Cover Tree with Friends is the fastest-known way to solve nearest-neighbors of pointclouds in Euclidean space, and they also allow fast approximations of the persistent homology of large data sets. By extending the notion of Friends, we developed CDER, a new method for supervised learning of labelled pointclouds in Euclidean space. It is a deterministic, multiscale, data-driven method that finds distinguishing regions of any scale with surprising efficiency. We think of CDER as “diff for pointclouds.” This talk is an overview of the preprint available at https://arxiv.org/abs/1702.07959 This is joint work with Paul Bendich, John Harer, and Jay Hineman.