle(self.screen, "green", (x, y), 40, 1)
def _draw_cross(self, x, y) -> None:
if self.headless:
return
pg.draw.line(self.screen, "red", (x - 40, y - 40), (x + 40, y + 40), 1)
pg.draw.line(self.screen, "red", (x + 40, y - 40), (x - 40, y + 40), 1)
def _draw_board(self):
for x in range(0, self.board_size * 100, 100):
for y in range(0, self.board_size * 100, 100):
pg.draw.rect(self.screen, (255, 255, 255), pg.Rect(x, y, 100, 100), 1)
def is_empty(self, index: tuple) -> bool:
return self.board[index] == 0
def get_valid_moves(self, observation: np.ndarray, player: int or None = None) -> list:
"""
Legal moves are the empty spaces on the board.
:param observation: A 2D numpy array representing the current state of the game.
:param player: The player to check for. Since the game is symmetric, this is ignored.
:return: A list of legal moves.
"""
legal_moves = []
observation = observation.reshape(self.board_size, self.board_size)
for row in range(self.board_size):
for col in range(self.board_size):
if observation[row][col] == 0:
legal_moves.append([row, col])
return legal_moves
def pygame_quit(self) -> bool:
if self.headless:
return False
pg.quit()
return True
def get_click_coords(self):
if self.headless:
return
mouse_pos = (x // 100 for x in pg.mouse.get_pos())
if pg.mouse.get_pressed()[0]: # left click
return mouse_pos
def get_human_input(self, board: np.ndarray):
if self.headless:
return
while True:
self.check_pg_events()
if self.get_click_coords() is not None:
x, y = self.get_click_coords()
if board[y][x] == 0:
return y, x
# time.sleep(1 / 60)
def check_pg_events(self):
if self.headless:
return
for event in pg.event.get():
if event.type == pg.QUIT:
self.pygame_quit()
sys.exit(0)
def network_to_board(self, move):
"""
Converts an integer move from the network to a board index.
:param move: An integer move selected from the network probabilities.
:return: A tuple representing the board index (int,int).
"""
return np.unravel_index(move, self.board.shape)
@staticmethod
def get_canonical_form(board, player) -> np.ndarray:
return board * player
def get_next_state(self, board: np.ndarray, action: int or tuple, player: int) -> np.ndarray:
if isinstance(action, int):
action = self.network_to_board(action)
board_ = board.copy()
board_[action] = player
return board_
def set_headless(self, val: bool):
self.headless = val
def get_invalid_actions(self, state: np.ndarray, player: int):
mask = np.where(state == 0, 1, 0)
return mask
def __str__(self):
return str(self.board).replace('1', 'X').replace('-1', 'O')
#
# if __name__ == "__main__":
# sample_arr = np.array([[-1,1,-1,1,1],[-1,1,1,-1,-1],[1,-1,1,1,0],[-1,1,1,-1,0],[0,0,-1,1,1]])
# game_manager = GameManager(5, True,num_to_win=3)
# res = game_manager.check_partial_win_vectorized(1,3,sample_arr)
# res = game_manager.game_result(-1,sample_arr)
# print(res)
# print(game_manager.check_partial_win(1,3,sample_arr))
| 16,109 | Python | .tac | 340 | 36.097059 | 128 | 0.567581 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,585 | tictactoe_game.cpython-311.pyc | Skirlax_MuAlphaZeroLibrary/mu_alpha_zero/Game/__pycache__/tictactoe_game.cpython-311.pyc | §
Uxf"D ã ó~ — d dl Z d dlZd dlZd dlmZ d dlZd dlZ d dl
Z d dl Z
d dlmZ d dlmZ G d„ de¦ « ZdS )é N)ÚImage)Ú
AlphaZeroGamec óŞ — e Zd ZdZd?dededdfd„Zdededdfd „Zd
edz defd „Z d „ Z
d
e j de
fd„Zd„ Zde j dede
fd„Zd@de j dededz fd„Zde j defd„Zde j fd„Zd?dedefd„Zd?dedefd„Zd?dededefd „Zd?dededefd „Zd?dededeeef ez fd „ZdAded edefd!„Z d"„ Z ded#edefd$„Z ded#edefd%„Z d?defd&„Z d'„ Z!d(„ Z"d)„ Z#defd*„Z$dBd+„Z%dBd,„Z&d-„ Z'dedefd.„Z(d?d/e j depdde
fd0„Z)defd1„Z*d2„ Z+de j fd3„Z,d4„ Z-d5„ Z.d6„ Z/e0de j fd7„¦ « Z1de j d8epedede j fd9„Z2d:efd;„Z3d<e j defd=„Z4d>„ Z5dS )CÚTicTacToeGameManagerzV
This class is the game manager for the game of Tic Tac Toe and its variants.
NÚ
board_sizeÚheadlessÚreturnc óØ — d| _ d| _ || _ | ¦ « | _ || _ | |¦ « | _ | |¦ « | _ d S )Né éÿÿÿÿ)
ÚplayerÚ enemy_playerr Úinitialize_boardÚboardr Úinit_num_to_winÚ
num_to_winÚ
create_screenÚscreen)Úselfr r r s úP/home/skyr/PycharmProjects/MuAlphaZeroBuild/mu_alpha_zero/Game/tictactoe_game.pyÚ__init__z TicTacToeGameManager.__init__ s` € àˆŒ Ø ˆÔØ$ˆŒØ×*Ò*Ñ,Ô,ˆŒ
Ø ˆŒ
Ø ×.Ò.¨zÑ:Ô:ˆŒØ×(Ò(¨Ñ2Ô2ˆŒ ˆ ˆ ó r
Úindexc ó — || j |< d S ©N©r )r r
r s r ÚplayzTicTacToeGameManager.play s € Ø "ˆŒ
�5ÑĞĞr r c óL — |€| j }|| j k rt d¦ « ‚|S )Nz+Num to win can't be greater than board size)r Ú Exception)r r s r r z$TicTacToeGameManager.init_num_to_win! s1 € Ø Ğ Ø œ ˆJØ ˜œÒ 'Ğ 'İĞ IÑJÔJĞ JØĞr c ó^ — t j | j | j ft j ¬¦ « }|S )N©Údtype)ÚnpÚzerosr Úint8©r r s r r z%TicTacToeGameManager.initialize_board( s& € İ”˜$œ/¨4¬?Ğ;Å2Ä7ĞKÑKÔKˆØˆ r Ú observationsc ó˜ — | |¦ « }t |¦ « dk rt d¦ « ‚t j |¦ « S )Nr zNo valid moves)Úget_valid_movesÚlenr ÚrandomÚchoice)r r' ÚkwargsÚ valid_movess r Úget_random_valid_actionz,TicTacToeGameManager.get_random_valid_action, sG € Ø×*Ò*¨<Ñ8Ô8ˆ İ ˆ{Ñ Ô ˜qÒ Ğ İĞ ,Ñ-Ô-Ğ -İŒ}˜[Ñ)Ô)Ğ)r c óÈ — |rd S t j ¦ « t j d¦ « | j dz }t j ||f¦ « }|S )Nz Tic Tac Toeéd )ÚpgÚinitÚdisplayÚ set_captionr Úset_mode)r r Úboard_rect_sizer s r r z"TicTacToeGameManager.create_screen2 s[ € Ø ğ Ø ˆFå
Œ‰ Œ ˆ İ
Œ
× Ò ˜}Ñ-Ô-Ğ-Ø œ/¨CÑ/ˆİ”×$Ò$ o°Ğ%GÑHÔHˆØˆ
r ÚarrÚfuncc ó^ ‡— g }| d¦ « ‰D ]3}| || d¦ « ¦ « ¦ « Œ4| d¦ « ‰j D ]3}| || d¦ « ¦ « ¦ « Œ4ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }| |¦ « t
|¦ « D ]}\ }} |dk r| d¦ « n+|t |¦ « d z k r| d
¦ « | || d¦ « ¦ « ¦ « Œ~|S ) an
This function iterates over all rows, columns and the two main diagonals of supplied array,
applies the supplied function to each of them and returns the results in a list.
:param arr: a 2D numpy array.
:param func: a callable function that takes a 1D array as input and returns a result.
:return: A list of results.
Úrowr Úcolc ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS ©)Úk©r# Údiag©Ú.0Úir8 s €r ú
<listcomp>z;TicTacToeGameManager.full_iterate_array.<locals>.<listcomp>M ó( ø€ ĞSĞSĞS q•”˜ Ğ"Ñ"Ô"ĞSĞSĞSr r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> ©r# rA ÚfliplrrB s €r rE z;TicTacToeGameManager.full_iterate_array.<locals>.<listcomp>N ó0 ø€ ĞfĞfĞf¸!�œ¥¤¨3¡ ¤ °1Ğ5Ñ5Ô5ĞfĞfĞfr Ú diag_lefté Ú
diag_right)ÚappendÚreshapeÚTÚrangeÚshapeÚextendÚ enumerater* )
r r8 r9 Úresultsr; r< ÚdiagsÚ
flipped_diagsÚidxrA s
` r Úfull_iterate_arrayz'TicTacToeGameManager.full_iterate_array< s¶ ø€ ğ ˆØ�Š�uÑ Ô Ğ Øğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1à�Š�uÑ Ô Ğ Ø”5ğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1àSĞSĞSĞSE°3´9¸Q´<°-À!Ñ2CÀSÄYÈqÄ\Ñ,RÔ,RĞSÑSÔSˆØfĞfĞfĞf½uÀcÄiĞPQÄlÀ]ĞUVÑEVĞX[ÔXaĞbcÔXdÑ?eÔ?eĞfÑfÔfˆ
Ø
� Š �]Ñ#Ô#Ğ#İ" 5Ñ)Ô)ğ 3ğ 3‰IˆC�Ø�aŠxˆxØ—’˜{Ñ+Ô+Ğ+Ğ+Ø�˜E™
œ
a™Ò'Ğ'Ø—’˜|Ñ,Ô,Ğ,Ø �NŠN˜4˜4 § ¢ ¨RÑ 0Ô 0Ñ1Ô1Ñ 2Ô 2Ğ 2Ğ 2ğ ˆr Tr Ú check_endc óB — | |¦ « rdS d}t | j | j dz d¦ « D ]J}| d| j |¬¦ « }| d| j |¬¦ « }|r|dz
}|r|dz }ŒK|r| d|¦ « €d n|S |S )Nr rL r r r )Ú
is_board_fullrQ r Úcheck_partial_winÚ game_result)r r rZ ÚscorerD Ú current_wonÚopp_wons r Ú
eval_boardz TicTacToeGameManager.eval_board] sÎ € Ø × Ò ˜eÑ $Ô $ğ Ø�1؈İ�t”¨¬¸1Ñ(<¸bÑAÔAğ ğ ˆAØ ×0Ò0°°T´_ÈEĞ0ÑRÔRˆKØ×,Ò,¨Q°´ÀuĞ,ÑMÔMˆGØğ
ؘ‘�Øğ
ؘ‘
�øØ ğ JØ ×+Ò+¨B°Ñ6Ô6Ğ>�4�4ÀEĞ I؈ r c ó\ ‡— g }‰D ]3}| || d¦ « ¦ « ¦ « Œ4‰j D ]3}| || d¦ « ¦ « ¦ « Œ4ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « }| |¦ « |D ]3}| || d¦ « ¦ « ¦ « Œ4|S )Nr c ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS r> r@ rB s €r rE zETicTacToeGameManager.full_iterate_array_all_diags.<locals>.<listcomp>u rF r r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> rH rB s €r rE zETicTacToeGameManager.full_iterate_array_all_diags.<locals>.<listcomp>v rJ r )rN rO rP rQ rR rS ) r r8 r9 rU r; r< rV rW rA s ` r Ú full_iterate_array_all_diagsz1TicTacToeGameManager.full_iterate_array_all_diagsl s? ø€ àˆØğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1à”5ğ 2ğ 2ˆCØ �NŠN˜4˜4 § ¢ ¨B¡¤Ñ0Ô0Ñ 1Ô 1Ğ 1Ğ 1àSĞSĞSĞSE°3´9¸Q´<°-À!Ñ2CÀSÄYÈqÄ\Ñ,RÔ,RĞSÑSÔSˆØfĞfĞfĞf½uÀcÄiĞPQÄlÀ]ĞUVÑEVĞX[ÔXaĞbcÔXdÑ?eÔ?eĞfÑfÔfˆ
Ø
� Š �]Ñ#Ô#Ğ#Øğ 3ğ 3ˆDğ
�NŠN˜4˜4 § ¢ ¨RÑ 0Ô 0Ñ1Ô1Ñ 2Ô 2Ğ 2Ğ 2àˆr c óÊ ‡ ‡— ‰ ¦ « ‰j ¦ « z }| ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « ¦ « | ˆfd„t ‰j d dz ‰j d ¦ « D ¦ « ¦ « ˆ fd„|D ¦ « }t j |¦ « S )Nc ó< •— g | ]}t j ‰|¬ ¦ « ‘ŒS r> r@ ©rC rD r s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>� s( ø€ Ğ`Ğ`Ğ`°�œ ¨Ğ+Ñ+Ô+Ğ`Ğ`Ğ`r r r c ó` •— g | ]*}t j t j ‰¦ « |¬ ¦ « ‘Œ+S r> rH ri s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>‚ s2 ø€ ĞkĞkĞk¸1�œ¥¤ ¨%Ñ 0Ô 0°AĞ6Ñ6Ô6ĞkĞkĞkr c
ól •— g | ]0}t j |d ‰j t |¦ « z
fd¬¦ « ‘Œ1S )r éıÿÿÿ)Úconstant_values)r# Úpadr r* )rC Úvectorr s €r rE z<TicTacToeGameManager.extract_all_vectors.<locals>.<listcomp>„ s> ø€ ĞpĞpĞpĞ^d•2”6˜& 1 d¤o½¸F¹ ¼ Ñ&CĞ"DĞVXĞYÑYÔYĞpĞpĞpr )ÚtolistrP rS rQ rR r# Úarray)r r Úvectorss `` r Úextract_all_vectorsz(TicTacToeGameManager.extract_all_vectors sŞ øø€ Ø—,’,‘.”. 5¤7§>¢>Ñ#3Ô#3Ñ3ˆØ�ŠĞ`Ğ`Ğ`Ğ`µU¸E¼KȼN¸?ÈQÑ;NĞPUÔP[Ğ\]ÔP^Ñ5_Ô5_Ğ`Ñ`Ô`ÑaÔaĞaØ�ŠĞkĞkĞkĞkÅÀuÄ{ĞSTÄ~ÀoĞXYÑFYĞ[`Ô[fĞghÔ[iÑ@jÔ@jĞkÑkÔkÑlÔlĞlàpĞpĞpĞpĞhoĞpÑpÔpˆİŒx˜Ñ Ô Ğ r c ó< — | || j |¬¦ « S )Nr )r] r )r r
r s r Ú check_winz TicTacToeGameManager.check_win‡ s! € ğ ×%Ò% f¨d¬oÀUĞ%ÑKÔKĞKr c ó¤ ‡— |€| ¦ « }| |ˆfd„¦ « }|D ] }t |t ¦ « s|r dS Œ dS )aC
This function checks if the supplied player has won the game with a full win (num_to_win == board_size).
