def run(self):
log_dir = "./logs/bastille/MOISMCTSWithRandomRollouts_Vs_TabularQLearning/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(MOISMCTSWithRandomRolloutsAgent(
100, SafeWindJammersRunner(RandomAgent(), RandomAgent())),
TabularQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(1000000))
def run(self):
log_dir = "./logs/bastille/Random_Vs_RandomRollout_100/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
RandomRolloutAgent(
100,
SafeWindJammersRunner(
RandomAgent(),
RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(1000000))
def run(self):
log_dir = "./logs/bastille/TabularQLearning_MOISMCTSWithRandomRolloutsExpertThenApprentice/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
TabularQLearningAgent(),
MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent(
100, SafeWindJammersRunner(RandomAgent(), RandomAgent()),
8, 12),
checkpoint=100,
log_dir=log_dir).run(1000000))
def run(self):
log_dir = "./logs/bastille/TabularQLearning_MOISMCTSWithValueNetwork/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(TabularQLearningAgent(),
MOISMCTSWithValueNetworkAgent(
100,
SafeWindJammersRunner(
RandomAgent(),
RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(1000000))
def main():
# Set log level.
logging.basicConfig(level=logging.INFO)
# Set a random seed for consistency in agent AND environment.
if config.RANDOM_SEED is not None:
np.random.seed(config.RANDOM_SEED)
# Make environment.
env = EnvCatcher(grid_size=config.GRID_SIZE,
env_type='episodic',
verbose=False,
random_seed=config.RANDOM_SEED)
# Make agent.
agent = RandomAgent(actions=list(range(env.action_space)))
# Run the RL Loop.
episode_returns = rl.run_loop(agent=agent,
env=env,
max_num_steps=config.MAX_NUM_STEPS,
report_every_n=config.REPORT_EVERY_N)
# Save the data.
date_string = time.strftime("%Y%m%d-%H%M%S")
filename = ('random_grid_{}_nep_{}_'.format(
config.GRID_SIZE, len(episode_returns)) + date_string + '.csv')
rl.save_episode_returns(filename=filename, episode_returns=episode_returns)
def run_training_batch(n_games):
random.seed(time.localtime())
agent = Agent()
agent_types = [agent, CustomAgent(), RandomAgent()]
# agent_weights = [1.0/3.0, 1.0/3.0, 1]
# ^ Weights needed for even distribution of learning agent and random agent
# agent_weights = [1.0/3.0, 1, 1]
# ^ Weights needed for even distribution of learning agent and custom agent
agent_weights = [1.0 / 9.0, 4.0 / 9.0, 1]
start = time.perf_counter()
for i in range(n_games):
try:
list_of_agents = []
list_of_agents.append(agent)
for i in range(3):
rolled_weight = random.random()
for index in range(len(agent_types)):
if rolled_weight <= agent_weights[index]:
list_of_agents.append(agent_types[index])
break
random.shuffle(list_of_agents)
active_game = Game(i, list_of_agents)
active_game.play_game()
except Exception as err:
print(f"Error encountered... {err}")
end = time.perf_counter()
print(f"Ran {n_games} in {(end-start)} seconds.")
