"""
This modules creates a deterministic policy network.
A neural network can be used as policy method in different RL algorithms.
It accepts an observation of the environment and predicts an action.
"""
import torch
from garage.torch.modules import MLPModule
from garage.torch.policies import Policy
[docs]class DeterministicMLPPolicy(MLPModule, Policy):
"""
Implements a deterministic policy network.
The policy network selects action based on the state of the environment.
It uses a PyTorch neural network module to fit the function of pi(s).
"""
def __init__(self, env_spec, **kwargs):
"""
Initialize class with multiple attributes.
Args:
env_spec (garage.envs.env_spec.EnvSpec): Environment specification.
nn_module (nn.Module): Neural network module in PyTorch.
"""
self._env_spec = env_spec
self._obs_dim = env_spec.observation_space.flat_dim
self._action_dim = env_spec.action_space.flat_dim
MLPModule.__init__(self,
input_dim=self._obs_dim,
output_dim=self._action_dim,
**kwargs)
[docs] def forward(self, input_val):
"""Forward method."""
input_val = torch.FloatTensor(input_val)
return super().forward(input_val)
[docs] def get_action(self, observation):
"""Return a single action."""
with torch.no_grad():
x = self.forward(observation.unsqueeze(0))
return x.squeeze(0).numpy(), dict()
[docs] def get_actions(self, observations):
"""Return multiple actions."""
with torch.no_grad():
x = self.forward(observations)
return x.numpy(), dict()
[docs] def reset(self, dones=None):
"""Reset policy."""
pass