:param player: The player to check for (1 or -1).
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
Nc ó4 •— t j | ‰k ¦ « S r )r# Úall)Úpartr
s €r ú<lambda>z5TicTacToeGameManager.check_full_win.<locals>.<lambda>– s ø€ ½b¼fÀTÈVÂ^Ñ>TÔ>T€ r TF)Ú get_boardrY Ú
isinstanceÚstr)r r
r ÚmatchesÚmatchs ` r Úcheck_full_winz#TicTacToeGameManager.check_full_win� sm ø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×)Ò)¨%Ğ1TĞ1TĞ1TĞ1TÑUÔUˆØ ğ ğ ˆEİ ˜e¥SÑ)Ô)ğ
¨eğ
Ø�t�tøàˆur Únc óª ‡‡— |€| ¦ « }| |ˆˆfd„¦ « }|D ] }t j |‰k ¦ « r dS Œ dS )aˆ
This function checks if the supplied player has won the game with a partial win (num_to_win < board_size).
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: True if the player has won, False otherwise.
Nc ót •— t j | ‰k t j ‰t j ¬¦ « d¦ « S ©Nr! Úvalid)r# ÚconvolveÚonesr% ©ry r� r
s €€r rz z8TicTacToeGameManager.check_partial_win.<locals>.<lambda>© s: ø€ İ46´KÀÈÂ ÕRTÔRYĞZ[ÕceÔcjĞRkÑRkÔRkØ@Gñ5Iô 5Iğ r TF)r{ rf r# Úany)r r
r� r r~ r s `` r r] z&TicTacToeGameManager.check_partial_win� s‹ øø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×3Ò3°Eğ5Iğ 5Iğ 5Iğ 5Iğ 5IñJô Jˆğ
ğ ğ ˆEİŒv�e˜q’jÑ!Ô!ğ
Ø�t�tğ
ğ ˆur c óâ — |€| ¦ « }t j | |¦ « ¦ « d¦ « }t j ddd|ft j ¬¦ « }t j ||k dd¦ « ¦ « }t j j
||¬¦ « }t j ||k ¦ « ¦ « S )Nr r r! )Úweight)
r{ ÚthÚtensorrs Ú unsqueezer‡ ÚlongÚwhereÚnnÚ
functionalÚconv2dr‰ Úitem)r r
r� r rr r‹ Úvectors_where_playerÚress r Ú check_partial_win_vectorizedz1TicTacToeGameManager.check_partial_win_vectorized³ s € Ø ˆ=Ø—N’NÑ$Ô$ˆEİ”)˜D× 4Ò 4°UÑ ;Ô ;Ñ<Ô<×FÒFÀqÑIÔIˆİ”˜!˜Q 1˜ R¬WĞ5Ñ5Ô5ˆİ !œx¨°6Ò(9¸1¸aÑ @Ô @× EÒ EÑ GÔ GĞ İŒeÔ ×%Ò%Ğ&:À6Ğ%ÑJÔJˆİŒv�c˜Q’hÑ Ô ×$Ò$Ñ&Ô&Ğ&r c óP ‡‡— |€| ¦ « }| |ˆˆfd„¦ « }d„ |D ¦ « }d}|D ]ß}t |t ¦ « r|}Œ|D ]Â}|‰k rº| |¦ « }| |¦ « } || k r|| dz z
n| || ¦ « }
|
| j k rd| j dz
|
z
dfin
dd|
| j z
fi} || d< |xdk r |
df|d œc c S xd
k r d|
f|d œc c S xd k r | c c S d k r| c c S ŒÃŒàddd œS )
aß
This variation of check_partial_win returns the index of the first partial win found.
The index being the index of the first piece in the winning sequence.
:param player: The player to check for (1 or -1).
:param n: The number of consecutive pieces needed to win.
:param board: The board to check on. If None, the current board is used.
:return: A dictionary containing the index and the position of the winning sequence.
Nc ój •— t j | ‰k t j ‰t ¬¦ « d¦ « S r„ )r# r† r‡ Úintrˆ s €€r rz zATicTacToeGameManager.check_partial_win_to_index.<locals>.<lambda>È s3 ø€ İ*,¬+°t¸v²~ÍÌĞPQÕY\ĞH]ÑH]ÔH]Ø6=ñ+?ô +?ğ r c ód — g | ]-}t |t ¦ « s| ¦ « n|‘Œ.S © )r| r} rp ©rC Úxs r rE zCTicTacToeGameManager.check_partial_win_to_index.<locals>.<listcomp>Ë s3 € ĞPĞPĞPÀ1¥Z°µ3Ñ%7Ô%7Ğ>�1—8’8‘:”:�:¸QĞPĞPĞPr r; r r r Úpos)r rŸ r< rK rM )r{ rY r| r} r r ) r r
r� r ÚindicesrŸ ro ÚelÚ vector_indexÚ pos_indexÚ
element_indexÚdiag_idxs `` r Úcheck_partial_win_to_indexz/TicTacToeGameManager.check_partial_win_to_index¼ sõ øø€ ğ ˆ=Ø—N’NÑ$Ô$ˆEØ×)Ò)¨%ğ+?ğ +?ğ +?ğ +?ğ +?ñ@ô @ˆğ QĞPÈĞPÑPÔPˆàˆØ ğ ,ğ ,ˆFݘ&¥#Ñ&Ô&ğ
Ø �ØØ ğ
,ğ
,�ؘ’7�7Ø#*§=¢=°Ñ#8Ô#8�LØ '§
¢
¨cÑ 2Ô 2�IàFRĞU^ÒF^ĞF^ L°IÀ±MÑ$BĞ$BĞdk×dqÒdqØ ñe+ô e+�Mğ Q^Ğ`dÔ`oÒPoĞPo˜ 4¤?°QÑ#6¸-Ñ"GÈĞ!Kğ Mğ Mà ! ]°T´_Ñ%DĞ!EğvGğ ğ '*�H˜U‘OØ Ø "˜UšU˜U˜UØ.;¸QĞ-?ÈĞ#LĞ#LĞ LĞ LĞ LĞ LĞ LØ "˜UšU˜U˜UØ./° Ğ-?ÈĞ#LĞ#LĞ LĞ LĞ LĞ LĞ LØ (˜[š[˜[˜[Ø#+˜O˜O˜O˜O˜OØ )š\˜\Ø#+˜O˜O˜O˜O˜Oøğ'
,ğ. dĞ+Ğ+Ğ+r rœ Úreturn_on_failc ó~ — |d dk s.|d | j k s |d dk s|d | j k r|S |S ©Nr r ©r )r r r§ s r Úreturn_index_if_validz*TicTacToeGameManager.return_index_if_validë sI € Ø �Œ8�aŠ<ˆ<˜5 œ8 t¤Ò6Ğ6¸%À¼(ÀQº,¸,È%ĞPQÌ(ĞVZÔVeÒJeĞJeØ!Ğ !؈ r c óD — t | j | j | j ¬¦ « S )N)r )r r r r ©r s r Úmake_fresh_instancez(TicTacToeGameManager.make_fresh_instanceğ s € İ# D¤O°T´]ÈtÌĞ_Ñ_Ô_Ğ_r rŸ c óà — t j |¦ « ¦ « dk r|S |xdk r) | |d |d |z
f|¬¦ « S xdk r) | |d |z
|d f|¬¦ « S xdk r, | |d |z
|d |z
f|¬¦ « S dk r+| |d |z
|d |z f|¬¦ « S d S )Nr r; r ©r§ r< rK rM )r# rq rx r« ©r r rŸ r� s r Ú get_previousz!TicTacToeGameManager.get_previousó s € İ
Œ8�E‰?Œ?× Ò Ñ Ô AÒ %Ğ %؈LØØ�’��Ø×1Ò1°5¸´8¸UÀ1¼Xȹ\Ğ2JĞ[`Ğ1ÑaÔaĞaØ�’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄĞ2JĞ[`Ğ1ÑaÔaĞaØ �’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeØ ’ � Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeğ � r c óğ — t j |¦ « ¦ « | j dz
k r|S |xdk r) | |d |d |z f|¬¦ « S xdk r) | |d |z |d f|¬¦ « S xdk r, | |d |z |d |z f|¬¦ « S dk r+| |d |z |d |z
f|¬¦ « S d S )Nr r; r r° r< rK rM )r# rq rx r r« r± s r Úget_nextz TicTacToeGameManager.get_next s € İ
Œ8�E‰?Œ?× Ò Ñ Ô D¤O°aÑ$7Ò 7Ğ 7؈LØØ�’��Ø×1Ò1°5¸´8¸UÀ1¼Xȹ\Ğ2JĞ[`Ğ1ÑaÔaĞaØ�’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄĞ2JĞ[`Ğ1ÑaÔaĞaØ �’��Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeØ ’ � Ø×1Ò1°5¸´8¸a±<ÀÀqÄÈAÁ Ğ2NĞ_dĞ1ÑeÔeĞeğ � r c ó^ — |€| ¦ « }t j |dk ¦ « S ©Nr )r{ r# rx r& s r r\ z"TicTacToeGameManager.is_board_full
s* € Ø ˆ=Ø—N’NÑ$Ô$ˆEİŒv�e˜q’jÑ!Ô!Ğ!r c ó4 — | j ¦ « S r )r Úcopyr s r r{ z TicTacToeGameManager.get_board s € ØŒz�ŠÑ Ô Ğ r c óf — | ¦ « | _ | j ¦ « S r )r r r¸ r s r ÚresetzTicTacToeGameManager.reset s' € Ø×*Ò*Ñ,Ô,ˆŒ
ØŒz�ŠÑ Ô Ğ r c ó — | j S r rª r s r Úget_board_sizez#TicTacToeGameManager.get_board_size s
€ ،Рr c óB — | j rdS | j d¦ « | ¦ « t | j ¦ « D ]–}t | j ¦ « D ]}| j | | | j k r#| |dz dz |dz dz ¦ « ŒA| j | | | j k r"|
|dz dz |dz dz ¦ « Œ€Œ—t j
¦ « t j ¦ « dS )NF)r r r r1 é2 T)r r ÚfillÚ _draw_boardrQ r r r
Ú _draw_circler Ú _draw_crossr2 ÚeventÚpumpr4 Úflip)r r; r< s r ÚrenderzTicTacToeGameManager.render s € Ø Œ=ğ Ø�5à Œ ×Ò˜Ñ#Ô#Ğ#Ø ×ÒÑÔĞݘœ Ñ)Ô)ğ Eğ EˆCİ ˜Tœ_Ñ-Ô-ğ
Eğ
E�Ø”:˜c”? 3Ô'¨4¬;Ò6Ğ6Ø×%Ò% c¨C¡i°"¡n°c¸C±iÀ"±nÑEÔEĞEĞEà”Z ”_ SÔ)¨TÔ->Ò>Ğ>Ø×$Ò$ S¨3¡Y°¡^°S¸3±YÀ±^ÑDÔDĞDøğ
Eõ Œ�
Š
‰Œˆİ
Œ
�ŠÑÔĞ؈tr c ól — | j rd S t j | j d||fdd¦ « d S )NÚgreené( r )r r2 ÚdrawÚcircler ©r r� Úys r rÁ z!TicTacToeGameManager._draw_circle. s9 € Ø Œ=ğ Ø ˆFİ
Œ�Š�t”{ G¨a°¨V°R¸Ñ;Ô;Ğ;Ğ;Ğ;r c óø — | j rd S t j | j d|dz
|dz
f|dz |dz fd¦ « t j | j d|dz |dz
f|dz
|dz fd¦ « d S )NÚredrÉ r )r r2 rÊ Úliner rÌ s r r z TicTacToeGameManager._draw_cross3 s‡ € Ø Œ=ğ Ø ˆFİ
Œ� Š �T”[ %¨!¨b©&°!°b±&Ğ)9¸AÀ¹FÀAÈÁFĞ;KÈQÑOÔOĞOİ
Œ� Š �T”[ %¨!¨b©&°!°b±&Ğ)9¸AÀ¹FÀAÈÁFĞ;KÈQÑOÔOĞOĞOĞOr c óî — t d| j dz d¦ « D ]Z}t d| j dz d¦ « D ]>}t j | j dt j ||dd¦ « d¦ « Œ?Œ[d S )Nr r1 )éÿ rÒ rÒ r )rQ r r2 rÊ Úrectr ÚRectrÌ s r rÀ z TicTacToeGameManager._draw_board9 sŒ € İ�q˜$œ/¨CÑ/°Ñ5Ô5ğ Wğ WˆAݘ1˜dœo°Ñ 3°SÑ9Ô9ğ
Wğ
W�İ”— ’ ˜Tœ[¨/½2¼7À1ÀaÈÈcÑ;RÔ;RĞTUÑVÔVĞVĞVğ
Wğ Wğ Wr c ó$ — | j | dk S r¶ r )r r s r Úis_emptyz TicTacToeGameManager.is_empty> s € ØŒz˜%Ô AÒ%Ğ%r Ú observationc óø — g }| | j | j ¦ « }t | j ¦ « D ]B}t | j ¦ « D ]+}|| | dk r| ||g¦ « Œ,ŒC|S )a
Legal moves are the empty spaces on the board.
:param observation: A 2D numpy array representing the current state of the game.
:param player: The player to check for. Since the game is symmetric, this is ignored.
:return: A list of legal moves.
r )rO r rQ rN )r r× r
Ú legal_movesr; r< s r r) z$TicTacToeGameManager.get_valid_movesA s� € ğ ˆ Ø!×)Ò)¨$¬/¸4¼?ÑKÔKˆ ݘœ Ñ)Ô)ğ 3ğ 3ˆCİ ˜Tœ_Ñ-Ô-ğ
3ğ
3�Ø ˜sÔ# CÔ(¨AÒ-Ğ-Ø ×&Ò&¨¨S zÑ2Ô2Ğ2øğ
3ğ Ğr c ó> — | j rdS t j ¦ « dS )NFT)r r2 Úquitr s r Ú pygame_quitz TicTacToeGameManager.pygame_quitP s! € Ø Œ=ğ Ø�5İ
Œ‰ Œ ˆ ؈tr c ó´ — | j rd S d„ t j ¦ « D ¦ « }t j ¦ « d r|S d S )Nc 3 ó K — | ] }|d z V — Œ
dS )r1 Nrœ r� s r ú <genexpr>z8TicTacToeGameManager.get_click_coords.<locals>.<genexpr>Y s& è è € Ğ:Ğ: !�Q˜#‘XĞ:Ğ:Ğ:Ğ:Ğ:Ğ:r r )r r2 ÚmouseÚget_posÚ get_pressed)r Ú mouse_poss r Úget_click_coordsz%TicTacToeGameManager.get_click_coordsV s] € Ø Œ=ğ Ø ˆFØ:Ğ:¥r¤x×'7Ò'7Ñ'9Ô'9Ğ:Ñ:Ô:ˆ İ
Œ8× Ò Ñ !Ô ! !Ô $ğ Ø Ğ ğ ğ r c ó — | j rd S | ¦ « | ¦ « �-| ¦ « \ }}|| | dk r||fS ŒV)NTr )r Úcheck_pg_eventsrä )r r r� rÍ s r Úget_human_inputz$TicTacToeGameManager.get_human_input] sp € Ø Œ=ğ Ø ˆFğ Ø × Ò Ñ "Ô "Ğ "Ø×$Ò$Ñ&Ô&Ğ2Ø×,Ò,Ñ.Ô.‘��1ؘ”8˜A”; !Ò#Ğ#Ø ˜a˜4�Kğ r c óÔ — | j rd S t j ¦ « D ]?}|j t j k r(| ¦ « t j d¦ « Œ@d S r¶ ) r r2 rà ÚgetÚtypeÚQUITrÜ ÚsysÚexit)r rà s r ræ z$TicTacToeGameManager.check_pg_eventsi sc € Ø Œ=ğ Ø ˆFİ”X—\’\‘^”^ğ ğ ˆEØŒz�RœWÒ$Ğ$Ø× Ò Ñ"Ô"Ğ"İ”˜‘ ” � øğ ğ r c ó@ — t j || j j ¦ « S )zÜ
Converts an integer move from the network to a board index.