return True
def run(self):
log_dir = "./logs/bastille/Random_Vs_ReinforceClassic/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
ReinforceClassicAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(1000000))
def test_on_round_finish(self):
initial_player_points = []
for player in self.temp_players:
player.set_controlling_agent(RandomAgent())
self.temp_deck.deal_cards_to(player)
initial_player_points.append(player.get_round_points())
self.test_round.begin_play()
self.assertEqual(len(self.temp_deck.get_card_list()), 0)
def run(self):
log_dir = "./logs/bastille/Random_Vs_DoubleQLearning/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
DoubleQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(1000000))
def test_full_replay_cycle():
from environment.battlesnake_environment import BattlesnakeEnvironment
from agents.RandomAgent import RandomAgent
from environment.Battlesnake.importer.Importer import Importer
from environment.Battlesnake.renderer.game_renderer import GameRenderer
agents = [RandomAgent(), RandomAgent()]
env = BattlesnakeEnvironment(
width=15,
height=15,
agents=agents,
act_timeout=0.1,
export_games=True
)
env.reset()
while not env.game.is_game_over():
env.step()
assert os.path.exists(env.exporter.outpath)
assert os.path.getsize(env.exporter.outpath) > 0
game, turns, move_list = Importer.load_replay_file(env.exporter.outpath)
width, height = move_list[0].width, move_list[0].width
num_snakes = len(move_list[0].snakes)
renderer = GameRenderer(width, height, num_snakes)
assert len(move_list) >= 1
assert len(move_list) == turns
assert game
for move in move_list:
renderer.display(move)
os.unlink(env.exporter.outpath)
# TODO: Could be moved special place, so that always at the end
import shutil
shutil.rmtree("replays", ignore_errors=True)
shutil.rmtree("replay_test", ignore_errors=True)
def run(self):
log_dir = "./logs/bastille/Random_Vs_PPOWithMultipleTrajectoriesMultiOutputs" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
RandomAgent(),
PPOWithMultipleTrajectoriesMultiOutputsAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(1000000))
def setUp(self):
# Set a random seed for consistency in agent AND environment.
if RANDOM_SEED is not None:
np.random.seed(RANDOM_SEED)
# Make environment.
self.env = EnvCatcher(grid_size=GRID_SIZE,
env_type='episodic',
verbose=False,
random_seed=RANDOM_SEED)
# Make agent.
self.agent = RandomAgent(actions=list(range(self.env.action_space)))
def sample_env():
from agents.RandomAgent import RandomAgent
from agents.SimpleAgent_solution import SimpleAgent
from environment.battlesnake_environment import BattlesnakeEnvironment
agents = [RandomAgent(), SimpleAgent()]
env = BattlesnakeEnvironment(
width=15,
height=15,
agents=agents,
act_timeout=0.2
)
env.reset()
return env
def test_file_out_behavior(self):
for player in self.temp_players:
player.set_controlling_agent(RandomAgent())
self.temp_game.play_game()
with open(self.test_round.get_file_out_name(), 'rb') as input:
file_data = pickle.load(input)
for i in range(len(file_data)):
self.assertTrue(self.test_round._get_file_out_data()[i]["trick_history"].all() == file_data[i]["trick_history"].all())
self.assertGreater(self.test_round._get_file_out_data()[i]["trick_history"].sum(), 0)
self.assertTrue(self.test_round._get_file_out_data()[i]["trick_point_history"].all() == file_data[i]["trick_point_history"].all())
self.assertGreaterEqual(self.test_round._get_file_out_data()[i]["trick_point_history"].sum(), 0)
self.assertTrue(self.test_round._get_file_out_data()[i]["player_partners"].all() == file_data[i]["player_partners"].all())
self.assertGreater(self.test_round._get_file_out_data()[i]["player_partners"].sum(), 0)
self.assertTrue(self.test_round._get_file_out_data()[i]["call_matrix"].all() == file_data[i]["call_matrix"].all())
self.assertGreater(self.test_round._get_file_out_data()[i]["call_matrix"].sum(), 0)
self.assertTrue(self.test_round._get_file_out_data()[i]["player_cards_in_hand_history"].all() == file_data[i]["player_cards_in_hand_history"].all())
