def __init__(
self,
environment_factory: jax_types.EnvironmentFactory,
network_factory: NetworkFactory,
sac_fd_config: SACfDConfig,
lfd_iterator_fn: Callable[[], Iterator[builder.LfdStep]],
seed: int,
num_actors: int,
environment_spec: Optional[specs.EnvironmentSpec] = None,
max_number_of_steps: Optional[int] = None,
log_to_bigtable: bool = False,
log_every: float = 10.0,
evaluator_factories: Optional[Sequence[
distributed_layout.EvaluatorFactory]] = None,
):
logger_fn = functools.partial(loggers.make_default_logger,
'learner',
log_to_bigtable,
time_delta=log_every,
asynchronous=True,
serialize_fn=utils.fetch_devicearray,
steps_key='learner_steps')
sac_config = sac_fd_config.sac_config
lfd_config = sac_fd_config.lfd_config
sac_builder = sac.SACBuilder(sac_config, logger_fn=logger_fn)
lfd_builder = builder.LfdBuilder(sac_builder, lfd_iterator_fn,
lfd_config)
if evaluator_factories is None:
eval_policy_factory = (
lambda n: sac.apply_policy_and_sample(n, True))
evaluator_factories = [
distributed_layout.default_evaluator_factory(
environment_factory=environment_factory,
network_factory=network_factory,
policy_factory=eval_policy_factory,
log_to_bigtable=log_to_bigtable)
]
super().__init__(
seed=seed,
environment_factory=environment_factory,
network_factory=network_factory,
environment_spec=environment_spec,
builder=lfd_builder,
policy_network=sac.apply_policy_and_sample,
evaluator_factories=evaluator_factories,
num_actors=num_actors,
max_number_of_steps=max_number_of_steps,
prefetch_size=sac_config.prefetch_size,
log_to_bigtable=log_to_bigtable,
actor_logger_fn=distributed_layout.get_default_logger_fn(
log_to_bigtable, log_every))
def main(_):
task = FLAGS.task
environment_factory = lambda seed: helpers.make_environment(task)
sac_config = sac.SACConfig(num_sgd_steps_per_step=64)
sac_builder = sac.SACBuilder(sac_config)
ail_config = ail.AILConfig(direct_rl_batch_size=sac_config.batch_size *
sac_config.num_sgd_steps_per_step)
def network_factory(spec: specs.EnvironmentSpec) -> ail.AILNetworks:
def discriminator(*args, **kwargs) -> networks_lib.Logits:
return ail.DiscriminatorModule(environment_spec=spec,
use_action=True,
use_next_obs=True,
network_core=ail.DiscriminatorMLP(
[4, 4], ))(*args, **kwargs)
discriminator_transformed = hk.without_apply_rng(
hk.transform_with_state(discriminator))
return ail.AILNetworks(ail.make_discriminator(
spec, discriminator_transformed),
imitation_reward_fn=ail.rewards.gail_reward(),
direct_rl_networks=sac.make_networks(spec))
def policy_network(
network: ail.AILNetworks,
eval_mode: bool = False) -> actor_core_lib.FeedForwardPolicy:
return sac.apply_policy_and_sample(network.direct_rl_networks,
eval_mode=eval_mode)
program = ail.DistributedAIL(
environment_factory=environment_factory,
rl_agent=sac_builder,
config=ail_config,
network_factory=network_factory,
seed=0,
batch_size=sac_config.batch_size * sac_config.num_sgd_steps_per_step,
make_demonstrations=functools.partial(
helpers.make_demonstration_iterator,
dataset_name=FLAGS.dataset_name),
policy_network=policy_network,
evaluator_policy_network=(lambda n: policy_network(n, eval_mode=True)),
num_actors=4,
max_number_of_steps=100,
discriminator_loss=ail.losses.gail_loss()).build()
# Launch experiment.
lp.launch(program, xm_resources=lp_utils.make_xm_docker_resources(program))
def __init__(self,
spec: specs.EnvironmentSpec,
sac_network: sac.SACNetworks,
sac_fd_config: SACfDConfig,
lfd_iterator_fn: Callable[[], Iterator[builder.LfdStep]],
seed: int,
counter: Optional[counting.Counter] = None):
"""New instance of a SACfD agent."""
