Leland McInnes

Duplicates ending up in different clusters *shouldn't* happen; it isn't intended behaviour. So something is a little odd there. It is possible if everything lines up just so (core distances,...

Semi-supervised clustering with partial labelling is a research interest of mine actually. There are certainly some ways to do it, but they are more in the "research" realm than the...

In that case the sstsne approach is the best thing I know of that there is a ready implementation of. Try that and let me know if it does any...

The ``y`` parameter is for conformity with sklearn's API. It is unused. In principle I think there are better things you can do now to build semi-supervision into the clustering...

You can get multiple granularities of clustering from one run; you can't get it by specifying a number of clusters however. The current best approach in a released version involves...

Not really; you can run a nearest neighbour search and look at the distribution of k-neighbour distances and see what looks useful based on that.

I don't have the time to fix this these days, so I think it is safest to assume the soft clustering is unmtaintained at this stage.

I think there may be an issue with code duplication that has fallen out of sync -- the membership / prediction data code needs to get a cluster tree, and...

It is certainly possible -- I don't have the exact approach to hand right now, but if you look at the code in the prediction.py file you'll see how the...

What version are you using? Can you also check if this is the case if you install directly from the master branch on github?