self.assertGreater(self.test_round._get_file_out_data()[i]["player_cards_in_hand_history"].sum(), 0)
self.assertTrue(self.test_round._get_file_out_data()[i]["player_point_history"].all() == file_data[i]["player_point_history"].all())
self.assertGreater(self.test_round._get_file_out_data()[i]["player_point_history"].sum(), 0)
self.assertTrue(self.test_round._get_file_out_data()[i]["player_partner_prediction_history"].all() == file_data[i]["player_partner_prediction_history"].all())
for e in range(len(file_data[i]["player_score_history"])):
self.assertTrue(self.test_round._get_file_out_data()[i]["player_score_history"][e] == file_data[i]["player_score_history"][e])
def main():
"""Running parameters"""
atari_game: str = "Qbert-v4"
# atari_game: str = "Pong-v4"
number_of_episodes: int = 20000 # Number of episodes to run
episode_number: int = 1
env = gym.make(atari_game)
"Create agents"
random_agent: RandomAgent = RandomAgent(env)
q_table_agent: QTableAgent = QTableAgent(env)
while episode_number <= number_of_episodes:
random_agent.play_an_episode()
q_table_agent.play_an_episode()
print()
if episode_number % 100 == 0:
q_table_agent.compare_performance_to(random_agent, 100)
print()
episode_number += 1
print(q_table_agent)
def run(self):
if self.opponent == "RandomAgent":
log_dir1 = self.log_dir_root + "DoubleQLearningAgent_VS_RandomAgent_" + self.time
print(log_dir1)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
RandomAgent(),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir1).run(100000000))
elif self.opponent == "TabularQLearningAgent":
log_dir2 = self.log_dir_root + "DoubleQLearningAgent_VS_TabularQLearningAgent_" + self.time
print(log_dir2)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
TabularQLearningAgent(),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir2).run(100000000))
elif self.opponent == "DeepQLearningAgent":
log_dir3 = self.log_dir_root + "DoubleQLearningAgent_VS_DeepQLearningAgent_" + self.time
print(log_dir3)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
DeepQLearningAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir3).run(100000000))
elif self.opponent == "ReinforceClassicAgent":
log_dir4 = self.log_dir_root + "DoubleQLearningAgent_VS_ReinforceClassicAgent_" + self.time
print(log_dir4)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
ReinforceClassicAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir4).run(100000000))
elif self.opponent == "ReinforceClassicWithMultipleTrajectoriesAgent":
log_dir5 = self.log_dir_root + "DoubleQLearningAgent_VS_ReinforceClassicWithMultipleTrajectoriesAgent_" + self.time
print(log_dir5)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
ReinforceClassicWithMultipleTrajectoriesAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir5).run(100000000))
elif self.opponent == "PPOWithMultipleTrajectoriesMultiOutputsAgent":
log_dir6 = self.log_dir_root + "DoubleQLearningAgent_VS_PPOWithMultipleTrajectoriesMultiOutputsAgent_" + self.time
print(log_dir6)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
PPOWithMultipleTrajectoriesMultiOutputsAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir6).run(100000000))
elif self.opponent == "MOISMCTSWithRandomRolloutsAgent":
log_dir7 = self.log_dir_root + "DoubleQLearningAgent_VS_MOISMCTSWithRandomRolloutsAgent_" + self.time
print(log_dir7)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
MOISMCTSWithRandomRolloutsAgent(100,
SafeTicTacToeRunner(RandomAgent(),
RandomAgent())),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir7).run(1000000000))
elif self.opponent == "MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent":
log_dir8 = self.log_dir_root + "DoubleQLearningAgent_VS_MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent_" + self.time
print(log_dir8)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent(100,
SafeTicTacToeRunner(
RandomAgent(),
RandomAgent()),9,9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir8).run(1000000000))
