def __init__(self, name, env_spec, encoder):
super().__init__(name, env_spec)
self._encoder = encoder
self._augmented_observation_space = akro.concat(
self._env_spec.observation_space, self.task_space)
def __init__(self,
env_spec,
encoder,
name='GaussianMLPTaskEmbeddingPolicy',
hidden_sizes=(32, 32),
hidden_nonlinearity=tf.nn.tanh,
hidden_w_init=tf.initializers.glorot_uniform(
seed=deterministic.get_tf_seed_stream()),
hidden_b_init=tf.zeros_initializer(),
output_nonlinearity=None,
output_w_init=tf.initializers.glorot_uniform(
seed=deterministic.get_tf_seed_stream()),
output_b_init=tf.zeros_initializer(),
learn_std=True,
adaptive_std=False,
std_share_network=False,
init_std=1.0,
min_std=1e-6,
max_std=None,
std_hidden_sizes=(32, 32),
std_hidden_nonlinearity=tf.nn.tanh,
std_output_nonlinearity=None,
std_parameterization='exp',
layer_normalization=False):
assert isinstance(env_spec.action_space, akro.Box)
assert not isinstance(env_spec.observation_space, akro.Dict)
self._env_spec = env_spec
self._name = name
self._encoder = encoder
self._augmented_observation_space = akro.concat(
self._env_spec.observation_space, self.task_space)
self._hidden_sizes = hidden_sizes
self._hidden_nonlinearity = hidden_nonlinearity
self._hidden_w_init = hidden_w_init
self._hidden_b_init = hidden_b_init
self._output_nonlinearity = output_nonlinearity
self._output_w_init = output_w_init
self._output_b_init = output_b_init
self._learn_std = learn_std
self._adaptive_std = adaptive_std
self._std_share_network = std_share_network
self._init_std = init_std
self._min_std = min_std
self._max_std = max_std
self._std_hidden_sizes = std_hidden_sizes
self._std_hidden_nonlinearity = std_hidden_nonlinearity
self._std_output_nonlinearity = std_output_nonlinearity
self._std_parameterization = std_parameterization
self._layer_normalization = layer_normalization
self.obs_dim = env_spec.observation_space.flat_dim
self.action_dim = env_spec.action_space.flat_dim
super().__init__(output_dim=self.action_dim,
hidden_sizes=hidden_sizes,
hidden_nonlinearity=hidden_nonlinearity,
hidden_w_init=hidden_w_init,
hidden_b_init=hidden_b_init,
output_nonlinearity=output_nonlinearity,
output_w_init=output_w_init,
output_b_init=output_b_init,
learn_std=learn_std,
adaptive_std=adaptive_std,
std_share_network=std_share_network,
init_std=init_std,
min_std=min_std,
max_std=max_std,
std_hidden_sizes=std_hidden_sizes,
std_hidden_nonlinearity=std_hidden_nonlinearity,
std_output_nonlinearity=std_output_nonlinearity,
std_parameterization=std_parameterization,
layer_normalization=layer_normalization,
name=name)
self._initialize()