TorchProbabilityDistribution¶
-
class
maze.distributions.torch_dist.TorchProbabilityDistribution(*args, **kwds)¶ Base class for wrapping Torch probability distributions.
- Parameters
dist – The torch probability distribution.
action_space – The gym action space.
-
entropy() → torch.Tensor¶ implementation of
ProbabilityDistributioninterface
-
kl(other: maze.distributions.torch_dist.TorchProbabilityDistribution) → torch.Tensor¶ implementation of
ProbabilityDistributioninterface
-
log_prob(actions: torch.Tensor) → torch.Tensor¶ implementation of
ProbabilityDistributioninterface
-
abstract classmethod
required_logits_shape(action_space: gym.spaces.Space) → Sequence[int]¶ Returns the required shape for the corresponding neural network logits output.
- Parameters
action_space – The respective action space to compute logits for.
- Returns
The required logits shape.
-
sample() → torch.Tensor¶ implementation of
ProbabilityDistributioninterface