Clustering · GMM

Interactive GMM simulator (Expectation-Maximization)

Each click advances half an EM iteration. In the E-step the point colors change to reflect each Gaussian's responsibility. In the M-step the ellipses move and adjust their covariance. Watch the log-likelihood grow until it converges.

simulatorGMM step by step
step 0 / 1
dataset
K = 3
phase: init
state
K (components)3
iteration0
next stepE-step
log-likelihood—
Δ likelihood—
statusiterating

Initialized with 3 isotropic Gaussians placed at the most spread-out points. The next step runs the E-step, computing each Gaussian's responsibility for each point.