:param move: An integer move selected from the network probabilities.
:return: A tuple representing the board index (int,int).
©r# Ú
unravel_indexr rR )r Úmoves r Únetwork_to_boardz%TicTacToeGameManager.network_to_boardq s € õ Ô d¤jÔ&6Ñ7Ô7Ğ7r c óŞ ‡ — ‰ j rd S t j d¬¦ « t | ¦ « � \ }}t |¦ « ˆ fd„|D ¦ « }t
j ||¬¦ « t j d¬¦ « t j d¦ « t j
d¦ « t j d ¦ « t j
¦ « }t j |d
d ¬ ¦ « | d
¦ « t! j |¦ « }t$ j ‰ j d¦ « }t! j d‰ j ¦ « |¦ « }|j |j z } t! j dt5 |j |j ¦ « | f¦ « }
|
|d¦ « |
|d
|j f¦ « |
|¦ « d S )N)é é
)Úfigsizec óª •— g | ]O}t j |‰j j ¦ « d › dt j |‰j j ¦ « d › �‘ŒPS )r ú;r rï )rC r� r s €r rE zKTicTacToeGameManager.save_screenshot_with_probabilities.<locals>.<listcomp> sf ø€ ğ ğ ğ Ğop•RÔ% a¨¬Ô)9Ñ:Ô:¸1Ô=ĞjĞjÅÔ@PĞQRĞTXÔT^ÔTdÑ@eÔ@eĞfgÔ@hĞjĞjğ ğ ğ r )r� rÍ éZ )ÚrotationÚMoveÚ ProbabilityzAction probabilitiesÚpngÚtight)ÚformatÚ bbox_inchesr ÚRGBAÚRGB)r r ) r ÚpltÚfigureÚzipÚitemsÚprintÚsnsÚbarplotÚxticksÚxlabelÚylabelÚtitleÚioÚBytesIOÚsavefigÚseekr Úopenr2 ÚimageÚtostringr Ú frombytesÚget_sizeÚheightÚnewÚmaxÚwidthÚpasteÚsave) r Ú action_probsÚpathÚlabelsÚ
probabilitiesÚbufÚplot_imgÚsurface_bufferÚ surface_imgÚ total_heightÚ combined_imgs ` r Ú"save_screenshot_with_probabilitiesz7TicTacToeGameManager.save_screenshot_with_probabilitiesy sÍ ø€ Ø Œ=ğ Ø ˆFİ Œ
˜8Ğ$Ñ$Ô$Ğ$İ # \×%7Ò%7Ñ%9Ô%9Ğ :Ñ ˆ�
İ
ˆmÑ Ô Ğ ğğ ğ ğ Øğñ ô ˆå Œ �f
Ğ.Ñ.Ô.Ğ.İ Œ
˜BĞ Ñ Ô Ğ İ Œ
�6ÑÔĞİ Œ
�=Ñ!Ô!Ğ!İ Œ Ğ(Ñ)Ô)Ğ)åŒj‰lŒlˆİ Œ �C °7Ğ;Ñ;Ô;Ğ;Ø �Š�‰ Œ ˆ İ”:˜c‘?”?ˆõ œ×*Ò*¨4¬;¸Ñ?Ô?ˆİ”o f¨d¬k×.BÒ.BÑ.DÔ.DÀnÑUÔUˆ 𠔨Ô);Ñ;ˆ İ ”y ¨X¬^¸[Ô=NÑ)OÔ)OĞQ]Ğ(^Ñ_Ô_ˆ Ø×Ò˜8 VÑ,Ô,Ğ,Ø×Ò˜;¨¨H¬OĞ(<Ñ=Ô=Ğ=Ø×Ò˜$Ñ Ô Ğ Ğ Ğ r c ó — | |z S r rœ )r r
s r Úget_canonical_formz'TicTacToeGameManager.get_canonical_form— s € à�v‰~Ğ r Úactionc óŒ — t |t ¦ « r| |¦ « }| ¦ « }|||< |S r )r| rš rò r¸ )r r r* r
Úboard_s r Úget_next_statez#TicTacToeGameManager.get_next_state› sB € İ �f�cÑ "Ô "ğ 3Ø×*Ò*¨6Ñ2Ô2ˆFØ—’‘ ” ˆØ ˆˆv‰Øˆ
r Úvalc ó — || _ d S r )r )r r. s r Ú set_headlessz!TicTacToeGameManager.set_headless¢ s
€ ؈Œ
ˆ
ˆ
r Ústatec ó: — t j |dk dd¦ « }|S r© )r# r� )r r1 r
Úmasks r Úget_invalid_actionsz(TicTacToeGameManager.get_invalid_actions¥ s € İŒx˜ š
A qÑ)Ô)ˆØˆ r c óz — t | j ¦ « dd¦ « dd¦ « S )NÚ1ÚXz-1ÚO)r} r Úreplacer s r Ú__str__z TicTacToeGameManager.__str__© s0 € İ�4”:‰Œ×&Ò& s¨CÑ0Ô0×8Ò8¸¸sÑCÔCĞCr r )T)rœ )r N)6Ú__name__Ú
__module__Ú __qualname__Ú__doc__rš Úboolr Útupler r r r# ÚndarrayÚlistr/ r ÚcallablerY rb rf rs ru r€ r] r— Údictr} r¦ r« r® r² r´ r\ r{ rº r¼ rÆ rÁ r rÀ rÖ r) rÜ rä rç ræ rò r' Ú staticmethodr) r- r0 r4 r: rœ r r r r s8 € € € € € ğğ ğ3ğ 3 3ğ 3°$ğ 3ÈDğ 3ğ 3ğ 3ğ 3ğ#˜3ğ # uğ #°ğ #ğ #ğ #ğ #ğ¨#°©*ğ ¸ğ ğ ğ ğ ğğ ğ ğ*°B´Jğ *ÈTğ *ğ *ğ *ğ *ğ ğ ğ ğ b¤jğ ¸ğ ÀTğ ğ ğ ğ ğB
ğ
¤
ğ
°tğ
ÀsÈTÁzğ
ğ
ğ
ğ
ğ °´
ğ À(ğ ğ ğ ğ ğ&!¨¬ğ !ğ !ğ !ğ !ğLğ L ğ L°Dğ Lğ Lğ Lğ Lğ ğ Sğ ¸ğ ğ ğ ğ ğ ğ ¨ğ °ğ ÀDğ ğ ğ ğ ğ,'ğ '°3ğ '¸3ğ 'Ètğ 'ğ 'ğ 'ğ 'ğ-,ğ -,°ğ -,¸ğ -,ÈTĞRWĞY\ĞR\ÔM]Ğ`dÑMdğ -,ğ -,ğ -,ğ -,ğ^ğ ¨5ğ À%ğ ĞQVğ ğ ğ ğ ğ
`ğ `ğ `ğ f %ğ f¨cğ f°cğ fğ fğ fğ fğ f˜eğ f¨#ğ f°#ğ fğ fğ fğ fğ"ğ "¨4ğ "ğ "ğ "ğ "ğ
!ğ !ğ !ğ!ğ !ğ !ğ ğ ğ ğ˜ğ ğ ğ ğ ğ$<ğ <ğ <ğ <ğ
Pğ Pğ Pğ Pğ Wğ Wğ Wğ
&˜eğ &¨ğ &ğ &ğ &ğ &ğ
ğ
¨2¬:ğ
¸s¸{Àdğ
ĞVZğ
ğ
ğ
ğ
ğ ˜Tğ ğ ğ ğ ğ ğ ğ ğ R¤Zğ ğ ğ ğ ğ ğ ğ ğ8ğ 8ğ 8ğ ğ ğ ğ< ğ ¨R¬Zğ ğ ğ ñ „\ğ ğ B¤J𠸸 ¸uğ Ècğ ĞVXÔV`ğ ğ ğ ğ ğ ğ ğ ğ ğ ğ¨¬ğ ¸Sğ ğ ğ ğ ğDğ Dğ Dğ Dğ Dr r )r r+ rì Úmatplotlib.pyplotÚpyplotr Únumpyr# Úpygamer2 Úseabornr ÚtorchrŒ ÚPILr Ú mu_alpha_zero.General.az_gamer r rœ r r ú<module>rN sÎ ğØ € € € Ø
€
€
€
Ø
€
€
€
à Ğ Ğ Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ø Ğ Ğ Ğ Ğ Ğ à 7Ğ 7Ğ 7Ğ 7Ğ 7Ğ 7ğ[Dğ [Dğ [Dğ [Dğ [D˜=ñ [Dô [Dğ [Dğ [Dğ [Dr | 31,194 | Python | .tac | 118 | 260.949153 | 2,695 | 0.344221 | Skirlax/MuAlphaZeroLibrary | 8 | 0 | 0 | GPL-2.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,586 | __init__.py | pblottiere_QSA/qsa-plugin/__init__.py | # -*- coding: utf-8 -*-
import os
import sys
import json
import time
import struct
import pickle
import socket
from osgeo import gdal
from pathlib import Path
from threading import Thread
from datetime import datetime
from qgis import PyQt
from qgis.utils import server_active_plugins
from qgis.server import QgsConfigCache, QgsServerFilter
from qgis.core import Qgis, QgsProviderRegistry, QgsApplication
LOG_MESSAGES = []
class ProbeFilter(QgsServerFilter):
def __init__(self, iface, task):
super().__init__(iface)
self.task = task
def onRequestReady(self) -> bool:
request = self.serverInterface().requestHandler()
params = request.parameterMap()
self.task["project"] = params.get("MAP", "")
self.task["service"] = params.get("SERVICE", "")
self.task["request"] = params.get("REQUEST", "")
self.task["start"] = datetime.now()
self.task["count"] += 1
return True
def onResponseComplete(self) -> bool:
self._clear_task()
return True
def onSendResponse(self) -> bool:
self._clear_task()
return True
def _clear_task(self):
count = self.task["count"]
self.task.clear()
self.task["count"] = count
def log_messages():
m = {}
m["logs"] = "\n".join(LOG_MESSAGES)
return m
def stats(task):
s = task
if "start" in s:
s["duration"] = int(
(datetime.now() - s["start"]).total_seconds() * 1000
)
return s
def metadata(iface) -> dict:
m = {}
m["plugins"] = server_active_plugins
m["versions"] = {}
m["versions"]["qgis"] = f"{Qgis.version().split('-')[0]}"
m["versions"]["qt"] = PyQt.QtCore.QT_VERSION_STR
m["versions"]["python"] = sys.version.split(" ")[0]
m["versions"]["gdal"] = gdal.__version__
m["providers"] = QgsProviderRegistry.instance().pluginList().split("\n")
m["cache"] = {}
m["cache"]["projects"] = []
for project in QgsConfigCache.instance().projects():
m["cache"]["projects"].append(Path(project.fileName()).name)
return m
def auto_connect(s: socket.socket, host: str, port: int) -> socket.socket:
while True:
print("Try to connect...", file=sys.stderr)
try:
s.connect((host, port))
break
except Exception as e:
if e.errno == 106:
s.close()
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
time.sleep(5)
print("Connected with QSA server", file=sys.stderr)
return s
def f(iface, host: str, port: int, task: dict) -> None:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s = auto_connect(s, host, port)
while True:
try:
data = s.recv(2000)
payload = {}
if b"metadata" in data:
payload = metadata(iface)
elif b"logs" in data:
payload = log_messages()
elif b"stats" in data:
payload = stats(task)
ser = pickle.dumps(payload)
s.sendall(struct.pack(">I", len(ser)))
s.sendall(ser)
except Exception as e:
print(e, file=sys.stderr)
s = auto_connect(s, host, port)
def capture_log_message(message, tag, level):
LOG_MESSAGES.append(message)
def serverClassFactory(iface):
QgsApplication.instance().messageLog().messageReceived.connect(
capture_log_message
)
host = str(os.environ.get("QSA_HOST", "localhost"))
port = int(os.environ.get("QSA_PORT", 9999))
task = {}
task["count"] = 0
t = Thread(
target=f,
args=(
iface,
host.replace('"', ""),
port,
task,
),
)
t.start()
iface.registerFilter(ProbeFilter(iface, task), 100)
| 3,856 | Python | .py | 118 | 25.474576 | 76 | 0.593463 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,587 | cli.py | pblottiere_QSA/qsa-cli/qsa/cli.py | # coding: utf8
import os
import json
import time
import click
import requests
from pathlib import Path
from tabulate import tabulate
QSA_URL = os.environ.get("QSA_SERVER_URL", "http://localhost:5000/")
@click.group()
def cli():
pass
@cli.command()
def ps():
"""
List QGIS Server instances
"""
url = f"{QSA_URL}/api/instances"
data = requests.get(url)
headers = ["INSTANCE ID", "IP", "STATUS"]
table = []
for s in data.json()["servers"]:
t = []
t.append(s["id"])
t.append(s["ip"])
t.append(f"Binded {s['binded']} seconds ago")
table.append(t)
print(tabulate(table, headers=headers))
@cli.command()
@click.argument("id")
def inspect(id):
"""
Returns metadata about a specific QGIS Server instance
"""
url = f"{QSA_URL}/api/instances/{id}"
data = requests.get(url)
print(json.dumps(data.json(), indent=2))
@cli.command()
@click.argument("id")
def logs(id):
"""
Returns logs of a specific QGIS Server instance
"""
url = f"{QSA_URL}/api/instances/{id}/logs"
data = requests.get(url)
print(data.json()["logs"])
@cli.command()
@click.argument("id", required=False)
def stats(id):
"""
Returns stats of QGIS Server instances
"""
ids = []
if id:
ids.append(id)
else:
url = f"{QSA_URL}/api/instances"
data = requests.get(url)
for s in data.json()["servers"]:
ids.append(s["id"])
headers = [
"INSTANCE ID",
"COUNT",
"TIME ",
"SERVICE",
"REQUEST",
"PROJECT",
]
try:
while 1:
table = []
for i in ids:
url = f"{QSA_URL}/api/instances/{i}/stats"
task = requests.get(url).json()
if "error" in task:
continue
t = []
t.append(i)
t.append(task["count"])
if "service" in task:
t.append(f"{task['duration']} ms")
t.append(task["service"])
t.append(task["request"])
p = Path(task["project"]).name
t.append(p)
else:
t.append("")
t.append("")
t.append("")
t.append("")
table.append(t)
s = tabulate(table, headers=headers)
os.system("cls" if os.name == "nt" else "clear")
print(s)
time.sleep(0.25)
except:
pass
| 2,610 | Python | .py | 97 | 18.463918 | 68 | 0.51027 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,588 | test_api_storage_filesystem.py | pblottiere_QSA/qsa-api/tests/test_api_storage_filesystem.py | import os
import unittest
from pathlib import Path
from .utils import TestClient
GPKG = Path(__file__).parent / "data.gpkg"
if "QSA_GPKG" in os.environ:
GPKG = os.environ["QSA_GPKG"]
GEOTIFF = Path(__file__).parent / "landsat_4326.tif"
if "QSA_GEOTIFF" in os.environ:
GEOTIFF = os.environ["QSA_GEOTIFF"]
TEST_PROJECT_0 = "qsa_test_project0"
TEST_PROJECT_1 = "qsa_test_project1"
class APITestCaseFilesystem(unittest.TestCase):
def setUp(self):
self.app = TestClient("/tmp/qsa/projects/qgis")
def test_projects(self):
# no projects
p = self.app.get("/api/projects/")
self.assertTrue(TEST_PROJECT_0 not in p.get_json())
self.assertTrue(TEST_PROJECT_1 not in p.get_json())
# add projects
data = {}
data["name"] = TEST_PROJECT_0
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "qsa_test_project1"
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
# 2 projects
p = self.app.get("/api/projects/")
self.assertTrue(TEST_PROJECT_0 in p.get_json())
self.assertTrue(TEST_PROJECT_1 in p.get_json())
# remove project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}")
self.assertEqual(p.status_code, 201)
# 1 projects
p = self.app.get("/api/projects/")
self.assertTrue(TEST_PROJECT_1 in p.get_json())
# get info about project
p = self.app.get(f"/api/projects/{TEST_PROJECT_1}")
j = p.get_json()
self.assertTrue("crs" in j)
self.assertTrue("creation_datetime" in j)
self.assertEqual(j["author"], "pblottiere")
self.assertEqual(j["storage"], "filesystem")
self.assertFalse("schema" in j)
# remove last project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_1}")
def test_vector_symbology_line(self):
# access symbol properties
p = self.app.get(
"/api/symbology/vector/line/single_symbol/line/properties"
)
j = p.get_json()
self.assertTrue("line_width" in j)
def test_vector_symbology_fill(self):
# list symbology for fill geometries
p = self.app.get(
"/api/symbology/vector/polygon/single_symbol/fill/properties"
)
j = p.get_json()
self.assertTrue("outline_style" in j)
def test_vector_symbology_marker(self):
# list symbology for marker geometries
p = self.app.get(
"/api/symbology/vector/point/single_symbol/marker/properties"
)
j = p.get_json()
self.assertTrue("outline_style" in j)
def test_vector_symbology_rendering(self):
p = self.app.get("/api/symbology/vector/rendering/properties")
j = p.get_json()
self.assertTrue("opacity" in j)
def test_raster_symbology_rendering(self):
p = self.app.get("/api/symbology/raster/rendering/properties")
j = p.get_json()
self.assertTrue("gamma" in j)
self.assertTrue("brightness" in j)
self.assertTrue("contrast" in j)
self.assertTrue("saturation" in j)
def test_raster_symbology_singlebandgray(self):
p = self.app.get("/api/symbology/raster/singlebandgray/properties")
j = p.get_json()
self.assertTrue("gray" in j)
self.assertTrue("contrast_enhancement" in j)
def test_raster_symbology_multibandcolor(self):
p = self.app.get("/api/symbology/raster/multibandcolor/properties")
j = p.get_json()
self.assertTrue("contrast_enhancement" in j)
def test_layers(self):
# add project
data = {}
data["name"] = TEST_PROJECT_0
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
# 0 layer
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers")
self.assertEqual(p.get_json(), [])
# add layer
data = {}
data["name"] = "layer0"
data["datasource"] = f"{GPKG}|layername=polygons"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer1"
data["datasource"] = f"{GPKG}|layername=lines"
# data["crs"] = 4326 # No CRS because it's optional
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer2"
data["datasource"] = f"{GPKG}|layername=points"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer3"
data["datasource"] = f"{GEOTIFF}"
data["crs"] = 4326