elif self.opponent == "MOISMCTSWithValueNetworkAgent":
log_dir9 = self.log_dir_root + "DoubleQLearningAgent_VS_MOISMCTSWithValueNetworkAgent_" + self.time
print(log_dir9)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
MOISMCTSWithValueNetworkAgent(100,
SafeTicTacToeRunner(RandomAgent(),
RandomAgent())),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir9).run(1000000000))
elif self.opponent == "DoubleQLearningAgent":
log_dir10 = self.log_dir_root + "DoubleQLearningAgent_VS_DoubleQLearningAgent_" + self.time
print(log_dir10)
print(TensorboardTicTacToeRunner(DoubleQLearningAgent(),
DoubleQLearningAgent(),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir9).run(1000000000))
elif self.opponent == "RandomRolloutAgent":
nb_rollouts = 3
log_dir11 = self.log_dir_root + "RandomAgent_VS_RandomRolloutAgent(" + str(nb_rollouts) + ")_" + self.time
print(log_dir11)
print(TensorboardTicTacToeRunner(RandomAgent(),
RandomRolloutAgent(nb_rollouts,
SafeTicTacToeRunner(
RandomAgent(),
RandomAgent())),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir11).run(1000000000))
else:
print("Unknown opponent")
def run(agent_id, is_training, load_dir):
env = gym.make('LunarLanderContinuous-v2')
time_format = strftime("%Y%m%d%H%M%S", gmtime())
outdir = './results/' + time_format + '-' + agent_id + '/'
if not os.path.exists(outdir + 'video/'):
os.makedirs(outdir + 'video/')
#Enable below comment to record video
#env = wrappers.Monitor(env, directory=outdir + 'video/', force=True)
env.seed(123)
np.random.seed(123)
#set up logger
logging.basicConfig(filename=outdir + 'application.log',
level=logging.INFO)
logger = logging.getLogger('gym')
agent = None
if (agent_id == 'RandomAgent'):
agent = RandomAgent(env)
elif (agent_id == 'DDPG'):
agent = DDPG(env)
agent.build_models()
elif (agent_id == 'RBFAgent'):
agent = RBFAgent(env)
else:
logger.error("Invalid Agent chosen!")
return
if load_dir is not None:
logger.info("Load model at " + load_dir)
#agent.load_weight(load_dir + '/')
agent.load_model(load_dir + '/')
# Initialize variables
episode = np.int16(1)
step = np.int16(0)
max_steps = 1000000
max_episodes = 1100 + episode
if not is_training:
logger.info("it is now testing")
K.set_learning_phase(0)
max_episodes = 5 + episode
else:
K.set_learning_phase(1)
file_output = outdir + 'reward.csv'
labels = ['episode', 'reward']
with open(file_output, 'w') as csvfile:
writer = csv.writer(csvfile)
writer.writerow(labels)
observation = None
episode_reward = None
try:
while step < max_steps and episode < max_episodes:
if observation is None: # start of a new episode
logger.info("Start episode " + str(episode))
episode_reward = np.float32(0)
# Obtain the initial observation by resetting the environment.
agent.reset_episode()
observation = deepcopy(env.reset())
action = agent.act(observation)
reward = np.float32(0)
done = False
next_state, r, done, info = env.step(action)
next_state = deepcopy(next_state)
reward = r
if is_training:
agent.step(observation, action, reward, next_state, done)
episode_reward += reward
step += 1
observation = next_state
if episode % 20 == 0:
logger.info(
"State-Action: {}, {}, {}, {}, reward: {}".format(
observation[0], observation[1], action[0],
action[1], reward))
if done:
# Act on the final state
# Step on final state but without adding to memory as next state is the reset state
if is_training:
action = agent.act(next_state)
agent.step_without_memory()
to_write = [episode] + [episode_reward]
writer.writerow(to_write)
if episode % 20 == 0:
csvfile.flush()
if episode % 20 == 0:
agent.save_weight(outdir)
episode += 1
observation = None
episode_reward = None
except KeyboardInterrupt:
csvfile.flush()
agent.save_weight(outdir)
# Close the env
env.close()
def test_play_with_learning_agent(self):
self.temp_player_list[0].set_controlling_agent(Agent())
for index in range(1, 4):
self.temp_player_list[index].set_controlling_agent(RandomAgent())
self.temp_game.play_game()
