Funsor is a tensor-like library for functions and distributions¶

Domains¶

Operations¶

Interpretations¶

Interpreter¶

Monte Carlo¶

Memoize¶

Funsors¶

Basic Funsors¶

Delta¶

PyTorch¶

JAX¶

Gaussian¶

Joint¶

Contraction¶

Integrate¶

Optimizer¶

Adjoint Algorithms¶

Sum-Product Algorithms¶

Affine Pattern Matching¶

Testing Utiltites¶

Pyro-Compatible Distributions¶

This interface provides a number of PyTorch-style distributions that use funsors internally to perform inference. These high-level objects are based on a wrapping class: FunsorDistribution which wraps a funsor in a PyTorch-distributions-compatible interface. FunsorDistribution objects can be used directly in Pyro models (using the standard Pyro backend).

FunsorDistribution Base Class¶

Hidden Markov Models¶

Conversion Utilities¶

Distribution Funsors¶

This interface provides a number of standard normalized probability distributions implemented as funsors.

Mini-Pyro Interface¶

This interface provides a backend for the Pyro probabilistic programming language. This interface is intended to be used indirectly by writing standard Pyro code and setting pyro_backend("funsor"). See examples/minipyro.py for example usage.

Einsum Interface¶

This interface implements tensor variable elimination among tensors. In particular it does not implement continuous variable elimination.

Indices and tables¶