Constrained Matrix Factorization Documentation¶
Warning
This is currently under rapid development, the API may change at any time. We suggest paying careful attention to the version number you are using.
Constrained Matrix Factorization (CMF) comes as an advancement on Non-negative Matrix Factorization (NMF). Initially called constrained non-negative matrix factorization, it was recognized that this was redundant, as the non-negativity is already a constraint. The goal for this package is to produce rapid matrix factorization approaches with effective constraints as related to beamline science. We use PyTorch as a backend to enable GPU acceleration and provide constraints via gradient management.
Total scattering (x-ray diffraction and pair distribution function) analysis, was used as a primary example with rigid constraints in Applied Physics Reviews (arXiv version).
Please feel free to file bug reports and feature requests *via* GitHub Issues!