Timely! We just added a super-efficient implementation of Online Non-negative Matrix Factorization to Gensim. Jupyter notebook tutorial: https://github.com/Ra

Timely! We just added a super-efficient implementation of Online Non-negative Matrix Factorization to Gensim. Jupyter notebook tutorial: https://github.com/RaRe-Technologies/gensim/blob/nmf_speedups/docs/notebooks/nmf_tutorial.ipynb https://github.com/RaRe-Technologies/gensim/blob/nmf_speedup... What's unique about this implementation is that it's streamed, i.e. can process arbitrary amounts of data in constant RAM. It's also heavily optimized for sparse inputs (~text corpora, not just dense images), just like the optimized sparse SVD, LDA, FastText etc in Gensim.