K-Means Seeding for EM Initialization
Running k-means first and initializing each component's mean at a cluster centroid, its covariance from the cluster's scatter, and its weight from cluster sizes gives EM a start near a good basin; random initialization with multiple restarts is the fallback.
This Concept is waiting for its first lesson!
Running k-means first and initializing each component's mean at a cluster centroid, its covariance from the cluster's scatter, and its weight from cluster sizes gives EM a start near a good basin; random initialization with multiple restarts is the fallback.
Are you a teacher? Sign in to start contributing.
Sign In