class Trainer:
def __init__(self, agent):
self.agent = agent
self.name = "Trainer"
def start_training_session(self, num_of_episodes):
for episode_num in range(num_of_episodes):
self.headline_print("start game of episode number {}".format(episode_num + 1))
self.game_id = rest_api.create_game(self.name).content.decode("utf-8")
print("Trainer created game with id: {}".format(self.game_id))
self.tcp = TCP(self.game_id, self.name, self.on_recieved)
self.start_game_with_agent(self.game_id)
self.headline_print("end of episode number {}".format(episode_num + 1))
def start_game_with_agent(self, game_id):
env = QuoridorEnv(game_id, "Agent")
cur_state = env.reset()
done = False
steps_num = 0
while not done:
steps_num += 1
action = self.agent.act(cur_state, env)
new_state, reward, done, _ = env.step(action)
self.agent.remember(cur_state, action, reward, new_state, done)
self.agent.replay(env) # internally iterates default (prediction) model
self.agent.target_train() # iterates target model
cur_state = new_state
print("\n\nGAME FINISHED IN {} STEPS!\n\n".format(steps_num))
def on_recieved(self, json_message):
if json_message["type"] == "NewTurnEvent":
if json_message["nextPlayerToPlay"] == self.name:
myLoc = json_message["currentPosition"]
actions = []
for move in json_message["avialiableMoves"]:
if int(move["x"]) > int(myLoc["x"]):
actions.append(3) # Move Right
elif int(move["x"]) < int(myLoc["x"]):
actions.append(2) # Move Left
elif int(move["y"]) > int(myLoc["y"]):
actions.append(1) # Move Down
elif int(move["y"]) < int(myLoc["y"]):
actions.append(0) # Move Up
actions_len = len(actions)
random_i = np.random.randint(0, actions_len)
act_json = utils.convert_action_to_server(actions[random_i])
self.tcp.write(act_json)
elif json_message["type"] == "GameOverEvent":
pass
elif json_message["type"] == "RoomStateResponse":
if len(json_message["players"]) == 2:
rest_api.start_game(self.game_id)
def headline_print(self, text):
print("====================================================")
print(text)
print("====================================================")
class QuoridorEnv(gym.Env):
"""
players: {
index: number,
start_location: number,
name: string,
targets: [number]
}
"""
def __init__(self, game_id, player_name, is_not_tcp):
self.game_id = game_id
self.player_name = player_name
self.is_my_turn = False
self.winner_status = GameWinnerStatus.NoWinner
self.last_turn_illegal = False
self.action_options = []
self.winning_points_dim = np.zeros(shape=(9, 9), dtype=int)
# join_game(self.game_id, self.player_name)
if not is_not_tcp:
self.tcp = TCP(game_id, player_name, self.on_recieved)
self.wait_for_my_turn()
self.action_space = spaces.Discrete(Global.num_of_actions)
self.observation_space = spaces.Tuple((
spaces.MultiBinary([9, 9]),
spaces.MultiBinary([9, 9]),
spaces.MultiBinary([9, 9]),
spaces.MultiBinary([9, 9]),
spaces.MultiBinary([9, 9])
))
self.seed()
# Start the first game
#self.reset()
def step(self, action):
assert self.action_space.contains(action)
action = int(action)
self.update_board(action)
self.wait_for_my_turn()
reward, done = self.calculate_reward()
self.is_my_turn = False
# self.print_board()
return self.board, reward, done, {}
def wait_for_my_turn(self):
while not self.is_my_turn:
pass
def reset(self):
# self.board = self.init_board()
return self.board
def calculate_reward(self):
reward = -1
done = False
if self.last_turn_illegal:
reward = -10
self.last_turn_illegal = False
if self.winner_status != GameWinnerStatus.NoWinner:
done = True
if self.winner_status == GameWinnerStatus.EnvWinner:
reward = 200
elif self.winner_status == GameWinnerStatus.EnvLoser:
reward = -200
return reward, done
def update_board(self, action):
operation = utils.convert_action_to_server(action)
self.send_to_server(operation) # WAITING
def print_board(self):
os.system('cls')
print('------- TEAM 600 --------')
arrays = np.dsplit(self.board, 4)
for y in range(9):
for x in range(9):
print('|', end='')
if arrays[0][y][x] != 0:
print(1, end='')
elif arrays[1][y][x] != 0:
print(2, end='')
else:
print(' ', end='')
if arrays[3][y][x] != 0:
print('|', end='')
elif y != 0 & arrays[3][y - 1][x] != 0:
print('|', end='')
else:
print(' ', end='')
print('')
for x_wall in range(9):
if arrays[2][y][x_wall] != 0:
print(' ==', end='')
elif x_wall != 0 & arrays[2][y][x_wall - 1] != 0:
print(' ==', end='')
else:
print(' __', end='')
print('')
def get_and_convert_board(self):
board = json.loads(get_board(self.game_id).content)
return convert_board(board, self.winning_points_dim)
def send_to_server(self, operation):
self.tcp.write(operation)
def on_recieved(self, json_message):
if json_message["type"] == "IllegalMove":
# self.is_my_turn = True
self.last_turn_illegal = True
elif json_message["type"] == "NewTurnEvent":
if json_message["nextPlayerToPlay"] == self.player_name:
self.board = self.get_and_convert_board()
self.is_my_turn = True
self.update_action_options(json_message)
elif json_message["type"] == "GameOverEvent":
self.is_my_turn = True
if json_message["winnerName"] == self.player_name:
self.winner_status = GameWinnerStatus.EnvWinner
else:
self.winner_status = GameWinnerStatus.EnvLoser
elif json_message["type"] == "StartGameMessage":
self.update_winning_locations(json_message["players"])
def action_shape(self):
return action_shape()
def observation_shape(self):
return observation_shape()
def update_action_options(self, moves_json):
self.action_options = []
myLoc = moves_json["currentPosition"]
for move in moves_json["avialiableMoves"]:
if int(move["x"]) > int(myLoc["x"]):
self.action_options.append(3) # Move Right
elif int(move["x"]) < int(myLoc["x"]):
self.action_options.append(2) # Move Left
elif int(move["y"]) > int(myLoc["y"]):
self.action_options.append(1) # Move Down
elif int(move["y"]) < int(myLoc["y"]):
self.action_options.append(0) # Move Up
for wall in moves_json["availableWalls"]:
wall_action = 4 + wall["position"]["x"] + (8 * wall["position"]["y"])
if wall["wallDirection"] == "Down":
wall_action += 64
self.action_options.append(wall_action)
def get_action_options(self):
return self.action_options
def update_winning_locations(self, players):
for player in players:
if player["name"] == self.player_name:
for loc in player["endLine"]:
x = int(loc["x"])
y = int(loc["y"])
self.winning_points_dim[y, x] = 1
print(self.winning_points_dim)
