A Symmetric Measure of Clustering Separation

Bringing some order to bear on unsupervised learning

Suppose you want to cluster 100 stocks. That is, you’re interested in finding out which stocks covary with which ones without bothering about why. Ideally, you want crisp clustering that assigns each stock to one and only one of the clusters. You also want your clusters to display a certain separability property. In some informative low-dimensional spac…