def __init__(self, initial):
""" Remplit la grille avec des valeurs qui respectent les carrés."""
state = np.array([frozenset(range(1, 10) if k == 0 else [k])
for k in initial]).reshape((9,9))
state = normalize_state(state)
self.initial = tuple(state.flatten())
def result(self, state, action):
"""
Calcule la configuration résultante à appliquer une action sur une
configuration.
Le nouvel état est une copie modifiée de l'état passé en argument.
"""
state = numpify_state(state)
i, j, k = action
state.itemset((i, j), frozenset([k]))
# normalize
normalized_state = normalize_state(state)
return tuple(normalized_state.flatten())
env_name = "Boxing-ram-v0"
try:
model_filename = sys.argv[2]
except IndexError:
model_filename = "Boxing-ram-v0_10000e.h5"
try:
num_episodes = int(sys.argv[3])
except IndexError:
num_episodes = 10
print("Playing agent {} for {} episodes.".format(model_filename, num_episodes))
env = gym.make(env_name)
agent = DQNAgent(env, epsilon=1.0, model_filename=model_filename)
for i_episode in range(num_episodes):
state = normalize_state(env.reset())
done = False
total_reward = 0
while not done:
env.render()
action = agent.act(state)
state, reward, done, info = env.step(action)
state = normalize_state(state)
total_reward += reward
print("Episode {} reward: {}".format(i_episode + 1, total_reward))
env.close()
num_episodes = int(sys.argv[2])
except IndexError:
num_episodes = 2000
env = gym.make(env_name)
agent = DQNAgent(env)
for i_episode in range(num_episodes):
# For timing every episode.
ts_start = datetime.now()
# For tracking accumulative reward.
total_reward = 0
lives = env.env.ale.lives()
state = normalize_state(env.reset())
done = False
while not done:
# Comment out env.render() for faster training.
# env.render()
action = agent.act(state)
next_state, reward, done, info = env.step(action)
next_state = normalize_state(next_state)
# If a life is lost, pass terminal state
if info["ale.lives"] < lives:
lives = info["ale.lives"]
done = True
agent.remember(state, action, reward, next_state, done)
state = next_state
total_reward += reward