Selecting the Number of Mixture Components
Because likelihood alone always improves with more components, K is chosen by penalized criteria: BIC (penalty (P/2) log N, consistent, favored for density estimation) or AIC (penalty P, lighter), scored across candidate K and covariance structures on held-in data.
This Concept is waiting for its first lesson!
Because likelihood alone always improves with more components, K is chosen by penalized criteria: BIC (penalty (P/2) log N, consistent, favored for density estimation) or AIC (penalty P, lighter), scored across candidate K and covariance structures on held-in data.
Are you a teacher? Sign in to start contributing.
Sign In