Looking for some software to deal with 50kx50k sparse matrix applying non-negative matrix factorization. Do you know any?
-
$\begingroup$ Third hit in google leads me here: cs.helsinki.fi/u/phoyer/software.html $\endgroup$ – Terry Loring Sep 2 '13 at 16:11
-
2$\begingroup$ It would help if you were a bit more verbose on what exactly you mean with "non-negative", and what kinds of factorizations you are looking for. What do you happen to know about your matrix? $\endgroup$ – Wolfgang Bangerth Sep 3 '13 at 3:02
-
1$\begingroup$ non-negative matrix factorization is a thing $\endgroup$ – k20 Sep 3 '13 at 21:32
-
$\begingroup$ @k20, several terms are "a thing", but it would not hurt to be descriptive. $\endgroup$ – nicoguaro♦ Mar 13 '18 at 13:19
$\begingroup$
$\endgroup$
The following paper uses MapReduce for distributed computation of non-negative matrix factorization:
- J. Yin, L. Gao, Zh. Zhang, "Scalable Nonnegative Matrix Factorization with Block-wise Updates," in Proc. Eu. Conf. Machine Learning Knowledge Discovery, Nancy, France, Sep. 2014, pp.337-352.
This paper describes an implementation and parallelization of the non-negative factorization and claims the results for matrices exceeding your specifications.
Regarding software libraries:
- python library NIMFA supports both dense and sparse input (associated paper)
- R-package NMF (no info on sparse that I was able to find quickly)
- NMF Matrix toolbox (v 1.4 added support for sparse input)
- high-performance library SmallK, supports both dense and sparse