Torchflows documentation

Torchflows is a library for generative modeling and density estimation using normalizing flows. It implements many normalizing flow architectures and their building blocks for:

• easy use of normalizing flows as trainable distributions;

• easy implementation of new normalizing flows.

Torchflows is structured according to the review paper Normalizing Flows for Probabilistic Modeling and Inference by Papamakarios et al. (2021), which classifies flow architectures as autoregressive, residual, or continuous. Visit the Github page to keep up to date and post any questions or issues here.

Installing

Torchflows can be installed easily using pip:

pip install torchflows

For other install options, see the install section.

Table of contents

• Installing
• Tutorial
  • Mathematical background
  • Basic usage
  • Event shapes
  • Image modeling
  • Choosing a base distribution
  • Using CUDA
  • Modifying normalizing flow architectures
• Full list of architectures (presets)
  • Autoregressive architectures
  • Multiscale architectures
  • Residual architectures
  • Continuous architectures
• API for standard architectures
  • Autoregressive architectures
  • Continuous architectures
  • Residual architectures
• API for multiscale architectures
  • Classic multiscale architectures
  • Glow-style multiscale architectures
• Developer reference
  • Base distributions
  • Bijections
  • Flow wrappers