def __init__(self, env_spec, name="CategoricalMLPPolicy", **kwargs):
self._obs_dim = env_spec.input_space.flat_dim
self._action_dim = env_spec.output_space.flat_dim
Policy.__init__(self, env_spec, name)
MLPModule.__init__(self, input_dim=self._obs_dim,
output_dim=self._action_dim,
**kwargs)
def __init__(self, env_spec, **kwargs):
self._obs_dim = env_spec.observation_space.flat_dim
self._action_dim = env_spec.action_space.flat_dim
Policy.__init__(self, env_spec)
GaussianMLPModule.__init__(self,
input_dim=self._obs_dim,
output_dim=self._action_dim,
**kwargs)
def __init__(self, env_spec, **kwargs):
assert isinstance(env_spec.action_space, akro.Discrete)
self._env_spec = env_spec
self._obs_dim = env_spec.observation_space.flat_dim
self._action_dim = env_spec.action_space.flat_dim
Policy.__init__(self, env_spec)
CategoricalMLPModule.__init__(self,
input_dim=self._obs_dim,
output_dim=self._action_dim,
**kwargs)
def __init__(self, env_spec, name='DeterministicMLPPolicy', **kwargs):
"""Initialize class with multiple attributes.
Args:
env_spec (garage.envs.env_spec.EnvSpec): Environment specification.
name (str): Policy name.
kwargs : Additional keyword arguments passed to the MLPModule.
"""
self._obs_dim = env_spec.observation_space.flat_dim
self._action_dim = env_spec.action_space.flat_dim
Policy.__init__(self, env_spec, name)
MLPModule.__init__(self,
input_dim=self._obs_dim,
output_dim=self._action_dim,
**kwargs)
def __init__(self,
env_spec,
hidden_sizes=(32, 32),
hidden_nonlinearity=torch.tanh,
hidden_w_init=nn.init.xavier_uniform_,
hidden_b_init=nn.init.zeros_,
output_nonlinearity=None,
output_w_init=nn.init.xavier_uniform_,
output_b_init=nn.init.zeros_,
learn_std=True,
init_std=1.0,
min_std=1e-6,
max_std=None,
std_parameterization='exp',
layer_normalization=False,
name='GaussianMLPPolicy'):
self._obs_dim = env_spec.observation_space.flat_dim
self._action_dim = env_spec.action_space.flat_dim
Policy.__init__(self, env_spec, name)
GaussianMLPModule.__init__(self,
input_dim=self._obs_dim,
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,
init_std=init_std,
min_std=min_std,
max_std=max_std,
std_parameterization=std_parameterization,
layer_normalization=layer_normalization)
def __init__(self,
env_spec,
n_agents,
hidden_sizes=(32, 32),
hidden_nonlinearity=torch.tanh,
hidden_w_init=nn.init.xavier_uniform_,
hidden_b_init=nn.init.zeros_,
output_nonlinearity=None,
output_w_init=nn.init.xavier_uniform_,
output_b_init=nn.init.zeros_,
layer_normalization=False,
name='DecCategoricalMLPPolicy'):
assert isinstance(env_spec.action_space, akro.Discrete), (
'CategoricalMLPPolicy only works with akro.Discrete action '
'space.')
self.centralized = True # centralized training
self._n_agents = n_agents
self._obs_dim = int(env_spec.observation_space.flat_dim /
n_agents) # dec obs_dim
self._action_dim = env_spec.action_space.n
Policy.__init__(self, env_spec, name)
CategoricalMLPModule.__init__(self,
input_dim=self._obs_dim,
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,
layer_normalization=layer_normalization)
def __init__(self,
env_spec,
hidden_sizes=(32, 32),
hidden_nonlinearity=nn.ReLU,
hidden_w_init=nn.init.xavier_uniform_,
hidden_b_init=nn.init.zeros_,
output_nonlinearity=None,
output_w_init=nn.init.xavier_uniform_,
output_b_init=nn.init.zeros_,
init_std=1.0,
min_std=np.exp(-20.),
max_std=np.exp(2.),
std_parameterization='exp',
layer_normalization=False):
self._obs_dim = env_spec.observation_space.flat_dim
self._action_dim = env_spec.action_space.flat_dim
Policy.__init__(self, env_spec, name='TanhGaussianPolicy')
GaussianMLPTwoHeadedModule.__init__(
self,
input_dim=self._obs_dim,
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,
init_std=init_std,
min_std=min_std,
max_std=max_std,
std_parameterization=std_parameterization,
layer_normalization=layer_normalization,
normal_distribution_cls=TanhNormal)
def __init__(self, env_spec, module):
nn.Module.__init__(self)
Policy.__init__(self, env_spec)
self._module = module