UMAP for visualizing clusters from sensor data

tSNE is so 2017

I would be lying if I said I anticipated I would say this in 2018.

UMAP is the new dimension reduction technique that is hot off the press. This post is an exposition of how UMAP can be deployed for unsupervised clustering of sensor data from Parkinson’s disease patients to distinguish the different traits. The source code is hosted at my Github.

Coming from the same class of manifold based algorithms as tSNE, UMAP goes beyond tSNE in that it also preserves the global structure when going from higher dimensions to lower dimensions while tSNE preserves the local relationships only.

(Under construction…)