Bayesian Clustering for CLV RFM Segmentation

[ColtAllen]

For the upcoming RFM Segmentation Notebook, I want to compare clv.utils.rfm_segments against more sophisticated (albeit computationally intensive) clustering methods.

I'm envisioning a Dirichlet process for estimating the number of segments:

And/or a Gaussian Mixture Model to identify "fringe" customers for conversion:

These could be great additions to the library!

(Images taken from the linked notebooks in the pymc docs)