data["type"] = "raster"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
# 3 layers
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers")
self.assertEqual(
p.get_json(), ["layer0", "layer1", "layer2", "layer3"]
)
# layer metadata
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers/layer1")
j = p.get_json()
self.assertEqual(j["type"], "vector")
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers/layer2")
j = p.get_json()
self.assertEqual(j["valid"], True)
# remove layer0
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}/layers/layer0")
self.assertEqual(p.status_code, 201)
# 2 layer
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers")
self.assertEqual(p.get_json(), ["layer1", "layer2", "layer3"])
# remove last project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}")
def test_raster_style(self):
# add project
data = {}
data["name"] = TEST_PROJECT_0
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
# 0 style
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles")
self.assertEqual(p.get_json(), [])
# add multibandcolor style to project
data = {}
data["type"] = "raster"
data["name"] = "style_multibandcolor"
data["symbology"] = {"type": "multibandcolor"}
data["symbology"]["properties"] = {
"red": {"band": 1},
"blue": {"band": 1},
"green": {"band": 1},
}
data["rendering"] = {
"brightness": 10,
"gamma": 1.0,
"contrast": 3,
"saturation": 2,
}
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/styles", data)
self.assertEqual(p.status_code, 201)
p = self.app.get(
f"/api/projects/{TEST_PROJECT_0}/styles/style_multibandcolor"
)
self.assertTrue("rendering" in p.get_json())
self.assertTrue("symbology" in p.get_json())
self.assertTrue("properties" in p.get_json()["symbology"])
# 1 style
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles")
self.assertTrue("style_multibandcolor" in p.get_json())
# add raster layer
data = {}
data["name"] = "layer0"
data["datasource"] = f"{GEOTIFF}"
data["crs"] = 4326
data["type"] = "raster"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
# update layer's style
data = {}
data["current"] = True
data["name"] = "style_multibandcolor"
p = self.app.post(
f"/api/projects/{TEST_PROJECT_0}/layers/layer0/style", data
)
self.assertEqual(p.status_code, 201)
# check style for layers
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers/layer0")
j = p.get_json()
self.assertEqual(j["styles"], ["default", "style_multibandcolor"])
self.assertEqual(j["current_style"], "style_multibandcolor")
# remove style
p = self.app.delete(
f"/api/projects/{TEST_PROJECT_0}/styles/style_multibandcolor"
)
self.assertEqual(p.status_code, 415) # style still in use
# update layer's style
data = {}
data["current"] = True
data["name"] = "default"
p = self.app.post(
f"/api/projects/{TEST_PROJECT_0}/layers/layer0/style", data
)
self.assertEqual(p.status_code, 201)
# remove style
p = self.app.delete(
f"/api/projects/{TEST_PROJECT_0}/styles/style_multibandcolor"
)
self.assertEqual(p.status_code, 201)
# 0 style
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles")
self.assertEqual(p.get_json(), [])
# remove last project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}")
def test_vector_style(self):
# add project
data = {}
data["name"] = TEST_PROJECT_0
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
# 0 style
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles")
self.assertEqual(p.get_json(), [])
# add line style to project
data = {}
data["type"] = "vector"
data["name"] = "style_line"
data["symbology"] = {"type": "single_symbol", "symbol": "line"}
data["symbology"]["properties"] = {"line_width": 0.5}
data["rendering"] = {"opacity": 0.4}
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/styles", data)
self.assertEqual(p.status_code, 201)
# add fill style to project
data = {}
data["type"] = "vector"
data["name"] = "style_fill"
data["symbology"] = {"type": "single_symbol", "symbol": "fill"}
data["symbology"]["properties"] = {"outline_width": 0.5}
data["rendering"] = {}
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/styles", data)
self.assertEqual(p.status_code, 201)
# 2 styles
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles")
self.assertTrue("style_line" in p.get_json())
self.assertTrue("style_fill" in p.get_json())
# style line metadata
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles/style_line")
j = p.get_json()
self.assertTrue(j["symbology"]["properties"]["line_width"], 0.75)
# style fill metadata
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles/style_fill")
j = p.get_json()
self.assertTrue(j["symbology"]["properties"]["outline_width"], 0.75)
# add layers
data = {}
data["name"] = "layer0"
data["datasource"] = f"{GPKG}|layername=polygons"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer1"
data["datasource"] = f"{GPKG}|layername=lines"
data["crs"] = 32637
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
# add style to layers
data = {}
data["current"] = False
data["name"] = "style_fill"
p = self.app.post(
f"/api/projects/{TEST_PROJECT_0}/layers/layer0/style", data
)
self.assertEqual(p.status_code, 201)
data = {}
data["current"] = True
data["name"] = "style_line"
p = self.app.post(
f"/api/projects/{TEST_PROJECT_0}/layers/layer1/style", data
)
self.assertEqual(p.status_code, 201)
# check style for layers
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers/layer0")
j = p.get_json()
self.assertEqual(j["styles"], ["default", "style_fill"])
self.assertEqual(j["current_style"], "default")
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers/layer1")
j = p.get_json()
self.assertEqual(j["styles"], ["default", "style_line"])
self.assertEqual(j["current_style"], "style_line")
# remove style
p = self.app.delete(
f"/api/projects/{TEST_PROJECT_0}/styles/style_fill"
)
self.assertEqual(p.status_code, 201)
# 1 style
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles")
self.assertEqual(p.get_json(), ["style_line"])
# remove last project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}")
def test_default_style(self):
# add project
data = {}
data["name"] = TEST_PROJECT_0
data["author"] = "pblottiere"
data["storage"] = "filesystem"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
# default styles
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles/default")
self.assertEqual(
p.get_json(),
{
"line": "default",
"polygon": "default",
"point": "default",
},
)
# add line style to project
data = {}
data["type"] = "vector"
data["name"] = "style_line"
data["symbology"] = {"type": "single_symbol", "symbol": "line"}
data["symbology"]["properties"] = {
"line_width": 0.75,
"line_style": "dash",
"customdash": "10;3",
"use_custom_dash": "1",
"line_color": "#0055FF",
}
data["rendering"] = {}
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/styles", data)
self.assertEqual(p.status_code, 201)
# add fill style to project
data = {}
data["type"] = "vector"
data["name"] = "style_fill"
data["symbology"] = {"type": "single_symbol", "symbol": "fill"}
data["symbology"]["properties"] = {
"color": "#00BBBB",
"style": "cross",
"outline_width": 0.16,
"outline_color": "#002222",
}
data["rendering"] = {}
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/styles", data)
self.assertEqual(p.status_code, 201)
# add marker style to project
data = {}
data["type"] = "vector"
data["name"] = "style_marker"
data["symbology"] = {"type": "single_symbol", "symbol": "marker"}
data["symbology"]["properties"] = {
"color": "#00BBBB",
"name": "star",
"size": 6,
"angle": 45,
}
data["rendering"] = {}
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/styles", data)
self.assertEqual(p.status_code, 201)
# set default styles for polygons/fill symbol
data = {}
data["symbology"] = "single_symbol"
data["geometry"] = "polygon"
data["symbol"] = "fill"
data["style"] = "style_fill"
p = self.app.post(
f"/api/projects/{TEST_PROJECT_0}/styles/default", data
)
self.assertEqual(p.status_code, 201)
# set default styles for line/line symbol
data = {}
data["symbology"] = "single_symbol"
data["geometry"] = "line"
data["symbol"] = "line"
data["style"] = "style_line"
p = self.app.post(
f"/api/projects/{TEST_PROJECT_0}/styles/default", data
)
self.assertEqual(p.status_code, 201)
# set default styles for point/marker symbol
data = {}
data["symbology"] = "single_symbol"
data["geometry"] = "point"
data["symbol"] = "marker"
data["style"] = "style_marker"
p = self.app.post(
f"/api/projects/{TEST_PROJECT_0}/styles/default", data
)
self.assertEqual(p.status_code, 201)
# check default style
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/styles/default")
self.assertEqual(
p.get_json(),
{
"line": "style_line",
"polygon": "style_fill",
"point": "style_marker",
},
)
# add layer
data = {}
data["name"] = "layer0"
data["datasource"] = f"{GPKG}|layername=polygons"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer1"
data["datasource"] = f"{GPKG}|layername=lines"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer2"
data["datasource"] = f"{GPKG}|layername=points"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
# check if default style is applied when adding a new layer in the project
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers/layer0")
j = p.get_json()
self.assertEqual(j["styles"], ["default", "style_fill"])
self.assertEqual(j["current_style"], "style_fill")
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers/layer1")
j = p.get_json()
self.assertEqual(j["styles"], ["default", "style_line"])
self.assertEqual(j["current_style"], "style_line")
if not self.app.is_flask_client:
# save polygon layer as png
r = self.app.get(
f"/api/projects/{TEST_PROJECT_0}/layers/layer0/map"
)
with open(
f"/tmp/{TEST_PROJECT_0}_layer0_style_fill.png", "wb"
) as out_file:
out_file.write(r.resp.content)
# save line layer as png
r = self.app.get(
f"/api/projects/{TEST_PROJECT_0}/layers/layer1/map"
)
with open(
f"/tmp/{TEST_PROJECT_0}_layer1_style_line.png", "wb"
) as out_file:
out_file.write(r.resp.content)
# save point layer as png
r = self.app.get(
f"/api/projects/{TEST_PROJECT_0}/layers/layer2/map"
)
with open(
f"/tmp/{TEST_PROJECT_0}_layer2_style_marker.png", "wb"
) as out_file:
out_file.write(r.resp.content)
# remove last project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}")
if __name__ == "__main__":
unittest.main()
| 19,356 | Python | .py | 482 | 30.605809 | 82 | 0.5596 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,589 | test_api_storage_postgresql.py | pblottiere_QSA/qsa-api/tests/test_api_storage_postgresql.py | import os
import unittest
from pathlib import Path
from .utils import TestClient
GPKG = Path(__file__).parent / "data.gpkg"
if "QSA_GPKG" in os.environ:
GPKG = os.environ["QSA_GPKG"]
GEOTIFF = Path(__file__).parent / "landsat_4326.tif"
if "QSA_GEOTIFF" in os.environ:
GEOTIFF = os.environ["QSA_GEOTIFF"]
TEST_PROJECT_0 = "qsa_test_project0"
TEST_PROJECT_1 = "qsa_test_project1"
class APITestCasePostgresql(unittest.TestCase):
def setUp(self):
self.app = TestClient("/tmp/qsa/projects/qgis", "qsa_test")
def test_projects(self):
# no projects
p = self.app.get("/api/projects/")
self.assertTrue(TEST_PROJECT_0 not in p.get_json())
self.assertTrue(TEST_PROJECT_1 not in p.get_json())
# add projects
data = {}
data["name"] = TEST_PROJECT_0
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "qsa_test_project1"
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
# 2 projects
p = self.app.get("/api/projects/")
self.assertTrue(TEST_PROJECT_0 in p.get_json())
self.assertTrue(TEST_PROJECT_1 in p.get_json())
# remove project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}")
self.assertEqual(p.status_code, 201)
# 1 projects
p = self.app.get("/api/projects/")
self.assertTrue(TEST_PROJECT_1 in p.get_json())
# get info about project
p = self.app.get(f"/api/projects/{TEST_PROJECT_1}")
j = p.get_json()
self.assertTrue("crs" in j)
self.assertTrue("creation_datetime" in j)
self.assertEqual(j["author"], "pblottiere")
self.assertEqual(j["storage"], "postgresql")
self.assertEqual(j["schema"], "public")
self.assertEqual(j["crs"], "EPSG:3857")
# remove last project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_1}")
def test_layers(self):
# add project
data = {}
data["name"] = TEST_PROJECT_0
data["author"] = "pblottiere"
p = self.app.post("/api/projects/", data)
self.assertEqual(p.status_code, 201)
# 0 layer
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers")
self.assertEqual(p.get_json(), [])
# add layer
data = {}
data["name"] = "layer0"
data["datasource"] = f"{GPKG}|layername=polygons"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer1"
data["datasource"] = f"{GPKG}|layername=lines"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer2"
data["datasource"] = f"{GPKG}|layername=points"
data["crs"] = 4326
data["type"] = "vector"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
data = {}
data["name"] = "layer3"
data["datasource"] = f"{GEOTIFF}"
