GitHub - mnandan/AENMF: Fast NMF solver

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 26 Commits
AENMF.cpp		AENMF.cpp
AENMF.gitignore		AENMF.gitignore
Makefile		Makefile
README		README
all_data.cpp		all_data.cpp
all_data.h		all_data.h
common.h		common.h
data_handler.h		data_handler.h
dense_mat.cpp		dense_mat.cpp
dense_mat.h		dense_mat.h
derive_ae.cpp		derive_ae.cpp
derive_ae.h		derive_ae.h
file_int.cpp		file_int.cpp
file_int.h		file_int.h
get_factors.cpp		get_factors.cpp
get_factors.h		get_factors.h
inp_params.cpp		inp_params.cpp
inp_params.h		inp_params.h
sparse_mat.cpp		sparse_mat.cpp
sparse_mat.h		sparse_mat.h

Repository files navigation

Fast NMF solver: version 0
--------------------

Find non-negative factors W and H, of a non-negative matrix F so that || F - WH || (frobineus norm) is minimized. A new algorithm similar to the algorithm used in Nandan et al. 2014 (JMLR) is used. 

Instructions:

1. Run make to compile.
2. Call as $ ./AENMF <input file in LibSVM format with X> R <desired file for W> <desired file for H>
The last three parameters are optional. example (1) $ ./AENMF ../data/mnist 20 w.dat h.dat
										example (2) $ ./AENMF ../data/mnist

3. The output files are also in LibSVM/SVMlight format.
4. The Frobenius norm error of each iteration is displayed.