def main(max_episodes: int, max_episode_len, env: gym.Env, gym_dir: str):
seed = 0
np.random.seed(seed)
tf.set_random_seed(seed)
env.seed(seed)
render_env = True
use_gym_monitor = True
agent = TensorFlowDDPGAgent(
state_dim=env.observation_space.shape[0],
action_space=env.action_space,
)
if use_gym_monitor:
env = wrappers.Monitor(env, gym_dir, force=True)
Experiment(
agent=agent,
env=env,
render_env=render_env,
num_episodes=max_episodes,
max_episode_len=max_episode_len,
).run(seed=seed)
if use_gym_monitor:
env.close()
def test_tf_ddpg_agent_trigger_train(cleanup):
replay_buffer = InMemoryReplayBuffer(lower_size_limit=2, batch_size=2)
agent = TensorFlowDDPGAgent(state_dim=STATE_DIM, action_space=ACTION_SPACE,
replay_buffer=replay_buffer)
with mock.patch.object(agent, "_train", wraps=agent._train) as train_spy:
agent.observe_episode(EPISODE)
agent.act(current_state=STATE_SPACE.sample())
train_spy.assert_called()
def test_tf_ddpg_agent_reject_invalid_seed(flower: TensorFlowDDPGAgent):
for invalid_seed in [None, "one"]:
with pytest.raises(TypeError):
flower.set_seed(invalid_seed)
def test_tf_ddpg_agent_set_seed(flower: TensorFlowDDPGAgent):
flower.set_seed(1)
def test_tf_ddpg_agent_rejects_invalid_episodes(flower: TensorFlowDDPGAgent):
for invalid_episode in [None, "one"]:
with pytest.raises(TypeError):
flower.observe_episode(invalid_episode)
def test_tf_ddpg_agent_observe_episode(flower: TensorFlowDDPGAgent, cleanup):
"""
Observing an episode may trigger model saving, so we need to remove
the created folder.
"""
flower.observe_episode(EPISODE)
def test_tf_ddpg_agent_act(flower: TensorFlowDDPGAgent):
action = flower.act(current_state=STATE_SPACE.sample())
assert ACTION_SPACE.contains(action)
def flower():
return TensorFlowDDPGAgent(state_dim=STATE_DIM, action_space=ACTION_SPACE)