torch.distributions

Sampling

The sample method generates samples from a distribution, and in this process, it does not consider the gradient of the parameters. This is used when differentiation through the sampling process is not necessary. On the other hand, rsample refers to reparameterization sampling, which enables gradient-based optimization through the sampling process (loss calculation Pathwise derivative).

torch.distributions members

torch.distributions.Categorical

torch.distributions.Normal

torch.distributions functions

torch.distribution log_prob()

torch.distribution.sample()

torch.distribution.ssample()

Probability distributions - torch.distributions — PyTorch 2.3 documentation

The distributions package contains parameterizable probability distributions and sampling functions. This allows the construction of stochastic computation graphs and stochastic gradient estimators for optimization. This package generally follows the design of the TensorFlow Distributions package.