Kernel Stein Discrepancy Descent

|GHActions|_ |PyPI|_

.. |GHActions| image:: https://github.com/pierreablin/ksddescent/workflows/unittests/badge.svg?branch=main&event=push .. _GHActions: https://github.com/pierreablin/ksddescent/actions

.. |PyPI| image:: https://badge.fury.io/py/ksddescent.svg .. _PyPI: https://badge.fury.io/py/ksddescent

Sampling by optimization of the Kernel Stein Discrepancy

The paper is available at arxiv.org/abs/2105.09994 <https://arxiv.org/abs/2105.09994>_.

The code uses Pytorch, and a numpy backend is available for svgd.

.. image:: https://pierreablin.github.io/figures/ksd_descent_small.png :width: 100 :alt: ksd_picture

Install

The code is available on pip::

$ pip install ksddescent

Documentation

The documentation is at pierreablin.github.io/ksddescent/ <https://pierreablin.github.io/ksddescent/>_.

Example

The main function is ksdd_lbfgs, which uses the fast L-BFGS algorithm to converge quickly. It takes as input the initial position of the particles, and the score function. For instance, to samples from a Gaussian (where the score is identity), you can use these simple lines of code:

.. code:: python

import torch from ksddescent import ksdd_lbfgs n, p = 50, 2 x0 = torch.rand(n, p) # start from uniform distribution score = lambda x: x # simple score function x = ksdd_lbfgs(x0, score) # run the algorithm

Reference

If you use this code in your project, please cite::

Anna Korba, Pierre-Cyril Aubin-Frankowski, Simon Majewski, Pierre Ablin
Kernel Stein Discrepancy Descent
International Conference on Machine Learning, 2021

Bug reports

Use the github issue tracker <https://github.com/pierreablin/ksddescent/issues>_ to report bugs.