sac_config = sac_fd_config.sac_config
lfd_config = sac_fd_config.lfd_config
sac_builder = sac.SACBuilder(sac_config)
lfd_builder = builder.LfdBuilder(sac_builder, lfd_iterator_fn,
lfd_config)
min_replay_size = sac_config.min_replay_size
# Local layout (actually agent.Agent) makes sure that we populate the
# buffer with min_replay_size initial transitions and that there's no need
# for tolerance_rate. In order for deadlocks not to happen we need to
# disable rate limiting that heppens inside the SACBuilder. This is achieved
# by the following two lines.
sac_config.samples_per_insert_tolerance_rate = float('inf')
sac_config.min_replay_size = 1
self.builder = lfd_builder
super().__init__(
builder=lfd_builder,
seed=seed,
environment_spec=spec,
networks=sac_network,
policy_network=sac.apply_policy_and_sample(sac_network),
batch_size=sac_config.batch_size,
prefetch_size=sac_config.prefetch_size,
samples_per_insert=sac_config.samples_per_insert,
min_replay_size=min_replay_size,
num_sgd_steps_per_step=sac_config.num_sgd_steps_per_step,
counter=counter,
)
def __init__(self, sac_fd_config: SACfDConfig,
lfd_iterator_fn: Callable[[], Iterator[builder.LfdStep]]):
sac_builder = sac.SACBuilder(sac_fd_config.sac_config)
super().__init__(sac_builder, lfd_iterator_fn, sac_fd_config.lfd_config)
def main(_):
# Create an environment, grab the spec, and use it to create networks.
environment = helpers.make_environment(task=FLAGS.env_name)
environment_spec = specs.make_environment_spec(environment)
agent_networks = sac.make_networks(environment_spec)
# Construct the agent.
# Local layout makes sure that we populate the buffer with min_replay_size
# initial transitions and that there's no need for tolerance_rate. In order
# for deadlocks not to happen we need to disable rate limiting that heppens
# inside the SACBuilder. This is achieved by the min_replay_size and
# samples_per_insert_tolerance_rate arguments.
sac_config = sac.SACConfig(
target_entropy=sac.target_entropy_from_env_spec(environment_spec),
num_sgd_steps_per_step=FLAGS.num_sgd_steps_per_step,
min_replay_size=1,
samples_per_insert_tolerance_rate=float('inf'))
sac_builder = sac.SACBuilder(sac_config)
sac_networks = sac.make_networks(environment_spec)
sac_networks = add_bc_pretraining(sac_networks)
ail_config = ail.AILConfig(direct_rl_batch_size=sac_config.batch_size *
sac_config.num_sgd_steps_per_step)
def discriminator(*args, **kwargs) -> networks_lib.Logits:
return ail.DiscriminatorModule(environment_spec=environment_spec,
use_action=True,
use_next_obs=True,
network_core=ail.DiscriminatorMLP(
[4, 4], ))(*args, **kwargs)
discriminator_transformed = hk.without_apply_rng(
hk.transform_with_state(discriminator))
ail_network = ail.AILNetworks(
ail.make_discriminator(environment_spec, discriminator_transformed),
imitation_reward_fn=ail.rewards.gail_reward(),
direct_rl_networks=sac_networks)
agent = ail.AIL(spec=environment_spec,
rl_agent=sac_builder,
network=ail_network,
config=ail_config,
seed=FLAGS.seed,
batch_size=sac_config.batch_size *
sac_config.num_sgd_steps_per_step,
make_demonstrations=functools.partial(
helpers.make_demonstration_iterator,
dataset_name=FLAGS.dataset_name),
policy_network=sac.apply_policy_and_sample(sac_networks),
discriminator_loss=ail.losses.gail_loss())
