Zoltan multilevel hypergraph partitioning doesn't generate enough levels on a network with communities.

[rsln-s]

I'm running zoltan phg on a matrix generated from youtube dataset with ground-truth communities in a following way: communities are hyperedges (rows) and users are vertices (i.e. a vertex is in a hyperedge if as user is a member of the corresponding community)

However, on that matrix with ~660k vertices zoltan only calls coarsening twice and therefore only generates 3 levels with ~590k vertices on the lowest one, despite the fact that I've specifically set PHG_COARSENING_LIMIT = 1000

Matrix and zdrive.inp files are attached. .txt had to be appended to the filenames so that github would let me upload them.

zdrive.inp.youtube.agg.2.txt
youtube.mtx.txt

[rsln-s]

The hypergraph in the example was not connected and had a lot of independent vertices. If you remove them, zoltan behaves correctly.