Package flow
This project implements basic Normalizing Flows in PyTorch and provides functionality for defining your own easily, following the conditioner-transformer architecture.
This is specially useful for lower-dimensional flows and for learning purposes. Nevertheless, work is being done on extending its functionalities to also accomodate for higher dimensional flows.
Supports conditioning flows, meaning, learning probability distributions conditioned by a given conditioning tensor. Specially useful for modelling causal mechanisms.
Expand source code
'''
.. include::documentation.md
'''
from .flow import Flow, Sequential, Transformer, Conditioner, inv_flowSub-modules
flow.conditioner-
Implementations for Flow-Conditioners …
flow.flow-
Abstract classes for the implementation of Flows …
flow.modules-
Miscellaneous Flows.
flow.prior-
Abstract class for U priors and implementations for common-use priors …
flow.training-
Train utilities for flows …
flow.transformer-
Implementations for Flow-transformers …
flow.utils-
Miscellaneous utility functions.