self.assertGreaterEqual(max(self.temp_game.get_score_list()), 42)
def test_play_game(self):
for player in self.temp_player_list:
player.set_controlling_agent(RandomAgent())
self.temp_game.play_game()
self.assertGreaterEqual(max(self.temp_game.get_score_list()), 42)
def test_update_scores(self):
for player in self.temp_player_list:
player.set_controlling_agent(RandomAgent())
self.temp_game.begin_round()
self.assertNotEqual(sum(self.temp_game.get_score_list()), 0)
# MOISMCTSWithRandomRolloutsAgent(100,
# SafeTicTacToeRunner(RandomAgent(), RandomAgent())),
# log_and_reset_score_history_threshold=10000,
# log_dir=log_dir).run(1000000000))
#
# log_dir = "./logs/Random_vs_all/RandomAgent_VS_MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent_" + str(time())
# print(log_dir)
# print(TensorboardTicTacToeRunner(RandomAgent(),
# MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent(100, SafeTicTacToeRunner(
# RandomAgent(), RandomAgent()),9,9),
# log_and_reset_score_history_threshold=10000,
# log_dir=log_dir).run(1000000000))
#
log_dir = "./logs/Random_vs_all/RandomAgent_VS_MOISMCTSWithValueNetworkAgent_" + str(time())
print(log_dir)
print(TensorboardTicTacToeRunner(RandomAgent(),
MOISMCTSWithValueNetworkAgent(100,"1",
SafeTicTacToeRunner(RandomAgent(), RandomAgent())),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir).run(1000000000))
# AGENTS EXAMPLES :
# CommandLineAgent()
# RandomAgent()
# RandomRolloutAgent(3, SafeTicTacToeRunner(RandomAgent(), RandomAgent()))
# TabularQLearningAgent()
# DeepQLearningAgent(9,9)
# ReinforceClassicAgent(9,9)
# ReinforceClassicWithMultipleTrajectoriesAgent(9,9)
# PPOWithMultipleTrajectoriesMultiOutputsAgent(9,9)
# MOISMCTSWithRandomRolloutsAgent(100, SafeTicTacToeRunner(RandomAgent(), RandomAgent()))
scores = self.generate_midgame_env(num_agent_init_card)["scores"]
print scores
cnt_win[np.argmin(scores)] += 1
print cnt_win
def test_66(self, num_round=100):
cnt_win = [
0,
] * len(self.agent_list)
for i in range(num_round):
scores = [
0,
] * len(self.agent_list)
while max(scores) < 66:
score = self.generate_midgame_env(
num_agent_init_card)["scores"]
scores = [
scores[j] + score[j] for j in range(len(self.agent_list))
]
cnt_win[np.argmin(scores)] += 1
print "Round %d :" % i, scores, cnt_win
print cnt_win
if __name__ == "__main__":
gameEmulator = GameEmulator()
#自己的Agent（调用execute_action的那个）编号默认为0，应当被第一个添加进去
gameEmulator.add_agent(KeyboardAgent())
gameEmulator.add_agent(RandomAgent())
gameEmulator.emulate(1)
if self.print_and_reset_score_history_threshold is not None and \
round_id % self.print_and_reset_score_history_threshold == 0:
print(score_history /
self.print_and_reset_score_history_threshold)
if self.prev_history is not None and \
score_history[0] == self.prev_history[0] and \
score_history[1] == self.prev_history[1] and \
score_history[2] == self.prev_history[2]:
self.stuck_on_same_score += 1
else:
self.prev_history = score_history
self.stuck_on_same_score = 0
if (self.
replace_player1_with_commandline_after_similar_results
is not None and self.stuck_on_same_score >= self.
replace_player1_with_commandline_after_similar_results
):
self.agents = (CommandLineAgent(), self.agents[1])
self.stuck_on_same_score = 0
return tuple(score_history)
if __name__ == "__main__":
print("MOISMCTSWithRandomRolloutsAgent VS RandomAgent")
print(
BasicTicTacToeRunner(
MOISMCTSWithRandomRolloutsAgent(
100, SafeTicTacToeRunner(RandomAgent(), RandomAgent())),
RandomAgent(),
print_and_reset_score_history_threshold=1000).run(1000))
def main(ppo=False, dqn=False, ddqn=False, pg=False, random=False):
"""
main
Parameters
----------
ppo : BOOLEAN, optional
SET TO TRUE TO RUN PPO AGENT. The default is False.
dqn : BOOLEAN, optional
SET TO TRUE TO RUN DQN AGENT. The default is False.
ddqn : BOOLEAN, optional
SET TO TRUE TO RUN DDQN AGENT. The default is False.
pg : BOOLEAN, optional
SET TO TRUE TO RUN PG AGENT. The default is False.
random : BOOLEAN, optional
SET TO TRUE TO RUN RANDOM AGENT. The default is False.
