Skip to content

mims-harvard/nimfa

Repository files navigation

Nimfa

build: passing build: passing GitHub release BSD license Conda Version

Nimfa is a Python module that implements many algorithms for nonnegative matrix factorization. Nimfa is distributed under the BSD license.

The project was started in 2011 by Marinka Zitnik as a Google Summer of Code project, and since then many volunteers have contributed. See AUTHORS file for a complete list of contributors.

It is currently maintained by a team of volunteers.

[News:] Scikit-fusion, collective latent factor models, matrix factorization for data fusion and learning over heterogeneous data.

[News:] fastGNMF, fast implementation of graph-regularized non-negative matrix factorization using Facebook FAISS.

Important links

Dependencies

Nimfa is tested to work under Python 2.7 and Python 3.4.

The required dependencies to build the software are NumPy >= 1.7.0, SciPy >= 0.12.0.

For running the examples Matplotlib >= 1.1.1 is required.

Install

This package uses setuptools, which is a common way of installing python modules. To install in your home directory, use:

python setup.py install --user

To install for all users on Unix/Linux:

sudo python setup.py install

For more detailed installation instructions, see the web page http://ai.stanford.edu/~marinka/nimfa.

Alternatively, you may also install this package using conda:

conda install -c conda-forge nimfa

Use

Run alternating least squares nonnegative matrix factorization with projected gradients and Random Vcol initialization algorithm on medulloblastoma gene expression data:

>>> import nimfa
>>> V = nimfa.examples.medulloblastoma.read(normalize=True)
>>> lsnmf = nimfa.Lsnmf(V, seed='random_vcol', rank=50, max_iter=100)
>>> lsnmf_fit = lsnmf()
>>> print('Rss: %5.4f' % lsnmf_fit.fit.rss())
Rss: 0.2668
>>> print('Evar: %5.4f' % lsnmf_fit.fit.evar())
Evar: 0.9997
>>> print('K-L divergence: %5.4f' % lsnmf_fit.distance(metric='kl'))
K-L divergence: 38.8744
>>> print('Sparseness, W: %5.4f, H: %5.4f' % lsnmf_fit.fit.sparseness())
Sparseness, W: 0.7297, H: 0.8796

Cite

@article{Zitnik2012,
  title     = {Nimfa: A Python Library for Nonnegative Matrix Factorization},
  author    = {Zitnik, Marinka and Zupan, Blaz},
  journal   = {Journal of Machine Learning Research},
  volume    = {13},
  pages     = {849-853},
  year      = {2012}
}

Selected publications (Methods)

Selected publications (Applications)

Tutorials