def __init__(self,
state_shape,
action_dim,
is_discrete,
actor=None,
critic=None,
actor_critic=None,
max_action=1.,
actor_units=[256, 256],
critic_units=[256, 256],
lr_actor=1e-3,
lr_critic=3e-3,
fix_std=False,
const_std=0.3,
hidden_activation_actor="relu",
hidden_activation_critic="relu",
name="VPG",
**kwargs):
super().__init__(name=name, **kwargs)
self._is_discrete = is_discrete
# TODO: clean codes
if actor_critic is not None:
self.actor_critic = actor_critic
self.actor_critic_optimizer = tf.keras.optimizers.Adam(
learning_rate=lr_actor)
self.actor = None
self.critic = None
else:
self.actor_critic = None
if actor is None:
if is_discrete:
self.actor = CategoricalActor(state_shape, action_dim,
actor_units)
else:
self.actor = GaussianActor(
state_shape,
action_dim,
max_action,
actor_units,
hidden_activation=hidden_activation_actor,
fix_std=fix_std,
const_std=const_std,
state_independent_std=True)
else:
self.actor = actor
if critic is None:
self.critic = CriticV(
state_shape,
critic_units,
hidden_activation=hidden_activation_critic)
else:
self.critic = critic
self.actor_optimizer = tf.keras.optimizers.Adam(
learning_rate=lr_actor)
self.critic_optimizer = tf.keras.optimizers.Adam(
learning_rate=lr_critic)
# This is used to check if input state to `get_action` is multiple (batch) or single
self._state_ndim = np.array(state_shape).shape[0]
def __init__(self,
state_shape,
action_dim,
is_discrete,
max_action=1.,
actor_units=[256, 256],
critic_units=[256, 256],
lr_actor=1e-3,
lr_critic=3e-3,
fix_std=False,
tanh_std=False,
const_std=0.3,
name="VPG",
**kwargs):
super().__init__(name=name, **kwargs)
self._is_discrete = is_discrete
if is_discrete:
self.actor = CategoricalActor(state_shape, action_dim, actor_units)
else:
self.actor = GaussianActor(state_shape,
action_dim,
max_action,
actor_units,
fix_std=fix_std,
tanh_std=tanh_std,
const_std=const_std)
self.critic = CriticV(state_shape, critic_units)
self._action_dim = action_dim
self.actor_optimizer = tf.keras.optimizers.Adam(learning_rate=lr_actor)
self.critic_optimizer = tf.keras.optimizers.Adam(
learning_rate=lr_critic)
def setUpClass(cls):
super().setUpClass()
cls.policy = CategoricalActor(
state_shape=cls.discrete_env.observation_space.shape,
action_dim=cls.discrete_env.action_space.n,
units=[4, 4])