## Issue Constructing and indexing the training data space for k-means is too expensive. ## Solution - Use sampling (e.g. MCMC) to repeatedly sample data and construct reduced space. - Run process until space properties become stationary.