Returns
-------
None.
"""
# Running DQN Agent
if dqn:
env = gym.make('PongDeterministic-v4')
action_size = env.action_space.n
agent = DqnAgent(action_size)
nn = AtariModel()
nn.build_atari_model(action_size)
model = nn.model
memory = RingBuffer(dqn_settings.MEMORY_SIZE)
for i in range(dqn_settings.NUMBER_EPISODES):
frame = env.reset()
frame = agent.image_preprocessing(frame)
state = (frame, frame, frame, frame)
env.render()
finished = False
summed_reward = 0
while not finished:
action = agent.choose_action(state, model)
next_frame, reward, finished, _ = env.step(action)
next_frame = agent.image_preprocessing(next_frame)
reward = agent.transform_reward(reward)
next_state = (next_frame, state[0], state[1], state[2])
summed_reward += reward
memory.append((state, action, next_state, reward, finished))
state = next_state
env.render()
if (i > dqn_settings.ITERATIONS_BEFORE_FIT):
minibatch = memory.sample_random_batch(
dqn_settings.BATCH_SIZE)
agent.fit_batch(model, minibatch, action_size)
if (agent.epsilon > dqn_settings.FINAL_EPSILON):
agent.epsilon = agent.epsilon * dqn_settings.GAMMA
print("Iteration:", i, " Reward:", summed_reward, "Epsilon:",
agent.epsilon)
f = open("rewards.txt", "a")
f.write(str(summed_reward) + "\n")
f.close()
if (i % 100 == 0):
model.save('models/saved_model')
# Running PPO Agent
elif ppo:
policy = PpoAgent()
run_ppo(policy)
# Running DDQN Agent
elif ddqn:
ddqn_agent = DdqnAgent()
ddqn_agent.run_ddqn()
# Running PG Agent
elif pg:
pg_agent = PgAgent()
pg_agent.run_pg()
# Running Random Agent
elif random:
random_agent = RandomAgent()
random_agent.run_random()
# If no agent is selected a message will print asking to select one
else:
print(
'No agent selected to run! Please select an agent: dqn, ppo, ddqn, pg, random'
)
from environment.battlesnake_environment import BattlesnakeEnvironment
from agents.RandomAgent import RandomAgent
from agents.HumanPlayer import HumanPlayer
import pygame
nb_random_player = 2
num_fruits = 1
agents = [HumanPlayer()]
for i in range(nb_random_player):
agents.append(RandomAgent())
env = BattlesnakeEnvironment(
width=15,
height=15,
num_fruits=num_fruits,
agents=agents,
)
env.reset()
env.render()
while True:
env.handle_input()
env.step()
env.render()
pygame.time.wait(250)
self.prev_history = score_history
self.stuck_on_same_score = 0
if (self.replace_player1_with_commandline_after_similar_results is not None and
self.stuck_on_same_score >= self.replace_player1_with_commandline_after_similar_results):
self.agents = (CommandLineAgent(), self.agents[1])
self.stuck_on_same_score = 0
score_history = np.array((0, 0, 0))
self.execution_time = np.array((0.0, 0.0))
return tuple(score_history)
if __name__ == "__main__":
number = [1000, 10000, 100000, 1000000]
versus_name = ['RandomAgent', 'Tabular', 'DQN', 'DDQN', 'Reinforce', 'Reinforce A2C Style', 'PPO', 'MCTS']
versus_agent = [RandomAgent(),
TabularQLearningAgent(),
DeepQLearningAgent(9, 9),
DoubleDeepQLearningAgent(9, 9),
ReinforceClassicAgent(9, 9),
ReinforceClassicWithMultipleTrajectoriesAgent(9, 9),
PPOWithMultipleTrajectoriesMultiOutputsAgent(9, 9),
MOISMCTSWithValueNetworkAgent(9, 9, 2)]
versus = [versus_name, versus_agent]
for num in number:
for i in range(len(versus_name)):
with open("D:/DEEP_LEARNING/Reinforcement/TabularVS" + str(versus[0][i]) +"_NB_"+ str(num) + ".csv", 'w+') as f: #Ici change TabularVS par le nom de l'agent que tu lance contre tout le reste
print("New Fight" + str(versus[0][i]) + " " + str(num))
begin = time()
# TabularQLearningAgent(),
# log_and_reset_score_history_threshold=1000,
# log_dir=log_dir).run(1000000000))
# log_dir = "./logs/TabularQLearning_VS_TabularQLearning_" + str(time())
# print(log_dir)
# print(TensorboardPower4Runner(TabularQLearningAgent(),
# TabularQLearningAgent(),
# log_and_reset_score_history_threshold=1000,
# log_dir=log_dir).run(1000000000))
log_dir = "./logs/RandomRollout_VS_TabularQLearning_" + str(time())
