def train_agent(self, random_agents_interval):
self.log_training_info()
for i in range(self.num_games):
if i % 100 == 0:
self.logger.info("Playing training round {}".format(i))
wiz = Wizard(num_players=self.num_players, players=self.players)
wiz.play()
if i > 0 and i % random_agents_interval == 0:
self.play_evaluation_games()
def test_play_with_RLAgent(self):
from Environment.Wizard import Wizard
seed(2)
players = [RandomPlayer() for _ in range(5)]
players.append(RLAgent())
wiz = Wizard(players=players)
self.assertIsNotNone(wiz.play())
def train_agent(self):
self.log_training_info()
for i in range(self.num_games):
if i % 100 == 0:
self.logger.info("Playing round {}".format(i))
wiz = Wizard(num_players=self.num_players, players=self.players)
scores = wiz.play()
self.update_scores(scores, i)
def collect_training_data(self, players):
x = None
y = None
for i in range(self.training_rounds):
wizard = Wizard(players=players, track_tricks=True)
wizard.play()
temp_x, temp_y = wizard.get_history()
if x is None:
x = temp_x
y = temp_y
else:
x = np.concatenate((x, temp_x), axis=0)
y = np.concatenate((y, temp_y), axis=0)
# temporärer Tracker
if i % 100 == 0:
self.logger.info("Trick Prediction Initializer: Round {} finished".format(i))
return x, y
def test_several_RL_one_estimator(self):
from Environment.Wizard import Wizard
games = 20000
players = [RLAgent()]
players[0].load_estimator()
for rl_player in range(3):
players.append(RLAgent(estimator=players[0].estimator))
players.append(AverageRandomPlayer())
players.append(RandomPlayer())
scores = []
for i in range(games):
if i % 100 == 0:
print("{}/{}".format(i, games))
wiz = Wizard(players=players)
scores.append(wiz.play())
players[0].save_estimator()
scores = np.array(scores)
plotting.plot_moving_average_scores(scores, 100)
def play_evaluation_games(self):
self.disable_training()
self.t_eval += 1
self.logger.info("Starting a train evaluation game with random players")
players = [AverageRandomPlayer() for _ in range(len(self.players) - 1)]
players.append(self.players[0])
# arrays for evaluation scores and
eval_scores = np.zeros((self.evaluation_games, len(players)))
eval_wins = np.zeros((self.evaluation_games, len(players)))
for i in range(self.evaluation_games):
wiz = Wizard(num_players=len(players), players=players)
scores = wiz.play()
# evaluate scores
eval_scores[i] = scores
eval_wins[i][scores == np.max(scores)] = 1
self.update_scores(eval_scores, eval_wins)
self.enable_training()
def test_play(self):
from Environment.Wizard import Wizard
seed(2)
wiz = Wizard(4)
self.assertIsNotNone(wiz.play())