maksimt / rri_nmf

Non-negative Matrix Factorization using Rank-One Residue Iterations implemented in Python

rri_nmf

Non-negative Matrix Factorization using Rank-One Residue Iterations implemented in Python

Ngoc Diep Ho's Thesis

CJ Lin's Paper

sklearn NMF

TODOs

RRI NMF Base algorithm

NMF for Topic Modeling

1. Add a score method to the estimator that uses a bunch of scores.

NMF for Recommender Systems

1. Initialize using elementwise weighted (masked) SVD. E.g. BIRSVD
2. Cythonized implementation of the elementwise division in the gradient step loop. (Algorithm 10, WRRI in Ho's Thesis)