# Create the environment loop used for training.
train_logger = experiment_utils.make_experiment_logger(
label='train', steps_key='train_steps')
train_loop = acme.EnvironmentLoop(environment,
agent,
counter=counting.Counter(prefix='train'),
logger=train_logger)
# Create the evaluation actor and loop.
eval_logger = experiment_utils.make_experiment_logger(
label='eval', steps_key='eval_steps')
eval_actor = agent.builder.make_actor(
random_key=jax.random.PRNGKey(FLAGS.seed),
policy_network=sac.apply_policy_and_sample(agent_networks,
eval_mode=True),
variable_source=agent)
eval_env = helpers.make_environment(task=FLAGS.env_name)
eval_loop = acme.EnvironmentLoop(eval_env,
eval_actor,
counter=counting.Counter(prefix='eval'),
logger=eval_logger)
assert FLAGS.num_steps % FLAGS.eval_every == 0
for _ in range(FLAGS.num_steps // FLAGS.eval_every):
eval_loop.run(num_episodes=5)
train_loop.run(num_steps=FLAGS.eval_every)
eval_loop.run(num_episodes=5)
def test_ail_flax(self):
shutil.rmtree(flags.FLAGS.test_tmpdir)
batch_size = 8
# Mujoco environment and associated demonstration dataset.
environment = fakes.ContinuousEnvironment(
episode_length=EPISODE_LENGTH,
action_dim=CONTINUOUS_ACTION_DIM,
observation_dim=CONTINUOUS_OBS_DIM,
bounded=True)
spec = specs.make_environment_spec(environment)
networks = sac.make_networks(spec=spec)
config = sac.SACConfig(batch_size=batch_size,
samples_per_insert_tolerance_rate=float('inf'),
min_replay_size=1)
base_builder = sac.SACBuilder(config=config)
direct_rl_batch_size = batch_size
behavior_policy = sac.apply_policy_and_sample(networks)
discriminator_module = DiscriminatorModule(spec, linen.Dense(1))
def apply_fn(params: networks_lib.Params,
policy_params: networks_lib.Params,
state: networks_lib.Params, transitions: types.Transition,
is_training: bool,
rng: networks_lib.PRNGKey) -> networks_lib.Logits:
del policy_params
variables = dict(params=params, **state)
return discriminator_module.apply(variables,
transitions.observation,
transitions.action,
transitions.next_observation,
is_training=is_training,
rng=rng,
mutable=state.keys())
def init_fn(rng):
variables = discriminator_module.init(rng,
dummy_obs,
dummy_actions,
dummy_obs,
is_training=False,
rng=rng)
init_state, discriminator_params = variables.pop('params')
return discriminator_params, init_state
dummy_obs = utils.zeros_like(spec.observations)
dummy_obs = utils.add_batch_dim(dummy_obs)
dummy_actions = utils.zeros_like(spec.actions)
dummy_actions = utils.add_batch_dim(dummy_actions)
discriminator_network = networks_lib.FeedForwardNetwork(init=init_fn,
apply=apply_fn)
networks = ail.AILNetworks(discriminator_network, lambda x: x,
networks)
builder = ail.AILBuilder(
base_builder,
config=ail.AILConfig(is_sequence_based=False,
share_iterator=True,
direct_rl_batch_size=direct_rl_batch_size,
discriminator_batch_size=2,
policy_variable_name=None,
min_replay_size=1),
discriminator_loss=ail.losses.gail_loss(),
make_demonstrations=fakes.transition_iterator(environment))
counter = counting.Counter()
# Construct the agent.
agent = local_layout.LocalLayout(
seed=0,
environment_spec=spec,
builder=builder,
networks=networks,
policy_network=behavior_policy,
min_replay_size=1,
batch_size=batch_size,
counter=counter,
)
# Train the agent.
train_loop = acme.EnvironmentLoop(environment, agent, counter=counter)
train_loop.run(num_episodes=1)
def test_ail(self,
algo,
airl_discriminator=False,
subtract_logpi=False,
dropout=0.,
lipschitz_coeff=None):
shutil.rmtree(flags.FLAGS.test_tmpdir, ignore_errors=True)
