Description

@GaelVaroquaux Maybe an online version of the NMF could be useful in big data settings.

Steps/Code to Reproduce

Expected Results

Actual Results

Here I show the results for an sparse matrix as input of size 1M x 2.8k
The factorization has n_components = 10 and each point is an entire pass on the data.
The current online implementation only supports the kullback-leibler divergence, but in practice
can be generalized to any beta divergence.

Versions