Skip to content

Efficient implementation of forward and backpropagation on angle parametrized orthogonal matrices

License

Notifications You must be signed in to change notification settings

microsoft/Givens-Orthogonal-Backprop

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Givens Rotation Parametrized Orthogonal Backpropagation

Code for: https://arxiv.org/abs/2106.00003 Allows for gradient descent over rotational matrices.

(Please note that the cpp extension is not optimized and is intended as a POC.)

Installation

PLease use python 3.8. You can replicate our development environment via conda, simply run:

conda env create -f environment.yml

conda activate rotMatDev

To setup the pytorch extension, go to the source folder ('cpp/' for serial execution on cpu, 'cuda/' for parallel execution on gpu) and run

python setup.py install

An example NN layer is included in the GivensRotations.py files in the source folders.

How to file issues and get help

This project uses GitHub Issues to track bugs and feature requests. Please search the existing issues before filing new issues to avoid duplicates. For new issues, file your bug or feature request as a new Issue. We'll try to address impactful issues. Please feel free to reach out if you have any questions!

About

Efficient implementation of forward and backpropagation on angle parametrized orthogonal matrices

Resources

License

Code of conduct

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published