Note: This repository is not maintained anymore. Please check the new repository of python package MALSpy.
This repository provides MATLAB and Python codes of our proposed methods in [1].
In MATLAB, you can run a demo script for NMF-SO (Nonnegative Matrix Factorization with Soft Orthogonality constraint):
demo_nmf_so
and NMF-ARD-SO (Nonnegative Matrix Factorization with Automatic Relevance Determination and Soft Orthogonality constraint):
demo_nmf_ard_so
SO is for resolving spatial overlaps among chemical components and ARD is for optimizing the number of chemical components.
Our python library code (supported on Python 3.5.1+) was updated on July 10, 2017. The new code defines a class for each NMF model and use method fit to learn, similarly to scikit-learn. See jupyter notebook demo_libnmf.ipynb.
[1]
Motoki Shiga, Kazuyoshi Tatsumi, Shunsuke Muto, Koji Tsuda, Yuta Yamamoto, Toshiyuki Mori, Takayoshi Tanji,
"Sparse Modeling of EELS and EDX Spectral Imaging Data by Nonnegative Matrix Factorization",
Ultramicroscopy, Vol.170, p.43-59, 2016.
http://dx.doi.org/10.1016/j.ultramic.2016.08.006
MIT License (see LICENSE file).