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

(overrides ProbabilityDistribution)

implementation of ProbabilityDistribution interface

kl(other: maze.distributions.torch_dist.TorchProbabilityDistribution) → torch.Tensor

(overrides ProbabilityDistribution)

implementation of ProbabilityDistribution interface

log_prob(actions: torch.Tensor) → torch.Tensor

(overrides ProbabilityDistribution)

implementation of ProbabilityDistribution interface

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

(overrides ProbabilityDistribution)

implementation of ProbabilityDistribution interface