Associate cluster label with centroid

[vihu]

Summary

It seems to me that there was currently no way to get the centroid of a cluster associated with its label. I need to know this for some external usage.

This patch makes it so that the center now also returns the associated cluster label making it easier to idenfity each cluster's centroid.

[tom-whitehead]

Hey @vihu, I can't merge this PR because it changes the library's API and therefore represents breaking changes that would need to be a major version bump.

I've double checked the existing centroids calculation and it looks fine to me. I just manually calculated the centroids on some test data to verify this. The centroid vector returned is in the order of the labels. For example if you want to get the centroid of the cluster with label 0 you can just do:

let centers = clusterer.calc_centers(Center::Centroid, &labels).unwrap();
let centroid = &centroids[0]

I took this from the Python HDBSCAN implementation. Let me know if it's not clear.

I will update the docstring when I get a chance to publish a new version to make it more obvious.

[vihu]

Hey @vihu, I can't merge this PR because it changes the library's API and therefore represents breaking changes that would need to be a major version bump.

I've double checked the existing centroids calculation and it looks fine to me. I just manually calculated the centroids on some test data to verify this. The centroid vector returned is in the order of the labels. For example if you want to get the centroid of the cluster with label 0 you can just do:

let centers = clusterer.calc_centers(Center::Centroid, &labels).unwrap();
let centroid = &centroids[0]

I took this from the Python HDBSCAN implementation. Let me know if it's not clear.

I will update the docstring when I get a chance to publish a new version to make it more obvious.

Ah I didn't realize, that makes sense. Thanks for the information! I'll close this PR since I can just lookup the centroids using the cluster label as index.