def test(env, base, batch_size, nested, rollout_length):
pi = Policy(base(env.observation_space, env.action_space))
if nested:
env = NestedVecObWrapper(env)
nenv = env.num_envs
data_manager = RolloutDataManager(env,
RolloutActor(pi),
'cpu',
batch_size=batch_size,
rollout_length=rollout_length)
for _ in range(3):
data_manager.rollout()
count = 0
for batch in data_manager.sampler():
assert 'key1' in batch
count += 1
assert 'done' in batch
data_manager.act(batch['obs'])
if data_manager.recurrent:
assert count == np.ceil(nenv / data_manager.batch_size)
else:
n = data_manager.storage.get_rollout()['reward'].data.shape[0]
assert count == np.ceil(n / data_manager.batch_size)
def test_policy_base(self):
"""Test Policy base."""
class Base(PolicyBase):
def forward(self, ob):
return DeltaDist(torch.ones([2], dtype=torch.float32))
env = gym.make('LunarLanderContinuous-v2')
pi = Policy(Base(env.observation_space, env.action_space))
pi2 = UnnormActionPolicy(
Base(env.observation_space, env.action_space))
ob = env.reset()
outs = pi(ob[None])
assert outs.value is None
assert outs.state_out is None
outs2 = pi2(ob[None])
assert outs.value is None
assert outs.state_out is None
ac_normed = 2 * (outs2.action - pi2.low) / (pi2.high -
pi2.low) - 1.
assert torch.allclose(ac_normed, outs.action)
assert torch.allclose(outs2.normed_action, outs.action)
def policy_fn(env):
return Policy(
FeedForwardActorCriticBase(env.observation_space,
env.action_space))
def policy_fn(env):
"""Create a policy."""
return Policy(PiBase(env.observation_space, env.action_space))
def policy_fn(env):
"""Create a policy network."""
return Policy(
FeedForwardPolicyBase(env.observation_space, env.action_space))
def continuous_policy_fn(env):
"""Create policy."""
return Policy(ContinuousPolicyBase(env.observation_space,
env.action_space))
def discrete_policy_fn(env):
"""Create policy."""
return Policy(DiscretePolicyBase(env.observation_space, env.action_space))
def drone_ppo_policy_fn(env, nunits=128):
"""Create a policy network."""
return Policy(
FeedForwardActorCriticBase(env.observation_space,
env.action_space,
nunits=nunits))
def policy_fn(env):
return Policy(NatureDQN(env.observation_space,
env.action_space))
def policy_fn(env):
"""Create policy."""
return Policy(NatureDQN(env.observation_space, env.action_space))
def policy_fn():
return Policy(TicTacToeNet())
def a3c_rnn_fn(env):
"""Create a3c recurrent policy."""
return Policy(A3CRNN(env.observation_space, env.action_space))
def a3c_cnn_fn(env):
"""Create a3c conv net policy."""
return Policy(A3CCNN(env.observation_space, env.action_space))