SquashedGaussianProbabilityDistribution

class maze.distributions.squashed_gaussian.SquashedGaussianProbabilityDistribution(*args, **kwds)

Tanh-squashed diagonal Gaussian (Normal) Torch probability distribution.

Parameters

• logits – the logits for both mean and standard deviation.

• action_space – the underlying gym.spaces action space.

deterministic_sample()

(overrides TorchProbabilityDistribution)

implementation of TorchProbabilityDistribution interface

entropy() → torch.Tensor

(overrides TorchProbabilityDistribution)

implementation of TorchProbabilityDistribution interface

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

(overrides TorchProbabilityDistribution)

implementation of TorchProbabilityDistribution interface

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

(overrides TorchProbabilityDistribution)

implementation of TorchProbabilityDistribution interface

classmethod required_logits_shape(action_space: gym.spaces.Space) → Sequence[int]

(overrides TorchProbabilityDistribution)

implementation of TorchProbabilityDistribution interface

sample()

(overrides TorchProbabilityDistribution)

implementation of TorchProbabilityDistribution interface