data["crs"] = 4326
data["type"] = "raster"
p = self.app.post(f"/api/projects/{TEST_PROJECT_0}/layers", data)
self.assertEqual(p.status_code, 201)
# 3 layers
p = self.app.get(f"/api/projects/{TEST_PROJECT_0}/layers")
self.assertEqual(
p.get_json(), ["layer0", "layer1", "layer2", "layer3"]
)
# remove last project
p = self.app.delete(f"/api/projects/{TEST_PROJECT_0}")
| 3,883 | Python | .py | 98 | 31.459184 | 73 | 0.585703 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,590 | utils.py | pblottiere_QSA/qsa-api/tests/utils.py | import os
import json
import shutil
import requests
from flask import Flask
from pathlib import Path
app = Flask(__name__)
from qsa_api.config import QSAConfig
from qsa_api.api.projects import projects
from qsa_api.api.symbology import symbology
app.register_blueprint(projects, url_prefix="/api/projects")
app.register_blueprint(symbology, url_prefix="/api/symbology")
class TestResponse:
def __init__(self, resp, flask_client):
self.flask_client = flask_client
self.resp = resp
@property
def status_code(self):
return self.resp.status_code
def get_json(self):
if self.flask_client:
return self.resp.get_json()
return self.resp.json()
class TestClient:
def __init__(self, projects_dir, projects_psql_service=""):
self.app = requests
self._url = ""
if "QSA_HOST" not in os.environ or "QSA_PORT" not in os.environ:
self.app = app.test_client()
# prepare app client
self.app = app.test_client()
os.environ["QSA_QGISSERVER_URL"] = "http://qgisserver/ogc/"
os.environ["QSA_QGISSERVER_PROJECTS_DIR"] = projects_dir
if projects_psql_service:
os.environ["QSA_QGISSERVER_PROJECTS_PSQL_SERVICE"] = (
projects_psql_service
)
self.app.application.config["CONFIG"] = QSAConfig()
self.app.application.config["DEBUG"] = True
# clear projects dir
tmpdir = Path("/tmp/qsa/projects")
shutil.rmtree(tmpdir, ignore_errors=True)
(tmpdir / "qgis").mkdir(parents=True, exist_ok=True)
(tmpdir / "mapproxy").mkdir(parents=True, exist_ok=True)
else:
host = os.environ["QSA_HOST"]
port = os.environ["QSA_PORT"]
self._url = f"http://{host}:{port}"
self.delete(f"/api/projects/{TEST_PROJECT_0}")
self.delete(f"/api/projects/{TEST_PROJECT_1}")
def post(self, url, data):
if self.is_flask_client:
r = self.app.post(
self.url(url),
data=json.dumps(data),
content_type="application/json",
)
else:
r = self.app.post(
self.url(url),
data=json.dumps(data),
headers={"Content-type": "application/json"},
)
return TestResponse(r, self.is_flask_client)
def delete(self, url):
r = self.app.delete(self.url(url))
return TestResponse(r, self.is_flask_client)
def get(self, url):
r = self.app.get(self.url(url))
return TestResponse(r, self.is_flask_client)
def url(self, url) -> str:
return f"{self._url}{url}"
@property
def is_flask_client(self) -> bool:
return not bool(self._url)
| 2,873 | Python | .py | 75 | 28.88 | 72 | 0.592513 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,591 | project.py | pblottiere_QSA/qsa-api/qsa_api/project.py | # coding: utf8
import sys
import shutil
import sqlite3
from pathlib import Path
from qgis.PyQt.QtCore import Qt, QDateTime
from qgis.core import (
Qgis,
QgsSymbol,
QgsProject,
QgsMapLayer,
QgsWkbTypes,
QgsFillSymbol,
QgsLineSymbol,
QgsApplication,
QgsVectorLayer,
QgsRasterLayer,
QgsMarkerSymbol,
QgsDateTimeRange,
QgsRasterMinMaxOrigin,
QgsContrastEnhancement,
QgsSingleSymbolRenderer,
QgsSimpleFillSymbolLayer,
QgsSimpleLineSymbolLayer,
QgsSimpleMarkerSymbolLayer,
QgsRasterLayerTemporalProperties,
)
from .mapproxy import QSAMapProxy
from .vector import VectorSymbologyRenderer
from .utils import StorageBackend, config, logger
from .raster import RasterSymbologyRenderer, RasterOverview
RENDERER_TAG_NAME = "renderer-v2" # constant from core/symbology/renderer.h
class QSAProject:
def __init__(self, name: str, schema: str = "public") -> None:
self.name: str = name
self.schema: str = "public"
if schema:
self.schema = schema
@property
def sqlite_db(self) -> Path:
p = self._qgis_project_dir / "qsa.db"
if not p.exists():
p.parent.mkdir(parents=True, exist_ok=True)
con = sqlite3.connect(p)
cur = con.cursor()
cur.execute("CREATE TABLE styles_default(geometry, style)")
cur.execute("INSERT INTO styles_default VALUES('line', 'default')")
cur.execute(
"INSERT INTO styles_default VALUES('polygon', 'default')"
)
cur.execute(
"INSERT INTO styles_default VALUES('point', 'default')"
)
con.commit()
con.close()
return p
@staticmethod
def projects(schema: str = "") -> list:
p = []
if StorageBackend.type() == StorageBackend.FILESYSTEM:
for i in QSAProject._qgis_projects_dir().glob("**/*.qgs"):
name = i.parent.name.replace(
QSAProject._qgis_project_dir_prefix(), ""
)
p.append(QSAProject(name))
else:
service = config().qgisserver_projects_psql_service
uri = f"postgresql:?service={service}&schema={schema}"
storage = (
QgsApplication.instance()
.projectStorageRegistry()
.projectStorageFromType("postgresql")
)
for pname in storage.listProjects(uri):
p.append(QSAProject(pname, schema))
return p
@property
def styles(self) -> list[str]:
s = []
for qml in self._qgis_project_dir.glob("**/*.qml"):
s.append(qml.stem)
self.debug(f"{len(s)} styles found")
return s
@property
def project(self) -> QgsProject:
project = QgsProject()
project.read(self._qgis_project_uri, Qgis.ProjectReadFlag.DontResolveLayers)
return project
@property
def layers(self) -> list:
layers = []
p = QgsProject()
p.read(self._qgis_project_uri, Qgis.ProjectReadFlag.DontResolveLayers)
for layer in p.mapLayers().values():
layers.append(layer.name())
self.debug(f"{len(layers)} layers found")
return layers
@property
def metadata(self) -> dict:
m = {}
p = QgsProject()
p.read(self._qgis_project_uri, Qgis.ProjectReadFlag.DontResolveLayers)
m["author"] = p.metadata().author()
m["creation_datetime"] = (
p.metadata().creationDateTime().toString(Qt.ISODate)
)
m["crs"] = p.crs().authid()
m["storage"] = StorageBackend.type().name.lower()
if StorageBackend.type() == StorageBackend.POSTGRESQL:
m["schema"] = self.schema
m["cache"] = "disabled"
if self._mapproxy_enabled:
m["cache"] = "mapproxy"
return m
def cache_metadata(self) -> (dict, str):
if self._mapproxy_enabled:
return QSAMapProxy(self.name).metadata(), ""
return {}, "Cache is disabled"
def cache_reset(self) -> (bool, str):
if self._mapproxy_enabled:
mp = QSAMapProxy(self.name)
rc, err = mp.read()
if not rc:
return False, err
p = QgsProject()
p.read(self._qgis_project_uri)
for layer in p.mapLayers().values():
t = layer.type()
bbox = QSAProject._layer_bbox(layer)
epsg_code = QSAProject._layer_epsg_code(layer)
mp.remove_layer(layer.name())
mp.add_layer(
layer.name(), bbox, epsg_code, t == Qgis.LayerType.Raster, None
)
mp.write()
return True, ""
return False, "Cache is disabled"
def style_default(self, geometry: str) -> bool:
con = sqlite3.connect(self.sqlite_db.as_posix())
cur = con.cursor()
sql = f"SELECT style FROM styles_default WHERE geometry = '{geometry}'"
res = cur.execute(sql)
default_style = res.fetchone()[0]
con.close()
return default_style
def style(self, name: str) -> (dict, str):
if name not in self.styles:
return {}, "Invalid style"
path = self._qgis_project_dir / f"{name}.qml"
if VectorSymbologyRenderer.style_is_vector(path):
return VectorSymbologyRenderer.style_to_json(path)
else:
return RasterSymbologyRenderer.style_to_json(path)
def style_update(self, geometry: str, style: str) -> None:
con = sqlite3.connect(self.sqlite_db.as_posix())
cur = con.cursor()
sql = f"UPDATE styles_default SET style = '{style}' WHERE geometry = '{geometry}'"
cur.execute(sql)
con.commit()
con.close()
def default_styles(self) -> list:
s = {}
s["polygon"] = self.style_default("polygon")
s["line"] = self.style_default("line")
s["point"] = self.style_default("point")
return s
def layer(self, name: str) -> dict:
project = QgsProject()
project.read(self._qgis_project_uri)
layers = project.mapLayersByName(name)
if layers:
layer = layers[0]
infos = {}
infos["name"] = layer.name()
infos["type"] = layer.type().name.lower()
if layer.type() == Qgis.LayerType.Vector:
infos["geometry"] = QgsWkbTypes.displayString(layer.wkbType())
elif layer.type() == Qgis.LayerType.Raster:
infos["bands"] = layer.bandCount()
infos["data_type"] = layer.dataProvider().dataType(1).name.lower()
infos["source"] = layer.source()
infos["crs"] = layer.crs().authid()
infos["current_style"] = layer.styleManager().currentStyle()
infos["styles"] = layer.styleManager().styles()
infos["valid"] = layer.isValid()
infos["bbox"] = layer.extent().asWktCoordinates()
return infos
return {}
def layer_update_style(
self, layer_name: str, style_name: str, current: bool
) -> (bool, str):
if layer_name not in self.layers:
return False, f"Layer '{layer_name}' does not exist"
if style_name != "default" and style_name not in self.styles:
return False, f"Style '{style_name}' does not exist"
flags = Qgis.ProjectReadFlags()
flags |= Qgis.ProjectReadFlag.ForceReadOnlyLayers
project = QgsProject()
project.read(self._qgis_project_uri, flags)
style_path = self._qgis_project_dir / f"{style_name}.qml"
layer = project.mapLayersByName(layer_name)[0]
if style_name not in layer.styleManager().styles():
self.debug(f"Add new style {style_name} in style manager")
l = layer.clone()
l.loadNamedStyle(style_path.as_posix()) # set "default" style
layer.styleManager().addStyle(
style_name, l.styleManager().style("default")
)
if current:
self.debug(f"Set default style {style_name}")
layer.styleManager().setCurrentStyle(style_name)
# refresh min/max for the current layer if necessary
# (because the style is built on an empty geotiff)
if layer.type() == QgsMapLayer.RasterLayer:
self.debug("Refresh symbology renderer min/max")
renderer = RasterSymbologyRenderer(layer.renderer().type())
renderer.refresh_min_max(layer)
if self._mapproxy_enabled:
self.debug("Clear MapProxy cache")
mp = QSAMapProxy(self.name)
mp.clear_cache(layer_name)
self.debug("Write project")
project.write()
return True, ""
def layer_exists(self, name: str) -> bool:
return bool(self.layer(name))
def remove_layer(self, name: str) -> bool:
# remove layer in qgis project
project = QgsProject()
project.read(self._qgis_project_uri, Qgis.ProjectReadFlag.DontResolveLayers)
ids = []
for layer in project.mapLayersByName(name):
ids.append(layer.id())
project.removeMapLayers(ids)
rc = project.write()
# remove layer in mapproxy config
if self._mapproxy_enabled:
mp = QSAMapProxy(self.name)
rc, err = mp.read()
if not rc:
self.debug(err)
return False
mp.remove_layer(name)
mp.write()
return rc
def exists(self) -> bool:
if StorageBackend.type() == StorageBackend.FILESYSTEM:
return self._qgis_project_dir.exists()
else:
service = config().qgisserver_projects_psql_service
uri = f"postgresql:?service={service}&schema={self.schema}"
storage = (
QgsApplication.instance()
.projectStorageRegistry()
.projectStorageFromType("postgresql")
)
projects = storage.listProjects(uri)
# necessary step if the project has been created without QSA
if self.name in projects:
self._qgis_projects_dir().mkdir(parents=True, exist_ok=True)
return self.name in projects and self._qgis_projects_dir().exists()
def create(self, author: str) -> (bool, str):
if self.exists():
return False
# create qgis directory for qsa sqlite database and .qgs file if
# filesystem storage based
self._qgis_project_dir.mkdir(parents=True, exist_ok=True)
# create qgis project
project = QgsProject()
m = project.metadata()
m.setAuthor(author)
project.setMetadata(m)
crs = project.crs()
crs.createFromString("EPSG:3857") # default to webmercator
project.setCrs(crs)
self.debug("Write QGIS project")
rc = project.write(self._qgis_project_uri)
# create mapproxy config file
if self._mapproxy_enabled:
self.debug("Write MapProxy configuration file")
mp = QSAMapProxy(self.name)
mp.create()
# init sqlite database
self.sqlite_db
return rc, project.error()
def remove(self) -> None:
# clear cache and stuff
for layer in self.layers:
self.remove_layer(layer)
# remove mapproxy config file
if self._mapproxy_enabled:
mp = QSAMapProxy(self.name)
mp.remove()
# remove qsa projects dir
shutil.rmtree(self._qgis_project_dir, ignore_errors=True)
# remove remove qgis prohect in db if necessary
if StorageBackend.type() == StorageBackend.POSTGRESQL:
storage = (
QgsApplication.instance()
.projectStorageRegistry()
.projectStorageFromType("postgresql")
)
storage.removeProject(self._qgis_project_uri)
def add_layer(
self,
datasource: str,
layer_type: str,
name: str,
epsg_code: int,
overview: bool,
datetime: QDateTime | None,
) -> (bool, str):
t = self._layer_type(layer_type)
if t is None:
return False, "Invalid layer type"
if name in self.layers:
return False, f"A layer {name} already exists"
provider = QSAProject._layer_provider(t, datasource)
lyr = None
if t == Qgis.LayerType.Vector:
self.debug("Init vector layer")
lyr = QgsVectorLayer(datasource, name, provider)
elif t == Qgis.LayerType.Raster:
self.debug("Init raster layer")
