Michel de Ruiter

Set `allow_single_cluster`=True?

Ah! I still would not expect a crash. Plus finding a difference where there is none is _not_ helpful as default behavior...

Even piping through `head` loses the colors. :-(

That helps, thanks! Still really slow though. I would expect there's a way to move stuff _in batch_, without interface updates along the way. Maybe that could further speed things...

Thanks. It takes a while but it does work. I might play a bit!

The problem is in the 'definition' of `neighbor_matrix`: `_compute_distance_and_neighbor_matrix` returns indexes _within_ the cluster, but `_prob_distances_ev` treats the numbers as being global.