print(log_dir)
print(
TensorboardPower4Runner(RandomRolloutAgent(
3, SafePower4Runner(RandomAgent(), RandomAgent())),
TabularQLearningAgent(),
log_and_reset_score_history_threshold=1000,
log_dir=log_dir).run(1000000000))
# AGENTS EXAMPLES :
# CommandLineAgent()
# RandomAgent()
# RandomRolloutAgent(3, SafeTicTacToeRunner(RandomAgent(), RandomAgent()))
# TabularQLearningAgent()
# DeepQLearningAgent(9,9)
# ReinforceClassicAgent(9,9)
# ReinforceClassicWithMultipleTrajectoriesAgent(9,9)
# PPOWithMultipleTrajectoriesMultiOutputsAgent(9,9)
# MOISMCTSWithRandomRolloutsAgent(100, SafeTicTacToeRunner(RandomAgent(), RandomAgent()))
# MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent(100, SafeTicTacToeRunner(RandomAgent(), RandomAgent()),9,9)
round_step += 1
self.writer.add_summary(
tf.Summary(value=[
tf.Summary.Value(
tag="agent1_action_mean_duration",
simple_value=self.mean_action_duration_sum[0] /
round_step),
tf.Summary.Value(
tag="agent2_action_mean_duration",
simple_value=self.mean_action_duration_sum[1] /
round_step),
tf.Summary.Value(
tag="agent1_accumulated_reward",
simple_value=self.mean_accumulated_reward_sum[0]),
tf.Summary.Value(
tag="agent2_accumulated_reward",
simple_value=self.mean_accumulated_reward_sum[1])
], ), episode_id)
episode_id += 1
if __name__ == "__main__":
print("Rdm vs Rdm")
print(
TensorboardInstrumentedWindJammersRunner(
TabularQLearningAgent(),
RandomAgent(),
log_dir_root="./logs/Rdm_Vs_Rdm").run(1000))
def run():
log_dir = "./logs/bastilleMP/ReinforceWithMultipleTraj_Vs_TabularQLearning/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
ReinforceClassicWithMultipleTrajectoriesAgent(8, 12),
TabularQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/DeepQLearning_Vs_TabularQLearning/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(DeepQLearningAgent(8, 12),
TabularQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/MOISMCTSWithRandomRollouts_Vs_TabularQLearning/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(MOISMCTSWithRandomRolloutsAgent(
100, SafeWindJammersRunner(RandomAgent(), RandomAgent())),
TabularQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/ReinforceClassicWithMultipleTrajectories_Vs_TabularQLearningAgent/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
ReinforceClassicWithMultipleTrajectoriesAgent(8, 12),
TabularQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_TabularQLearningAgent/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
TabularQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_RandomRollout_100/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
RandomRolloutAgent(
100,
SafeWindJammersRunner(
RandomAgent(), RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_DeepQLearning/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
DeepQLearningAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_DoubleQLearning/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
DoubleQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_ReinforceClassic/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
ReinforceClassicAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_ReinforceClassicWithMultipleTrajectories/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
RandomAgent(),
ReinforceClassicWithMultipleTrajectoriesAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_PPOWithMultipleTrajectoriesMultiOutputs" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
RandomAgent(),
PPOWithMultipleTrajectoriesMultiOutputsAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_MOISMCTSWithRandomRollouts/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