lyr = QgsRasterLayer(datasource, name, provider)
ovr = RasterOverview(lyr)
if overview:
if not ovr.is_valid():
self.debug("Build overviews")
rc, err = ovr.build()
if not rc:
return False, err
else:
self.debug("Overviews already exist")
if datetime:
self.debug("Activate temporal properties")
mode = QgsRasterLayerTemporalProperties.ModeFixedTemporalRange
props = lyr.temporalProperties()
props.setMode(mode)
dt_range = QgsDateTimeRange(datetime, datetime)
props.setFixedTemporalRange(dt_range)
props.setIsActive(True)
else:
return False, "Invalid layer type"
if lyr is None:
return False, "Invalid layer (None)"
if not lyr.isValid():
return False, f"Invalid layer ({lyr.error()})"
if epsg_code > 0:
crs = lyr.crs()
crs.createFromString(f"EPSG:{epsg_code}")
lyr.setCrs(crs)
if not lyr.isValid():
return False, f"Invalid layer ({lyr.error()})"
# create project
project = QgsProject()
project.read(self._qgis_project_uri, Qgis.ProjectReadFlag.DontResolveLayers)
project.addMapLayer(lyr)
self.debug("Write QGIS project")
project.write()
# set default style
if t == Qgis.LayerType.Vector:
self.debug("Set default style")
geometry = lyr.geometryType().name.lower()
default_style = self.style_default(geometry)
self.layer_update_style(name, default_style, True)
# add layer in mapproxy config file
if self._mapproxy_enabled:
self.debug("Update MapProxy configuration file")
bbox = QSAProject._layer_bbox(lyr)
epsg_code = QSAProject._layer_epsg_code(lyr)
if epsg_code < 0:
return False, f"Invalid CRS {lyr.crs().authid()}"
self.debug(f"EPSG code {epsg_code}")
mp = QSAMapProxy(self.name)
rc, err = mp.read()
if not rc:
return False, err
rc, err = mp.add_layer(
name, bbox, epsg_code, t == Qgis.LayerType.Raster, datetime
)
if not rc:
return False, err
mp.write()
return True, ""
def add_style(
self,
name: str,
layer_type: str,
symbology: dict,
rendering: dict,
) -> (bool, str):
t = self._layer_type(layer_type)
if t == Qgis.LayerType.Vector:
return self._add_style_vector(name, symbology, rendering)
elif t == Qgis.LayerType.Raster:
return self._add_style_raster(name, symbology, rendering)
elif t is None:
return False, "Invalid layer type"
def _add_style_raster(
self, name: str, symbology: dict, rendering: dict
) -> (bool, str):
# safety check
if "type" not in symbology:
return False, "`type` is missing in `symbology`"
if "properties" not in symbology:
return False, "`properties` is missing in `symbology`"
# init renderer
tif = Path(__file__).resolve().parent / "raster" / "empty.tif"
rl = QgsRasterLayer(tif.as_posix(), "", "gdal")
# symbology
renderer = RasterSymbologyRenderer(symbology["type"])
renderer.load(symbology["properties"])
# config rendering
if "gamma" in rendering:
rl.brightnessFilter().setGamma(float(rendering["gamma"]))
if "brightness" in rendering:
rl.brightnessFilter().setBrightness(int(rendering["brightness"]))
if "contrast" in rendering:
rl.brightnessFilter().setContrast(int(rendering["contrast"]))
if "saturation" in rendering:
rl.hueSaturationFilter().setSaturation(
int(rendering["saturation"])
)
# save style
if renderer.renderer:
rl.setRenderer(renderer.renderer)
# contrast enhancement needs to be managed after setting renderer
if renderer.contrast_algorithm:
rl.setContrastEnhancement(
renderer.contrast_algorithm, renderer.contrast_limits
)
# user defined min/max
if (
renderer.contrast_limits
== QgsRasterMinMaxOrigin.Limits.None_
):
if (
renderer.type
== RasterSymbologyRenderer.Type.SINGLE_BAND_GRAY
):
ce = QgsContrastEnhancement(
rl.renderer().contrastEnhancement()
)
if renderer.gray_min is not None:
ce.setMinimumValue(renderer.gray_min)
if renderer.gray_max is not None:
ce.setMaximumValue(renderer.gray_max)
rl.renderer().setContrastEnhancement(ce)
elif (
renderer.type
== RasterSymbologyRenderer.Type.MULTI_BAND_COLOR
):
# red
red_ce = QgsContrastEnhancement(
rl.renderer().redContrastEnhancement()
)
if renderer.red_min is not None:
red_ce.setMinimumValue(renderer.red_min)
if renderer.red_max is not None:
red_ce.setMaximumValue(renderer.red_max)
rl.renderer().setRedContrastEnhancement(red_ce)
# green
green_ce = QgsContrastEnhancement(
rl.renderer().greenContrastEnhancement()
)
if renderer.green_min is not None:
green_ce.setMinimumValue(renderer.green_min)
if renderer.green_max is not None:
green_ce.setMaximumValue(renderer.green_max)
rl.renderer().setGreenContrastEnhancement(green_ce)
# blue
blue_ce = QgsContrastEnhancement(
rl.renderer().blueContrastEnhancement()
)
if renderer.blue_min is not None:
blue_ce.setMinimumValue(renderer.blue_min)
if renderer.blue_max is not None:
blue_ce.setMaximumValue(renderer.blue_max)
rl.renderer().setBlueContrastEnhancement(blue_ce)
# save
path = self._qgis_project_dir / f"{name}.qml"
rl.saveNamedStyle(
path.as_posix(), categories=QgsMapLayer.AllStyleCategories
)
return True, ""
return False, "Error"
def _add_style_vector(
self, name: str, symbology: dict, rendering: dict
) -> (bool, str):
if "type" not in symbology:
return False, "`type` is missing in `symbology`"
if "symbol" not in symbology:
return False, "`symbol` is missing in `symbology`"
if "properties" not in symbology:
return False, "`properties` is missing in `symbology`"
if symbology["type"] != "single_symbol":
return False, "Invalid symbol"
r = None
vl = QgsVectorLayer()
symbol = symbology["symbol"]
properties = symbology["properties"]
if symbol == "line":
r = QgsSingleSymbolRenderer(
QgsSymbol.defaultSymbol(QgsWkbTypes.LineGeometry)
)
props = QgsSimpleLineSymbolLayer().properties()
for key in properties.keys():
if key not in props:
return False, "Invalid properties"
symbol = QgsLineSymbol.createSimple(properties)
r.setSymbol(symbol)
elif symbol == "fill":
r = QgsSingleSymbolRenderer(
QgsSymbol.defaultSymbol(QgsWkbTypes.PolygonGeometry)
)
props = QgsSimpleFillSymbolLayer().properties()
for key in properties.keys():
if key not in props:
return False, "Invalid properties"
symbol = QgsFillSymbol.createSimple(properties)
r.setSymbol(symbol)
elif symbol == "marker":
r = QgsSingleSymbolRenderer(
QgsSymbol.defaultSymbol(QgsWkbTypes.PointGeometry)
)
props = QgsSimpleMarkerSymbolLayer().properties()
for key in properties.keys():
if key not in props:
return False, "Invalid properties"
symbol = QgsMarkerSymbol.createSimple(properties)
r.setSymbol(symbol)
if "opacity" in rendering:
vl.setOpacity(float(rendering["opacity"]))
if r:
vl.setRenderer(r)
path = self._qgis_project_dir / f"{name}.qml"
vl.saveNamedStyle(
path.as_posix(), categories=QgsMapLayer.Symbology
)
return True, ""
return False, "Error"
def remove_style(self, name: str) -> bool:
if name not in self.styles:
return False, f"Style '{name}' does not exist"
p = QgsProject()
p.read(self._qgis_project_uri)
for layer in p.mapLayers().values():
if name == layer.styleManager().currentStyle():
return False, f"Style is used by {layer.name()}"
for layer in p.mapLayers().values():
layer.styleManager().removeStyle(name)
path = self._qgis_project_dir / f"{name}.qml"
path.unlink()
p.write()
return True, ""
def debug(self, msg: str) -> None:
caller = f"{self.__class__.__name__}.{sys._getframe().f_back.f_code.co_name}"
if StorageBackend.type() == StorageBackend.FILESYSTEM:
msg = f"[{caller}][{self.name}] {msg}"
else:
msg = f"[{caller}][{self.schema}:{self.name}] {msg}"
logger().debug(msg)
@staticmethod
def _qgis_projects_dir() -> Path:
return Path(config().qgisserver_projects_dir)
@staticmethod
def _layer_type(layer_type: str) -> Qgis.LayerType | None:
if layer_type.lower() == "vector":
return Qgis.LayerType.Vector
elif layer_type.lower() == "raster":
return Qgis.LayerType.Raster
return None
@property
def _mapproxy_enabled(self) -> bool:
return bool(config().mapproxy_projects_dir)
@property
def _qgis_project_dir(self) -> Path:
return (
self._qgis_projects_dir()
/ f"{QSAProject._qgis_project_dir_prefix(self.schema)}{self.name}"
)
@staticmethod
def _qgis_project_dir_prefix(schema: str = "") -> str:
prefix = ""
if StorageBackend.type() == StorageBackend.POSTGRESQL:
prefix = f"{schema}_"
return prefix
@staticmethod
def _layer_provider(layer_type: Qgis.LayerType, datasource: str) -> str:
provider = ""
if layer_type == Qgis.LayerType.Vector:
provider = "ogr"
if "table=" in datasource:
provider = "postgres"
elif layer_type == Qgis.LayerType.Raster:
provider = "gdal"
return provider
@staticmethod
def _layer_epsg_code(lyr) -> int:
authid_items = lyr.crs().authid().split(":")
if len(authid_items) < 2:
return -1
return int(authid_items[1])
@staticmethod
def _layer_bbox(lyr) -> list:
return list(
map(
float,
lyr.extent()
.asWktCoordinates()
.replace(",", "")
.split(" "),
)
)
@property
def _qgis_project_uri(self) -> str:
if StorageBackend.type() == StorageBackend.POSTGRESQL:
service = config().qgisserver_projects_psql_service
return f"postgresql:?service={service}&schema={self.schema}&project={self.name}"
else:
return (self._qgis_project_dir / f"{self.name}.qgs").as_posix()
| 25,580 | Python | .py | 622 | 29.051447 | 92 | 0.562888 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,592 | config.py | pblottiere_QSA/qsa-api/qsa_api/config.py | # coding: utf8
import os
import logging
from pathlib import Path
class QSAConfig:
@property
def is_valid(self) -> bool:
if self.qgisserver_url and self.qgisserver_projects_dir:
return True
return False
@property
def loglevel(self):
level = os.environ.get("QSA_LOGLEVEL", "INFO").lower()
logging_level = logging.INFO
if level == "debug":
logging_level = logging.DEBUG
elif level == "error":
logging_level = logging.ERROR
return logging_level
@property
def gdal_pam_proxy_dir(self) -> Path:
return Path(os.environ.get("GDAL_PAM_PROXY_DIR", ""))
@property
def monitoring_port(self) -> int:
return int(os.environ.get("QSA_QGISSERVER_MONITORING_PORT", "0"))
@property
def qgisserver_url(self) -> str:
return os.environ.get("QSA_QGISSERVER_URL", "")
@property
def qgisserver_projects_dir(self) -> str:
return os.environ.get("QSA_QGISSERVER_PROJECTS_DIR", "")
@property
def qgisserver_projects_psql_service(self) -> str:
return os.environ.get("QSA_QGISSERVER_PROJECTS_PSQL_SERVICE", "")
@property
def mapproxy_projects_dir(self) -> str:
return os.environ.get("QSA_MAPPROXY_PROJECTS_DIR", "").replace('"', "")
@property
def mapproxy_cache_s3_bucket(self) -> str:
return os.environ.get("QSA_MAPPROXY_CACHE_S3_BUCKET", "")
@property
def mapproxy_cache_s3_dir(self) -> str:
return os.environ.get("QSA_MAPPROXY_CACHE_S3_DIR", "/mapproxy/cache")
@property
def aws_access_key_id(self) -> str:
return os.environ.get("AWS_ACCESS_KEY_ID", "")
@property
def aws_secret_access_key(self) -> str:
return os.environ.get("AWS_SECRET_ACCESS_KEY", "")
| 1,810 | Python | .py | 49 | 30.163265 | 79 | 0.640893 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,593 | app.py | pblottiere_QSA/qsa-api/qsa_api/app.py | # coding: utf8
import click
from flask import Flask
from qsa_api.config import QSAConfig
from qsa_api.monitor import QSAMonitor
from qsa_api.api.projects import projects
from qsa_api.api.symbology import symbology
from qsa_api.api.instances import instances
from qsa_api.api.processing import processing
app = Flask(__name__)
class QSA:
def __init__(self) -> None:
self.cfg = QSAConfig()
if not self.cfg.is_valid:
return
self.monitor = None
if self.cfg.monitoring_port:
self.monitor = QSAMonitor(self.cfg)
self.monitor.start()
app.config["CONFIG"] = self.cfg
app.config["MONITOR"] = self.monitor
app.register_blueprint(projects, url_prefix="/api/projects")
app.register_blueprint(symbology, url_prefix="/api/symbology")
app.register_blueprint(instances, url_prefix="/api/instances")
app.register_blueprint(processing, url_prefix="/api/processing")
app.logger.setLevel(self.cfg.loglevel)
def run(self):
app.run(host="0.0.0.0", threaded=False)
qsa = QSA()
@click.command()
def run():
qsa.run()
| 1,149 | Python | .py | 32 | 30 | 72 | 0.6875 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,594 | utils.py | pblottiere_QSA/qsa-api/qsa_api/utils.py | # coding: utf8
import os
import sys
import boto3
import logging
import threading
from enum import Enum
from pathlib import Path
from flask import current_app
from botocore.exceptions import ClientError
from .config import QSAConfig
def config():
return current_app.config["CONFIG"]
def logger():
return current_app.logger
def s3_parse_uri(uri: str):
# /vsis3/{bucket}/{subdirs}/{filename}
bucket = uri.split("/")[2]
subdirs = Path(uri.split(f"/vsis3/{bucket}/")[1]).parent.as_posix()
filename = Path(uri.split(f"/vsis3/{bucket}/")[1]).name
return bucket, subdirs, filename
class StorageBackend(Enum):
FILESYSTEM = 0
POSTGRESQL = 1
@staticmethod
def type() -> "StorageBackend":
if config().qgisserver_projects_psql_service:
return StorageBackend.POSTGRESQL
return StorageBackend.FILESYSTEM
def qgisserver_base_url(project: str, psql_schema: str) -> str:
url = f"{config().qgisserver_url}"
if StorageBackend.type() == StorageBackend.FILESYSTEM:
url = f"{url}/{project}?"
