def exp_effect_of_action_restrict_for_time(depth=5, func_id=2):
gamma = 100000 # スレッショルドカットを実施しない
create_ev_table(ev_table, select_func(func_id)) # 評価関数は固定
state = State()
restrict_time = 0.0
no_restrict_time = 0.0
while True:
# ゲーム終了時
if state.is_done():
break
# 行動の取得
if state.is_first_player():
action = random_action(state) # ランダム行動
else:
# 行動数の削減あり
start = time.time()
for _ in range(50):
action = alpha_beta_action(state, gamma, depth, True)
restrict_time += time.time() - start
# 行動数の削減なし
start = time.time()
for _ in range(50):
action = alpha_beta_action(state, gamma, depth, False)
no_restrict_time += time.time() - start
action = random_action(state) # お互いにランダム行動をさせる
state = state.next(action)
print("restrict:", restrict_time, "no_restrict:", no_restrict_time)
def exp_gamma_time(depth=5, func_id=2, seed=random.random()):
print("seed", seed)
random.seed(seed)
state = State()
create_ev_table(ev_table, select_func(func_id))
keep_gamma_time = [0] * 30
# ゲーム終了までのループ
while True:
# ゲーム終了時
if state.is_done():
break
# 行動の取得
if state.is_first_player():
action = random_action(state)
else:
gamma = 0.0
for index, _ in enumerate(keep_gamma_time):
start = time.time()
for _ in range(100):
# action = alpha_beta_action(state, gamma)
action = alpha_beta_action(state, gamma, depth, False)
keep_gamma_time[index] += time.time() - start
gamma += 0.1
# データをばらつかせるためにランダム行動をとる
action = random_action(state)
print(keep_gamma_time)
# 次の状態の取得
state = state.next(action)
def exp_value_changing(depth=5, func_id=3, gamma=1.0, seed=random.random()):
record_values = [] # 評価値を記録
record_boards = [] # 評価値に連動して盤面を記録
for i in range(100):
random.seed(seed * (i + 1))
state = State()
ii_state = AccessableState()
values = []
boards = []
while True:
if state.is_done():
break
if state.is_first_player():
action = move_ordering_alpha_beta_action(state, 1, depth, i)
# 盤面の評価値を算出し記録
ii_state.create_ii_state_from_state(state)
values.append(evaluate_board_state(ii_state))
boards.append([state.pieces, state.enemy_pieces])
else:
action = random_action(state)
state = state.next(action)
record_values.apped(values)
record_boards.apped(boards)
# TODO: csvに出力する
print(record_values)
print(record_boards)
def playout(state):
if state.is_lose():
return -1
if state.is_draw():
return 0
return -playout(state.next(random_action(state))) # 再帰的に探索
def rollout(self, state):
cur_player = state.player
while state.winner == 0:
while True:
a = random_action()
if state.legal_move(a):
break
state.make_move(a)
return cur_player == state.winner
def exp_move_ordering_time(depth=5,
func_id=3,
gamma=1.0,
seed=random.random()):
print("seed", seed)
timer = [0.0] * (depth + 1)
random.seed(seed)
state = State()
while True:
if state.is_done():
break
for i in range(depth + 1):
start = time.time()
for _ in range(1):
move_ordering_alpha_beta_action(state, 1, depth, i)
timer[i] += time.time() - start
action = random_action(state)
state = state.next(action) # ランダムにゲームを進める
# else:
# pass
if __name__ == "__main__":
os.environ["OMP_NUM_THREADS"] = "1"
with open("config.yaml") as f:
args = yaml.safe_load(f)
# print(args)
# ここに実験用のコードを書く
state = State()
while True:
print(state.legal_actions())
state = state.next(random_action(state))
# path = "models/10000.pth"
# EvalHandyRL(100, path)
# policies = obs_to_policy_to_use_game(agent, obs, state)
# print(policies)
# convert_state_to_obs(state)
# test_predict()
# test_cigeister()
# 方策を持ってくる
# 接続部分
def main():
print(
"Please input a port number that you want to connect. (10000 or 10001)"
)
# port = 10000 # 先手
# port = 10001 # 後手
port = int(input())
# クライアントの作成
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
# サーバを指定
s.connect(("localhost", port))
# s.connect(("15.152.47.162", port)) # 1
# s.connect(("13.208.76.157", port)) # 2
# s.connect(("13.208.214.66", port)) # 3
# s.connect(("13.208.184.19", port)) # 4
# ネットワークのバッファサイズ。サーバからの文字列を取得(recv)する
data = s.recv(2048)
print(repr(data))
# 赤駒のセット
red_piece = ""
piece_char_list = ["A", "B", "C", "D", "E", "F", "G", "H"]
confirmed_info = KeepConfirmedInfo()
for i, color in enumerate(confirmed_info.my_pieces_color):
if color == 2:
red_piece += piece_char_list[i]
send_str = "SET:" + red_piece + "\r\n"
s.sendall(send_str.encode(encoding="utf-8"))
data = s.recv(2048)
print(repr(data))
if str(repr(data)) == str(b"OK \r\n"):
print("駒のセット成功")
else:
print("駒のセット失敗")
now_tcp_str = "" # 直前に受け取ったtcp
# 受け取ったdataをstrに変換
data = s.recv(2048)
now_tcp_str = data.decode(encoding="utf-8")
print(now_tcp_str)
# ちょっとルールベースな処理
rl_action = HandyAction("models/108000.pth")
turn_count = 0
while 1:
# 確定している情報からStateを生成
state = confirmed_info.create_state()
print(state)
# stateからactionを決定
action = alpha_beta_action(state, 5, False)
print(action)
# actionが変（0とか）だったらとりあえず修正
if action not in state.legal_actions():
action = random_action(state)
# 自分の行動をsendall可能なバイナリデータに変換
sendall_str_b = confirmed_info.action_to_sendall_str(action)
print("送信データ（行動）:", str(sendall_str_b))
s.sendall(sendall_str_b)
data = s.recv(2048) # OK or OKR
now_tcp_str = data.decode(encoding="utf-8")
print("データ送信の可否:", now_tcp_str)
# 自分の行動からconfirmed_infoを更新
confirmed_info.update_coo_from_my_aciton(action, now_tcp_str)
# 相手の行動
data = s.recv(2048)
now_tcp_str = data.decode(encoding="utf-8")
print("相手の行動:", now_tcp_str)
# 相手の行動からconfirmed_infoを更新
# confirmed_info.update_coo_from_enemy_aciton(now_tcp_str)
# 更新が正常に動作しないことがあるため、毎回リセットする（仮置き）
confirmed_info.reset_keep_confirmed_info(now_tcp_str)
# 自分と相手で2ターン経過
turn_count += 2