batch_size = 8
# Mujoco environment and associated demonstration dataset.
if algo == 'ppo':
environment = fakes.DiscreteEnvironment(
num_actions=NUM_DISCRETE_ACTIONS,
num_observations=NUM_OBSERVATIONS,
obs_shape=OBS_SHAPE,
obs_dtype=OBS_DTYPE,
episode_length=EPISODE_LENGTH)
else:
environment = fakes.ContinuousEnvironment(
episode_length=EPISODE_LENGTH,
action_dim=CONTINUOUS_ACTION_DIM,
observation_dim=CONTINUOUS_OBS_DIM,
bounded=True)
spec = specs.make_environment_spec(environment)
if algo == 'sac':
networks = sac.make_networks(spec=spec)
config = sac.SACConfig(
batch_size=batch_size,
samples_per_insert_tolerance_rate=float('inf'),
min_replay_size=1)
base_builder = sac.SACBuilder(config=config)
direct_rl_batch_size = batch_size
behavior_policy = sac.apply_policy_and_sample(networks)
elif algo == 'ppo':
unroll_length = 5
distribution_value_networks = make_ppo_networks(spec)
networks = ppo.make_ppo_networks(distribution_value_networks)
config = ppo.PPOConfig(unroll_length=unroll_length,
num_minibatches=2,
num_epochs=4,
batch_size=batch_size)
base_builder = ppo.PPOBuilder(config=config)
direct_rl_batch_size = batch_size * unroll_length
behavior_policy = jax.jit(ppo.make_inference_fn(networks),
backend='cpu')
else:
raise ValueError(f'Unexpected algorithm {algo}')
if subtract_logpi:
assert algo == 'sac'
logpi_fn = make_sac_logpi(networks)
else:
logpi_fn = None
if algo == 'ppo':
embedding = lambda x: jnp.reshape(x, list(x.shape[:-2]) + [-1])
else:
embedding = lambda x: x
def discriminator(*args, **kwargs) -> networks_lib.Logits:
if airl_discriminator:
return ail.AIRLModule(
environment_spec=spec,
use_action=True,
use_next_obs=True,
discount=.99,
g_core=ail.DiscriminatorMLP(
[4, 4],
hidden_dropout_rate=dropout,
spectral_normalization_lipschitz_coeff=lipschitz_coeff
),
h_core=ail.DiscriminatorMLP(
[4, 4],
hidden_dropout_rate=dropout,
spectral_normalization_lipschitz_coeff=lipschitz_coeff
),
observation_embedding=embedding)(*args, **kwargs)
else:
return ail.DiscriminatorModule(
environment_spec=spec,
use_action=True,
use_next_obs=True,
network_core=ail.DiscriminatorMLP(
[4, 4],
hidden_dropout_rate=dropout,
spectral_normalization_lipschitz_coeff=lipschitz_coeff
),
observation_embedding=embedding)(*args, **kwargs)
discriminator_transformed = hk.without_apply_rng(
hk.transform_with_state(discriminator))
discriminator_network = ail.make_discriminator(
environment_spec=spec,
discriminator_transformed=discriminator_transformed,
logpi_fn=logpi_fn)
networks = ail.AILNetworks(discriminator_network, lambda x: x,
networks)
builder = ail.AILBuilder(
base_builder,
config=ail.AILConfig(
is_sequence_based=(algo == 'ppo'),
share_iterator=True,
direct_rl_batch_size=direct_rl_batch_size,
discriminator_batch_size=2,
policy_variable_name='policy' if subtract_logpi else None,
min_replay_size=1),
discriminator_loss=ail.losses.gail_loss(),
make_demonstrations=fakes.transition_iterator(environment))
# Construct the agent.
agent = local_layout.LocalLayout(seed=0,
environment_spec=spec,
builder=builder,
networks=networks,
policy_network=behavior_policy,
min_replay_size=1,
batch_size=batch_size)
# Train the agent.
train_loop = acme.EnvironmentLoop(environment, agent)
train_loop.run(num_episodes=(10 if algo == 'ppo' else 1))