rJKOtt

The project aims at modelling probability distributions with use of entropy-regularized JKO scheme with tensor-train decomposition.

The goal is, given pointwise access to the unnormalized density of a complicated probability distribution in high dimension, to fit a compressed model which will allow to sample from it and solve some downstream tasks.

Installation

pip install --upgrade rJKOtt@git+https://github.com/viviaxenov/rJKOtt

Requirements:

• teneva (https://github.com/AndreiChertkov/teneva)
• TBD

Optional requirements

• POT (https://github.com/PythonOT/POT)
• geomloss (https://github.com/jeanfeydy/geomloss)
• TBD

Documentation, examples, etc

• Documentation can be found here.
• An example with explanation of basic theory
• Tests – TBD

Citation

TBD one day i hope...