reference/modules_distributions

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

Distribution Classes

Custom distribution classes for RL, extending PyTorch distributions.

Delta(param[, atol, rtol, batch_shape, ...])	Delta distribution.
IndependentNormal(loc, scale[, upscale, ...])	Implements a Normal distribution with location scaling.
MaskedCategorical([logits, probs, mask, ...])	MaskedCategorical distribution.
NormalParamExtractor([scale_mapping, scale_lb])	A non-parametric nn.Module that splits its input into loc and scale parameters.
OneHotCategorical([logits, probs, grad_method])	One-hot categorical distribution.
ReparamGradientStrategy(value[, names, ...])
TanhDelta(param[, low, high, event_dims, ...])	Implements a Tanh transformed_in Delta distribution.
TanhNormal(loc, scale[, upscale, low, high, ...])	Implements a TanhNormal distribution with location scaling.
TruncatedNormal(loc, scale[, upscale, low, ...])	Implements a Truncated Normal distribution with location scaling.

Sampling utilities

sample_and_log_prob(distribution[, ...])	Sample once and score the same draw atomically when supported.
rsample_and_log_prob(distribution[, ...])	Reparameterize once and score the same draw atomically when supported.
composite_entropy(distribution[, samples_mc])	Compute component entropy without inverse-scoring Monte Carlo samples.