NonnegMatrixFactor

This Julia repo implements a variety of Nonnegative Matrix Factorization (NMF) models in Julia, a fast, flexible and free programming language. NMF attempts to solve the optimization problem

minimize D(X || W H) s.t. W >= 0, H >= 0

where D(X || W H) is some distance metric between X and W H.

Usage

Fitting an NMF model is easy. Simply choose your algorithm and any relevant hyperparemeters. For example,

>> using NonnegMatrixFactor
>> results = fit_nmf(data, alg=:mult, K=3, max_itr=100)
>>
>> using Plots
>> plot(results.time_hist, results.loss_hist)

Installation

First, download and install the latest version of Julia. Then install the NonnegMatrixFactor package:

>> using Pkg
>> Pkg.add("NonnegMatrixFactor")

Supported Features

The following NMF models and algorithms are available

• Frobenius norm loss (square error)
  • Multiplicative updates (alg=:mult)
  • Hierachical alternating least squares (alg=:hals)
  • Projected alternaing least squares (alg=:projals)
• KL-divergence loss
  • Multiplicative updates (alg=:kl_mult)