EnsemblePursuit-- a sparse matrix factorization algorithm for extracting co-activating neurons from large-scale recordings

Ensemble Pursuit is a matrix factorization algorithm that extracts sparse neural components of co-activating cells.

The matrix U is a sparse matrix (because of an L0 penalty in the cost function) that encodes which neurons belong to a component. V is an average timecourse of these neurons, e.g. component time course.

You can install EnsemblePursuit with "pip install EnsemblePursuit"

For more details see the wiki and our Statistical Analysis of Neural Data 2019 workshop poster

Ensembles learned using EnsemblePursuit from recordings in V1 have Gabor receptive fields.

Some ensembles are well explained by behavior PC's extracted from mouse orofacial movies.