This package includes Python implementations (with Numpy and Scipy) of numerical algorithms for computing nonnegative matrix factorization.
Numpy (http://www.numpy.org) and Scipy (http://www.scipy.org) need to be installed. Versions of Numpy and Scipy tested with this code were 1.6.1 and 0.9.0, respectively.
Use setup.py to install this package:
sudo python setup.py installWhen A is a dense (numpy.array) or a sparse (scipy.sparse) matrix, the following code returns W and H as factor matrices of A with 10 as the lower rank.
from nonnegfac.nmf import NMF
W, H, info = NMF().run(A, 10)Try to execute example.py to see simple usage. Function run() executes
an NMF algorithm once, and Function run_repeat() executes an NMF algorithm
for the specified number of times and returns the best result based on the
norm of the error matrix. See nmf.py for the optional arguments and the
return information of run() and run_repeat().
There are several algorithms implemented and included as separate classes. A
specific algorithm can be used by creating an instance of one of the following
classes. By default, NMF() creates an instance of NMF_ANLS_BLOCKPIVOT;
another fast algorithm is NMF_HALS. Examples of using each of these
algorithms are also included in example.py. See nmf.py and the
following references for more information of algorithms.
NMF_ANLS_BLOCKPIVOT - ANLS with block principal pivotingNMF_ANLS_AS_NUMPY - ANLS with scipy.optimize.nnls solverNMF_ANLS_AS_GROUP - ANLS with active-set method and column groupingNMF_HALS - Hierarchical alternating least squaresNMF_MU - Multiplicative updatingJingu Kim, Yunlong He, and Haesun Park. Algorithms for Nonnegative Matrix and Tensor Factorizations: A Unified View Based on Block Coordinate Descent Framework. Journal of Global Optimization, 58(2), pp. 285-319, 2014. http://link.springer.com/content/pdf/10.1007%2Fs10898-013-0035-4.pdf
Jingu Kim and Haesun Park. Fast Nonnegative Matrix Factorization: An Active-set-like Method And Comparisons. SIAM Journal on Scientific Computing (SISC), 33(6), pp. 3261-3281, 2011. https://sites.google.com/site/jingukim/2011_paper_sisc_nmf.pdf
Please send bug reports, comments, or questions to Jingu Kim. Contributions and extensions with new algorithms are welcome.