Module: tfp.experimental

TensorFlow Probability API-unstable package.

This package contains potentially useful code which is under active development with the intention eventually migrate to TFP proper. All code in tfp.experimental should be of production quality, i.e., idiomatically consistent, well tested, and extensively documented. tfp.experimental code relaxes the TFP non-experimental contract in two regards:

  1. tfp.experimental has no API stability guarantee. The public footprint of tfp.experimental code may change without notice or warning.
  2. Code outside tfp.experimental cannot depend on code within tfp.experimental.

You are welcome to try any of this out (and tell us how well it works for you!).

Modules

auto_batching module: TensorFlow Probability auto-batching package.

edward2 module: Edward2 probabilistic programming language.

inference_gym module: Inference Gym package.

linalg module: Experimental tools for linear algebra.

marginalize module: Marginalizable probability distributions.

mcmc module: TensorFlow Probability experimental NUTS package.

nn module: Tools for building neural networks.

sequential module: TensorFlow Probability experimental sequential estimation package.

substrates module: TensorFlow Probability alternative substrates.

vi module: Experimental methods and objectives for variational inference.

Functions

as_composite(...): Returns a CompositeTensor equivalent to the given object.

register_composite(...): A decorator that registers a Distribution as composite-friendly.