MOISMCTSWithRandomRolloutsAgent(
100,
SafeWindJammersRunner(
RandomAgent(), RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_MOISMCTSWithRandomRolloutsExpertThenApprentice/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
RandomAgent(),
MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent(
100, SafeWindJammersRunner(RandomAgent(), RandomAgent()), 8,
12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/Random_Vs_MOISMCTSWithValueNetwork/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(RandomAgent(),
MOISMCTSWithValueNetworkAgent(
100,
SafeWindJammersRunner(
RandomAgent(), RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_RandomRollout_100/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(TabularQLearningAgent(),
RandomRolloutAgent(
100,
SafeWindJammersRunner(
RandomAgent(), RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_DeepQLearning/" + str(time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(TabularQLearningAgent(),
DeepQLearningAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_DoubleQLearning/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(TabularQLearningAgent(),
DoubleQLearningAgent(),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_ReinforceClassic/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(TabularQLearningAgent(),
ReinforceClassicAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_ReinforceClassicWithMultipleTrajectories/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
TabularQLearningAgent(),
ReinforceClassicWithMultipleTrajectoriesAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_PPOWithMultipleTrajectoriesMultiOutputs" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
TabularQLearningAgent(),
PPOWithMultipleTrajectoriesMultiOutputsAgent(8, 12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_MOISMCTSWithRandomRollouts/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(TabularQLearningAgent(),
MOISMCTSWithRandomRolloutsAgent(
100,
SafeWindJammersRunner(
RandomAgent(), RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_MOISMCTSWithRandomRolloutsExpertThenApprentice/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(
TabularQLearningAgent(),
MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent(
100, SafeWindJammersRunner(RandomAgent(), RandomAgent()), 8,
12),
checkpoint=100,
log_dir=log_dir).run(100000))
log_dir = "./logs/bastilleMP/TabularQLearning_MOISMCTSWithValueNetwork/" + str(
time())
print(str(log_dir))
print(
TensorboardWindJammersRunner(TabularQLearningAgent(),
MOISMCTSWithValueNetworkAgent(
100,
SafeWindJammersRunner(
RandomAgent(), RandomAgent())),
checkpoint=100,
log_dir=log_dir).run(100000))
input("Which call do you want to make?(enter an index)"))
return call_to_play_index
def play_card(self, a_player, a_game):
valid_play_list = a_player.get_valid_play_list()
cards_in_hand_list = [
card.get_card_id() for card in a_player.get_cards_in_hand()
]
print(f"Your cards in hand are: {cards_in_hand_list}")
playable_cards_list = [
cards_in_hand_list[index] for index in range(len(valid_play_list))
if valid_play_list[index] != 0
]
print(f"Your playable cards are: {playable_cards_list}")
current_trick = a_game.get_trick()
cards_on_trick = [
str(card) for card in current_trick.get_played_cards_list()
]
print(f"The cards on trick are: {cards_on_trick}")
card_to_play = -1
while card_to_play not in playable_cards_list:
card_to_play = int(input("Which card do you want to play?"))
for index in range(len(cards_in_hand_list)):
if card_to_play == cards_in_hand_list[index]:
return index
if __name__ == "__main__":
agent_types = [HumanAgent(), Agent(), CustomAgent(), RandomAgent()]
active_game = Game(0, agent_types)
active_game.play_game()
