Faiss k-means clustering
The faiss library has a k-means clustering implementation that seems to be quite handy.
I would like to test its performance, using a user-specified value of k (sqrt(N/2) by default), and perhaps use it to achieve better 2D projections in the large spaces. Sometimes I am obtaining unacceptable 2D projections (a big ball in the middle, satellite points around) and I thought I might be able to solve this issue by just clustering the centroids, and then mapping the rest of the molecules to the 2D projections of the centroids.
For this, I would need some help with the faiss library...