Feature suggestion: set minimum cluster size

[HLammers]

I'm using FasterPAM and FasterMSC on a project and I run into the issue that I sometimes get clusters on size n=1. For my usecase that is not desirable. Would it be possible to add a minimum cluster size parameter, if that is algorithmically possible?

[kno10]

It certainly is possible, but tends to make the optimization more complex.
E.g., similar to this k-means variation: https://elki-project.github.io/tutorial/same-size_k_means
I suggest you rather experiment with DynMSC that does not need to specify k.
Or you just drop those points as outliers, and rerun. Or increase k. if done right, you don't need to recompute your distance matrix (which is most of the effort), and you could also use the existing centers as initialization.

But either way, you may want to consider reviewing your objective. Isolating such points is typically not a bug, but optimal. So if these one-elemental clusters are an issue for your approach, maybe you are using the wrong optimization function. Make sure you know what you are optimizing for - do you have a "proof" why one-elemental clusters are bad? Try formalizing what you are trying to achieve; then pick the right tool for this objective.

[HLammers]

Thank you for your swift response. I've implemented a remove and rerun algorithm and it works well enough.