def test_nested_ob(self):
"""Test."""
env = make_env('CartPole-v1')
qf = QFunction(
FeedForwardBase(env.observation_space, env.action_space))
env = NestedVecObWrapper(env)
buffer = ReplayBuffer(2000, 1)
data_manager = ReplayBufferDataManager(buffer,
env,
act_fn=BufferActor(qf),
device='cpu',
learning_starts=50,
update_period=2)
for _ in range(11):
data_manager.step_until_update()
assert buffer.num_in_buffer == 70
batch = data_manager.sample(32)
data_manager.act(batch['obs'])
assert batch['action'].shape == batch['reward'].shape
assert batch['action'].shape == batch['done'].shape
if isinstance(batch['obs'], list):
assert batch['obs'][0].shape == batch['next_obs'][0].shape
assert len(batch['obs'][0].shape) == 2
else:
assert batch['obs'].shape == batch['next_obs'].shape
assert len(batch['obs'].shape) == 2
assert len(batch['action'].shape) == 1
def qf_fn(env):
"""Create a qfunction."""
return QFunction(QFBase(env.observation_space, env.action_space))
def nature_dqn_fn(env):
"""Create nature dqn qfunction."""
return QFunction(NatureDQN(env.observation_space, env.action_space))
def qf_fn(env):
"""Create a qfunction network."""
return QFunction(
AppendActionFeedForwardQFBase(env.observation_space, env.action_space))
def qf_fn_discrete(env):
"""Create a qfunction network."""
return QFunction(FeedForwardQFBase(env.observation_space,
env.action_space))
def qf_fn(env):
return QFunction(
NatureDQN(env.observation_space, env.action_space))