elif StorageBackend.type() == StorageBackend.POSTGRESQL:
service = config().qgisserver_projects_psql_service
url = f"{url}?MAP=postgresql:?service={service}%26schema={psql_schema}%26project={project}&"
return url
# see boto3 doc
class ProgressPercentage:
def __init__(self, filename):
self._filename = filename
self._size = float(os.path.getsize(filename))
self._seen_so_far = 0
self._lock = threading.Lock()
self._last = 0
def __call__(self, bytes_amount):
with self._lock:
self._seen_so_far += bytes_amount
percentage = (self._seen_so_far / self._size) * 100
if percentage < self._last + 5:
return
self._last = percentage
if QSAConfig().loglevel == logging.DEBUG:
print(
"\r%s %s / %s (%.2f%%)"
% (
self._filename,
self._seen_so_far,
self._size,
percentage,
),
file=sys.stderr,
)
sys.stdout.flush()
def s3_bucket_upload(bucket: str, source: str, dest: str) -> (bool, str):
size = float(os.path.getsize(source) >> 20)
logger().debug(
f"[utils.s3_bucket_upload] Upload {source} ({size}MB) to S3 bucket {bucket} in {dest}"
)
try:
s3 = boto3.resource("s3")
s3.Bucket(bucket).upload_file(
source,
dest,
Callback=ProgressPercentage(source),
)
except ClientError as e:
return False, "Upload to S3 bucket failed"
return True, ""
| 2,784 | Python | .py | 79 | 26.772152 | 100 | 0.596118 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,595 | __init__.py | pblottiere_QSA/qsa-api/qsa_api/__init__.py | # coding: utf8
import os
from qgis.core import QgsApplication
# avoid "Application path not initialized" message
os.environ["GDAL_PAM_PROXY_DIR"] = "/tmp"
os.environ["QT_QPA_PLATFORM"] = "offscreen"
QgsApplication.setPrefixPath("/usr", True)
qgs = QgsApplication([], False)
qgs.initQgis()
| 293 | Python | .py | 9 | 31.111111 | 50 | 0.771429 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,596 | monitor.py | pblottiere_QSA/qsa-api/qsa_api/monitor.py | # coding: utf8
import sys
import uuid
import time
import pickle
import socket
import struct
from datetime import datetime
from threading import Thread, Lock
from qsa_api.config import QSAConfig
class QSAMonitorThread(Thread):
def __init__(self, con, ip: str, port: int) -> None:
Thread.__init__(self)
self.con = con
self.ip = ip
self.port = port
self.now = datetime.now()
self.response = None
def run(self):
try:
while True:
# empty recv if error or size of the payload
data = self.con.recv(4)
if not data:
self.con.close()
return
data_size = struct.unpack(">I", data)[0]
self.response = self._recv_payload(data_size)
except BrokenPipeError:
return
@property
def metadata(self) -> dict:
self.response = None
try:
self.con.send(b"metadata")
return self._wait_recv()
except Exception:
return {}
@property
def logs(self) -> dict:
self.response = None
try:
self.con.send(b"logs")
return self._wait_recv()
except Exception as e:
print(e, file=sys.stderr)
return {}
@property
def stats(self) -> dict:
self.response = None
try:
self.con.send(b"stats")
return self._wait_recv()
except Exception as e:
print(e, file=sys.stderr)
return {}
def _wait_recv(self):
it = 0
while True:
if self.response:
return self.response
time.sleep(0.1)
it += 1
if it >= 20:
return {"error": "timeout"}
def _recv_payload(self, data_size):
received_payload = b""
reamining_payload_size = data_size
while reamining_payload_size != 0:
received_payload += self.con.recv(reamining_payload_size)
reamining_payload_size = data_size - len(received_payload)
return pickle.loads(received_payload)
class QSAMonitor:
def __init__(self, cfg: QSAConfig) -> None:
self.monitor: Thread
self.port: int = cfg.monitoring_port
self._lock = Lock()
self._conns: dict = {}
@property
def conns(self):
conns = {}
self._lock.acquire()
for uid in self._conns:
if self._conns[uid].is_alive():
conns[uid] = self._conns[uid]
else:
self._conns[uid].join()
self._conns = conns
self._lock.release()
return self._conns
def start(self) -> None:
self.monitor = Thread(target=self._start, args=())
self.monitor.start()
def _start(self) -> None:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
s.bind(("0.0.0.0", self.port))
while True:
s.listen(5)
try:
(con, (ip, port)) = s.accept()
except:
break
thread = QSAMonitorThread(con, ip, port)
thread.start()
self._lock.acquire()
uid = str(uuid.uuid4())[:8]
self._conns[uid] = thread
self._lock.release()
for uid in self._conns:
self._conns[uid].join()
| 3,488 | Python | .py | 111 | 21.468468 | 70 | 0.534467 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,597 | wms.py | pblottiere_QSA/qsa-api/qsa_api/wms.py | # coding: utf8
from .project import QSAProject
from .utils import qgisserver_base_url
class WMS:
@staticmethod
def getmap_url(project, psql_schema, layer):
p = QSAProject(project, psql_schema)
props = p.layer(layer)
if "bbox" not in props:
return "Invalid layer"
bbox = props["bbox"].replace(" ", ",").replace(",,", ",").split(",")
wms_bbox = f"{bbox[1]},{bbox[0]},{bbox[3]},{bbox[2]}"
return f"REQUEST=GetMap&WIDTH=400&HEIGHT=400&CRS={props['crs']}&VERSION=1.3.0&BBOX={wms_bbox}&LAYERS={layer}"
@staticmethod
def getmap(project, psql_schema, layer):
return f"{qgisserver_base_url(project, psql_schema)}{WMS.getmap_url(project, psql_schema, layer)}"
| 741 | Python | .py | 16 | 39.625 | 117 | 0.639276 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,598 | mapproxy.py | pblottiere_QSA/qsa-api/qsa_api/mapproxy/mapproxy.py | # coding: utf8
import sys
import yaml
import boto3
import shutil
from pathlib import Path
from qgis.PyQt.QtCore import Qt, QDateTime
from ..utils import config, logger, qgisserver_base_url
class QSAMapProxy:
def __init__(self, name: str, schema: str = "") -> None:
self.name = name
self.schema = "public"
if schema:
self.schema = schema
def create(self) -> None:
parent = Path(__file__).resolve().parent
template = parent / "mapproxy.yaml"
shutil.copy(template, self._mapproxy_project)
def remove(self) -> None:
self._mapproxy_project.unlink()
def write(self) -> None:
with open(self._mapproxy_project, "w") as file:
yaml.safe_dump(self.cfg, file, sort_keys=False)
def read(self) -> (bool, str):
# if a QGIS project is created manually without QSA, the MapProxy
# configuration file may not be created at this point.
if not self._mapproxy_project.exists():
self.create()
try:
with open(self._mapproxy_project, "r") as file:
self.cfg = yaml.safe_load(file)
except yaml.scanner.ScannerError as e:
return (
False,
f"Failed to load MapProxy configuration file {self._mapproxy_project}",
)
if self.cfg is None:
return (
False,
f"Failed to load MapProxy configuration file {self._mapproxy_project}",
)
return True, ""
def metadata(self) -> dict:
md = {}
md["storage"] = ""
md["valid"] = False
if self._mapproxy_project.exists():
md["valid"] = True
md["storage"] = "filesystem"
if config().mapproxy_cache_s3_bucket:
md["storage"] = "s3"
return md
def clear_cache(self, layer_name: str) -> None:
if config().mapproxy_cache_s3_bucket:
bucket_name = config().mapproxy_cache_s3_bucket
cache_dir = f"{config().mapproxy_cache_s3_dir}/{layer_name}"
if cache_dir[0] == "/":
cache_dir = cache_dir[1:]
self.debug(f"Clear S3 cache 's3://{bucket_name}/{cache_dir}'")
s3 = boto3.resource(
"s3",
aws_access_key_id=config().aws_access_key_id,
aws_secret_access_key=config().aws_secret_access_key,
)
bucket = s3.Bucket(bucket_name)
bucket.objects.filter(Prefix=cache_dir).delete()
else:
cache_dir = self._mapproxy_project.parent / "cache_data"
self.debug(f"Clear tiles cache '{cache_dir}'")
for d in cache_dir.glob(f"{layer_name}_cache_*"):
shutil.rmtree(d)
cache_dir = self._mapproxy_project.parent / "cache_data" / "legends"
self.debug(f"Clear legends cache '{cache_dir}'")
shutil.rmtree(cache_dir, ignore_errors=True)
def add_layer(
self,
name: str,
bbox: list,
srs: int,
is_raster: bool,
datetime: QDateTime | None,
) -> (bool, str):
if self.cfg is None:
return False, "Invalid MapProxy configuration"
if "layers" not in self.cfg:
self.cfg["layers"] = []
self.cfg["caches"] = {}
self.cfg["sources"] = {}
lyr = {"name": name, "title": name, "sources": [f"{name}_cache"]}
if datetime and is_raster:
lyr["dimensions"] = {}
lyr["dimensions"]["time"] = {
"values": [datetime.toString(Qt.ISODate)]
}
self.cfg["layers"].append(lyr)
c = {"grids": ["webmercator"], "sources": [f"{name}_wms"]}
if is_raster:
c["use_direct_from_level"] = 14
c["meta_size"] = [1, 1]
c["meta_buffer"] = 0
if config().mapproxy_cache_s3_bucket:
s3_cache_dir = f"{config().mapproxy_cache_s3_dir}/{name}"
c["cache"] = {}
c["cache"]["type"] = "s3"
c["cache"]["directory"] = s3_cache_dir
c["cache"]["bucket_name"] = config().mapproxy_cache_s3_bucket
self.cfg["caches"][f"{name}_cache"] = c
s = {
"type": "wms",
"wms_opts": {
"legendgraphic": True,
},
"req": {
"url": qgisserver_base_url(self.name, self.schema),
"layers": name,
"transparent": True,
},
"coverage": {"bbox": bbox, "srs": f"EPSG:{srs}"},
}
if datetime and is_raster:
s["forward_req_params"] = ["TIME"]
self.cfg["sources"][f"{name}_wms"] = s
return True, ""
def remove_layer(self, name: str) -> None:
# early return
if "layers" not in self.cfg:
return
# clear cache
self.clear_cache(name)
# clean layers
layers = []
for layer in self.cfg["layers"]:
if layer["name"] != name:
layers.append(layer)
self.cfg["layers"] = layers
# clean caches
cache_name = f"{name}_cache"
if cache_name in self.cfg["caches"]:
self.cfg["caches"].pop(cache_name)
# clean sources
source_name = f"{name}_wms"
if source_name in self.cfg["sources"]:
self.cfg["sources"].pop(source_name)
def debug(self, msg: str) -> None:
caller = f"{self.__class__.__name__}.{sys._getframe().f_back.f_code.co_name}"
msg = f"[{caller}][{self.name}] {msg}"
logger().debug(msg)
@staticmethod
def _mapproxy_projects_dir() -> Path:
return Path(config().mapproxy_projects_dir)
@property
def _mapproxy_project(self) -> Path:
return QSAMapProxy._mapproxy_projects_dir() / f"{self.name}.yaml"
| 5,918 | Python | .py | 153 | 28.078431 | 87 | 0.533694 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
2,287,599 | renderer.py | pblottiere_QSA/qsa-api/qsa_api/raster/renderer.py | # coding: utf8
from enum import Enum
from pathlib import Path
from qgis.core import (
QgsStyle,
QgsRasterLayer,
QgsRasterShader,
QgsColorRampShader,
QgsRasterBandStats,
QgsGradientColorRamp,
QgsRasterMinMaxOrigin,
QgsContrastEnhancement,
QgsSingleBandGrayRenderer,
QgsMultiBandColorRenderer,
QgsSingleBandPseudoColorRenderer,
)
ContrastEnhancementAlgorithm = (
QgsContrastEnhancement.ContrastEnhancementAlgorithm
)
class RasterSymbologyRenderer:
class Type(Enum):
SINGLE_BAND_GRAY = QgsSingleBandGrayRenderer(None, 1).type()
SINGLE_BAND_PSEUDOCOLOR = QgsSingleBandPseudoColorRenderer(
None, 1
).type()
MULTI_BAND_COLOR = QgsMultiBandColorRenderer(None, 1, 1, 1).type()
def __init__(self, name: str) -> None:
self.renderer = None
self.contrast_algorithm = None
self.contrast_limits = QgsRasterMinMaxOrigin.Limits.MinMax
self.gray_min = None
self.gray_max = None
self.red_min = None
self.red_max = None
self.green_min = None
self.green_max = None
self.blue_min = None
self.blue_max = None
if name == RasterSymbologyRenderer.Type.SINGLE_BAND_GRAY.value:
self.renderer = QgsSingleBandGrayRenderer(None, 1)
elif name == RasterSymbologyRenderer.Type.MULTI_BAND_COLOR.value:
self.renderer = QgsMultiBandColorRenderer(None, 1, 1, 1)
elif (
name == RasterSymbologyRenderer.Type.SINGLE_BAND_PSEUDOCOLOR.value
):
self.renderer = QgsSingleBandPseudoColorRenderer(None, 1)
@property
def type(self):
if (
self.renderer.type()
== RasterSymbologyRenderer.Type.SINGLE_BAND_GRAY.value
):
return RasterSymbologyRenderer.Type.SINGLE_BAND_GRAY
elif (
self.renderer.type()
== RasterSymbologyRenderer.Type.MULTI_BAND_COLOR.value
):
return RasterSymbologyRenderer.Type.MULTI_BAND_COLOR
elif (
self.renderer.type()
== RasterSymbologyRenderer.Type.SINGLE_BAND_PSEUDOCOLOR.value
):
return RasterSymbologyRenderer.Type.SINGLE_BAND_PSEUDOCOLOR
return None
def load(self, properties: dict) -> (bool, str):
if not self.renderer:
return False, "Invalid renderer"
if "contrast_enhancement" in properties:
self._load_contrast_enhancement(properties["contrast_enhancement"])
if self.type == RasterSymbologyRenderer.Type.MULTI_BAND_COLOR:
self._load_multibandcolor_properties(properties)
elif self.type == RasterSymbologyRenderer.Type.SINGLE_BAND_GRAY:
self._load_singlebandgray_properties(properties)
elif self.type == RasterSymbologyRenderer.Type.SINGLE_BAND_PSEUDOCOLOR:
self._load_singlebandpseudocolor_properties(properties)
