List of supported types#

Base types#

Functions#

A function is termed “pure” if its output depends only on its inputs. If you serialize a function as part of a reproducible workflow, this is generally what you want.

NumPy types#

SciPy distributions#

Univariate distributions in scipy.stats are implemented in a generic manner, which allows us to support all of them with the same type annotation UniDistribution. This is different for multivariate, which each require their own type.

MvDistribution is an abstract type for all multivariate distributions, and Distribution an abstract type for all univariate and multivariate ones.

Units#

Pint#

Quantities#

PyTorch#