def test_connect_four():
game = ConnectFour()
game.play_move(1)
game.play_move(2)
game.play_move(3)
board = game.board
game2 = ConnectFour.from_numpy(board, is_canonical=False)
board2 = game2.board
np.testing.assert_array_equal(board, board2)
assert type(game2) == type(game)
def test_mcts_search():
from pyoaz.thread_pool import ThreadPool
from pyoaz.search import Search, PlayerSearchProperties
from pyoaz.selection import UCTSelector
from pyoaz.evaluator.simulation_evaluator import SimulationEvaluator
from pyoaz.games.connect_four import ConnectFour
thread_pool = ThreadPool(n_workers=1)
evaluator = SimulationEvaluator(thread_pool=thread_pool)
selector = UCTSelector()
player_search_properties = [
PlayerSearchProperties(evaluator, selector),
PlayerSearchProperties(evaluator, selector)
]
game = ConnectFour()
_ = Search(
game=game,
player_search_properties=player_search_properties,
thread_pool=thread_pool,
n_concurrent_workers=1,
n_iterations=25000,
noise_epsilon=0.0,
noise_alpha=0.0,
)
from pyoaz.games.connect_four import apply_symmetry, ConnectFour import numpy as np POLICy_0 = np.array([1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]) POLICy_1 = np.array([0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0]) SYM_POLICy_0 = np.array([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]) SYM_POLICy_1 = np.array([0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]) game_0 = ConnectFour() game_0.play_move(0) BOARD_0 = game_0.board.copy() game_1 = ConnectFour() game_1.play_move(1) game_1.play_move(1) game_1.play_move(3) game_1.play_move(6) BOARD_1 = game_1.board.copy() sym_game_0 = ConnectFour() sym_game_0.play_move(6) SYM_BOARD_0 = sym_game_0.board.copy() sym_game_1 = ConnectFour() sym_game_1.play_move(5) sym_game_1.play_move(5) sym_game_1.play_move(3) sym_game_1.play_move(0) SYM_BOARD_1 = sym_game_1.board.copy()
def test_az_search_with_cache():
with tf.Session() as session:
# Neural network definition
input = tf.placeholder(
dtype=tf.float32, shape=[None, 6, 7, 2], name="input"
)
conv0_filters = tf.Variable(
[
[[[1.0], [1.0]]],
[[[-2.0], [-2.0]]],
[[[3.0], [3.0]]],
[[[4.0], [4.0]]],
],
name="conv0_filters",
dtype=tf.float32,
)
conv0 = tf.nn.conv2d(input, conv0_filters, 1, "SAME")
max_pool0 = tf.nn.max_pool2d(
conv0, [1, 2, 2, 1], [1, 2, 2, 1], padding="SAME"
)
flat = tf.reshape(max_pool0, [-1, 12], name="flat")
dense_value = tf.Variable([[1.0] for _ in range(12)], dtype=tf.float32)
dense_policy = tf.Variable(
[[1.0 for _ in range(7)] for _ in range(12)], dtype=tf.float32
)
dense_value = tf.Variable([[1.0] for _ in range(12)], dtype=tf.float32)
_ = tf.matmul(flat, dense_value, name="value")
dense_policy = tf.Variable(
[[1.0 for _ in range(7)] for _ in range(12)], dtype=tf.float32
)
_ = tf.matmul(flat, dense_value, name="value")
policy_logits = tf.matmul(flat, dense_policy, name="policy_logits")
_ = tf.nn.softmax(policy_logits, name="policy")
session.run(tf.global_variables_initializer())
# AZ search definition
model = Model(
session=session,
input_node_name="input",
value_node_name="value",
policy_node_name="policy",
)
thread_pool = ThreadPool(n_workers=1)
cache = SimpleCache(ConnectFour(), 100)
evaluator = NNEvaluator(
model=model,
cache=cache,
thread_pool=thread_pool,
dimensions=(6, 7, 2),
batch_size=1,
)
selector = AZSelector()
player_search_properties = [
PlayerSearchProperties(evaluator, selector),
PlayerSearchProperties(evaluator, selector)
]
game = ConnectFour()
_ = Search(
game=game,
player_search_properties=player_search_properties,
thread_pool=thread_pool,
n_concurrent_workers=1,
n_iterations=100,
noise_epsilon=0.25,
noise_alpha=1,
)
def worker(id, pool, evaluator):
print(f"Thread {id} started")
game = ConnectFour()
self_play(game, pool, evaluator)
import random
from tqdm.auto import tqdm
from pyoaz.games.connect_four import ConnectFour
boards = []
print("Playing random games")
for i in tqdm(range(1000)):
game = ConnectFour()
while not game.finished:
moves = game.available_moves
move = random.choice(moves)
game.play_move(move)
boards.append(game.board)
print(f"Collected {len(boards)} positions")
new_boards = []
print("Playing from collected boards")
games = [ConnectFour.from_numpy(board) for board in boards]
games = [game for game in games if not game.finished]
for game in tqdm(games, total=len(games)):
while not game.finished:
moves = game.available_moves