return True, ""
def refresh_min_max(self, layer: QgsRasterLayer) -> None:
# see QgsRasterMinMaxWidget::doComputations
# early break
if (
layer.renderer().minMaxOrigin().limits()
== QgsRasterMinMaxOrigin.Limits.None_
):
return
# refresh according to renderer
if self.type == RasterSymbologyRenderer.Type.SINGLE_BAND_GRAY:
self._refresh_min_max_singlebandgray(layer)
elif self.type == RasterSymbologyRenderer.Type.MULTI_BAND_COLOR:
self._refresh_min_max_multibandcolor(layer)
elif self.type == RasterSymbologyRenderer.Type.SINGLE_BAND_PSEUDOCOLOR:
self._refresh_min_max_singlebandpseudocolor(layer)
@staticmethod
def style_to_json(path: Path) -> (dict, str):
tif = Path(__file__).resolve().parent / "empty.tif"
rl = QgsRasterLayer(tif.as_posix(), "", "gdal")
rl.loadNamedStyle(path.as_posix())
renderer = rl.renderer()
renderer_type = RasterSymbologyRenderer(renderer.type()).type
m = {}
m["name"] = path.stem
m["type"] = "raster"
m["symbology"] = {}
m["symbology"]["type"] = renderer.type()
props = {}
if renderer_type == RasterSymbologyRenderer.Type.SINGLE_BAND_GRAY:
props = RasterSymbologyRenderer._singlebandgray_properties(
renderer
)
elif renderer_type == RasterSymbologyRenderer.Type.MULTI_BAND_COLOR:
props = RasterSymbologyRenderer._multibandcolor_properties(
renderer
)
elif (
renderer_type
== RasterSymbologyRenderer.Type.SINGLE_BAND_PSEUDOCOLOR
):
props = RasterSymbologyRenderer._singlebandpseudocolor_properties(
renderer
)
m["symbology"]["properties"] = props
m["rendering"] = {}
m["rendering"]["brightness"] = rl.brightnessFilter().brightness()
m["rendering"]["contrast"] = rl.brightnessFilter().contrast()
m["rendering"]["gamma"] = rl.brightnessFilter().gamma()
m["rendering"]["saturation"] = rl.hueSaturationFilter().saturation()
return m, ""
@staticmethod
def _multibandcolor_properties(renderer) -> dict:
props = {}
# limits
limits = renderer.minMaxOrigin().limits()
props["contrast_enhancement"] = {}
props["contrast_enhancement"]["limits_min_max"] = "UserDefined"
if limits == QgsRasterMinMaxOrigin.Limits.MinMax:
props["contrast_enhancement"]["limits_min_max"] = "MinMax"
# bands
props["red"] = {}
props["red"]["band"] = renderer.redBand()
props["blue"] = {}
props["blue"]["band"] = renderer.blueBand()
props["green"] = {}
props["green"]["band"] = renderer.greenBand()
# red band
if renderer.redContrastEnhancement():
red_ce = QgsContrastEnhancement(renderer.redContrastEnhancement())
props["red"]["min"] = red_ce.minimumValue()
props["red"]["max"] = red_ce.maximumValue()
# blue band
blue_ce = QgsContrastEnhancement(
renderer.blueContrastEnhancement()
)
props["blue"]["min"] = blue_ce.minimumValue()
props["blue"]["max"] = blue_ce.maximumValue()
# green band
green_ce = QgsContrastEnhancement(
renderer.greenContrastEnhancement()
)
props["green"]["min"] = green_ce.minimumValue()
props["green"]["max"] = green_ce.maximumValue()
# ce
alg = red_ce.contrastEnhancementAlgorithm()
props["contrast_enhancement"]["algorithm"] = "NoEnhancement"
if (
alg
== QgsContrastEnhancement.ContrastEnhancementAlgorithm.StretchToMinimumMaximum
):
props["contrast_enhancement"][
"algorithm"
] = "StretchToMinimumMaximum"
else:
# default behavior
props["contrast_enhancement"][
"algorithm"
] = "StretchToMinimumMaximum"
return props
@staticmethod
def _singlebandgray_properties(renderer) -> dict:
props = {}
props["gray"] = {}
props["gray"]["band"] = renderer.grayBand()
ce = renderer.contrastEnhancement()
props["gray"]["min"] = ce.minimumValue()
props["gray"]["max"] = ce.maximumValue()
gradient = renderer.gradient()
if gradient == QgsSingleBandGrayRenderer.Gradient.BlackToWhite:
props["color_gradient"] = "BlackToWhite"
else:
props["color_gradient"] = "WhiteToBlack"
props["contrast_enhancement"] = {}
alg = ce.contrastEnhancementAlgorithm()
props["contrast_enhancement"]["algorithm"] = "NoEnhancement"
if (
alg
== QgsContrastEnhancement.ContrastEnhancementAlgorithm.StretchToMinimumMaximum
):
props["contrast_enhancement"][
"algorithm"
] = "StretchToMinimumMaximum"
limits = renderer.minMaxOrigin().limits()
props["contrast_enhancement"]["limits_min_max"] = "UserDefined"
if limits == QgsRasterMinMaxOrigin.Limits.MinMax:
props["contrast_enhancement"]["limits_min_max"] = "MinMax"
return props
@staticmethod
def _singlebandpseudocolor_properties(renderer) -> dict:
props = {}
if renderer.shader() is None:
return {}, "Invalid shader in singlebandpseudocolor renderer"
if renderer.shader().rasterShaderFunction().sourceColorRamp() is None:
return {}, "Invalid color ramp in singlebandpseudocolor renderer"
props["band"] = {}
props["band"]["band"] = renderer.band()
props["band"]["min"] = renderer.classificationMin()
props["band"]["max"] = renderer.classificationMax()
props["ramp"] = {}
shader_fct = renderer.shader().rasterShaderFunction()
color_1 = (
shader_fct.sourceColorRamp().properties()["color1"].split("rgb")[0]
)
color_2 = (
shader_fct.sourceColorRamp().properties()["color2"].split("rgb")[0]
)
stops = shader_fct.sourceColorRamp().properties()["stops"]
props["ramp"]["color1"] = color_1
props["ramp"]["color2"] = color_2
props["ramp"]["stops"] = stops
ramp_type = shader_fct.colorRampType()
if ramp_type == QgsColorRampShader.Discrete:
props["ramp"]["interpolation"] = "Discrete"
elif ramp_type == QgsColorRampShader.Exact:
props["ramp"]["interpolation"] = "Exact"
elif ramp_type == QgsColorRampShader.Interpolated:
props["ramp"]["interpolation"] = "Interpolated"
props["contrast_enhancement"] = {}
limits = renderer.minMaxOrigin().limits()
props["contrast_enhancement"]["limits_min_max"] = "UserDefined"
if limits == QgsRasterMinMaxOrigin.Limits.MinMax:
props["contrast_enhancement"]["limits_min_max"] = "MinMax"
return props
def _refresh_min_max_multibandcolor(self, layer: QgsRasterLayer) -> None:
renderer = layer.renderer()
red_ce = QgsContrastEnhancement(renderer.redContrastEnhancement())
green_ce = QgsContrastEnhancement(renderer.greenContrastEnhancement())
blue_ce = QgsContrastEnhancement(renderer.blueContrastEnhancement())
# early break
alg = red_ce.contrastEnhancementAlgorithm()
if (
alg == ContrastEnhancementAlgorithm.NoEnhancement
or alg == ContrastEnhancementAlgorithm.UserDefinedEnhancement
):
return
# compute min/max with "Accuracy: estimate"
min_max_origin = renderer.minMaxOrigin().limits()
if min_max_origin == QgsRasterMinMaxOrigin.Limits.MinMax:
red_band = renderer.redBand()
red_stats = layer.dataProvider().bandStatistics(
red_band,
QgsRasterBandStats.Min | QgsRasterBandStats.Max,
layer.extent(),
250000,
)
red_ce.setMinimumValue(red_stats.minimumValue)
red_ce.setMaximumValue(red_stats.maximumValue)
green_band = renderer.greenBand()
green_stats = layer.dataProvider().bandStatistics(
green_band,
QgsRasterBandStats.Min | QgsRasterBandStats.Max,
layer.extent(),
250000,
)
green_ce.setMinimumValue(green_stats.minimumValue)
green_ce.setMaximumValue(green_stats.maximumValue)
blue_band = renderer.blueBand()
blue_stats = layer.dataProvider().bandStatistics(
blue_band,
QgsRasterBandStats.Min | QgsRasterBandStats.Max,
layer.extent(),
250000,
)
blue_ce.setMinimumValue(blue_stats.minimumValue)
blue_ce.setMaximumValue(blue_stats.maximumValue)
layer.renderer().setRedContrastEnhancement(red_ce)
layer.renderer().setGreenContrastEnhancement(green_ce)
layer.renderer().setBlueContrastEnhancement(blue_ce)
def _refresh_min_max_singlebandgray(self, layer: QgsRasterLayer) -> None:
ce = QgsContrastEnhancement(layer.renderer().contrastEnhancement())
# early break
alg = ce.contrastEnhancementAlgorithm()
if (
alg == ContrastEnhancementAlgorithm.NoEnhancement
or alg == ContrastEnhancementAlgorithm.UserDefinedEnhancement
):
return
# compute min/max
min_max_origin = layer.renderer().minMaxOrigin().limits()
if min_max_origin == QgsRasterMinMaxOrigin.Limits.MinMax:
# Accuracy : estimate
stats = layer.dataProvider().bandStatistics(
1,
QgsRasterBandStats.Min | QgsRasterBandStats.Max,
layer.extent(),
250000,
)
ce.setMinimumValue(stats.minimumValue)
ce.setMaximumValue(stats.maximumValue)
layer.renderer().setContrastEnhancement(ce)
def _refresh_min_max_singlebandpseudocolor(
self, layer: QgsRasterLayer
) -> None:
# compute min/max
min_max_origin = layer.renderer().minMaxOrigin().limits()
if min_max_origin == QgsRasterMinMaxOrigin.Limits.MinMax:
# Accuracy : estimate
stats = layer.dataProvider().bandStatistics(
1,
QgsRasterBandStats.Min | QgsRasterBandStats.Max,
layer.extent(),
250000,
)
layer.renderer().setClassificationMin(stats.minimumValue)
layer.renderer().setClassificationMax(stats.maximumValue)
layer.renderer().shader().rasterShaderFunction().classifyColorRamp()
def _load_multibandcolor_properties(self, properties: dict) -> None:
if "red" in properties:
red = properties["red"]
self.renderer.setRedBand(int(red["band"]))
if self.contrast_limits == QgsRasterMinMaxOrigin.Limits.None_:
if "min" in red:
self.red_min = float(red["min"])
if "max" in red:
self.red_max = float(red["max"])
if "blue" in properties:
blue = properties["blue"]
self.renderer.setBlueBand(int(blue["band"]))
if self.contrast_limits == QgsRasterMinMaxOrigin.Limits.None_:
if "min" in blue:
self.blue_min = float(blue["min"])
if "max" in blue:
self.blue_max = float(blue["max"])
if "green" in properties:
green = properties["green"]
self.renderer.setGreenBand(int(green["band"]))
if self.contrast_limits == QgsRasterMinMaxOrigin.Limits.None_:
if "min" in green:
self.green_min = float(green["min"])
if "max" in green:
self.green_max = float(green["max"])
def _load_singlebandgray_properties(self, properties: dict) -> None:
if "gray" in properties:
gray = properties["gray"]
self.renderer.setGrayBand(int(gray["band"]))
if self.contrast_limits == QgsRasterMinMaxOrigin.Limits.None_:
if "min" in gray:
self.gray_min = float(gray["min"])
if "max" in gray:
self.gray_max = float(gray["max"])
if "color_gradient" in properties:
gradient = properties["color_gradient"]
if gradient == "blacktowhite":
self.renderer.setGradient(
QgsSingleBandGrayRenderer.Gradient.BlackToWhite
)
elif gradient == "whitetoblack":
self.renderer.setGradient(
QgsSingleBandGrayRenderer.Gradient.WhiteToBlack
)
def _load_singlebandpseudocolor_properties(self, properties: dict) -> None:
# always stretch to min/max in case of the singlepseudocolor renderer
self.contrast_algorithm = (
ContrastEnhancementAlgorithm.StretchToMinimumMaximum
)
band_min = None
band_max = None
if "band" in properties:
band = properties["band"]
self.renderer.setBand(int(band["band"]))
if self.contrast_limits == QgsRasterMinMaxOrigin.Limits.None_:
if "min" in band:
band_min = float(band["min"])
if "max" in band:
band_max = float(band["max"])
if "ramp" in properties:
ramp = properties["ramp"]
shader_type = QgsColorRampShader.Type.Interpolated
if "interpolation" in ramp:
interpolation = ramp["interpolation"]
if interpolation == "Discrete":
shader_type = QgsColorRampShader.Type.Discrete
elif interpolation == "Exact":
shader_type = QgsColorRampShader.Type.Exact
color_ramp = QgsStyle().defaultStyle().colorRamp("Spectral")
if "name" in ramp and ramp["name"]:
color_ramp = QgsStyle().defaultStyle().colorRamp(ramp["name"])
elif "color1" in ramp and "color2" in ramp:
color_ramp = QgsGradientColorRamp.create(ramp)
ramp_shader = QgsColorRampShader()
ramp_shader.setSourceColorRamp(color_ramp)
ramp_shader.setColorRampType(shader_type)
shader = QgsRasterShader()
shader.setRasterShaderFunction(ramp_shader)
self.renderer.setShader(shader)
if band_min is not None:
self.renderer.setClassificationMin(band_min)
if band_max is not None:
self.renderer.setClassificationMax(band_max)
self.renderer.shader().rasterShaderFunction().classifyColorRamp()
def _load_contrast_enhancement(self, properties: dict) -> None:
if "algorithm" in properties:
alg = properties["algorithm"]
if alg == "StretchToMinimumMaximum":
self.contrast_algorithm = (
ContrastEnhancementAlgorithm.StretchToMinimumMaximum
)
elif alg == "NoEnhancement":
self.contrast_algorithm = (
ContrastEnhancementAlgorithm.NoEnhancement
)
if "limits_min_max" in properties:
limits = properties["limits_min_max"]
if limits == "UserDefined":
self.contrast_limits = QgsRasterMinMaxOrigin.Limits.None_
elif limits == "MinMax":
self.contrast_limits = QgsRasterMinMaxOrigin.Limits.MinMax
| 19,024 | Python | .py | 423 | 33.3026 | 94 | 0.605393 | pblottiere/QSA | 8 | 4 | 12 | GPL-3.0 | 9/5/2024, 10:48:18 PM (Europe/Amsterdam) |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.