Clustering without labels, and finding the rare event you have almost no examples of.
8h · 11 lessons · 1 challenge
Before this: Decision Trees & Ensembles
Clustering is the first thing to reach for when nothing is labelled. K-means is derived here from its objective rather than described as a procedure, which is what makes initialization and the choice of k make sense.
Anomaly detection follows, and uses the multivariate Gaussian from the probability module. The last lesson is the one that matters in practice: when to model anomalies and when to just train a classifier.
Six two-dimensional projections of the same data. Say which structure is real and which the projection invented.
Worth 350 XP.
