Population sample Distribution mean Best-ever point
drag = pan · scroll = zoom
best f (log) σ (log)

CMA-ES 2D: Covariance Adaptation on a Loss Contour Map

This is the 2D companion to the 3D CMA-ES landscape simulator: the same real update equations — weighted-mean recombination, rank-1 and rank-μ covariance adaptation, and cumulative step-size control — run independently here over a top-down contour map (Sphere, Rosenbrock or Rastrigin), with a live convergence chart tracking best fitness and step size alongside. Pan and zoom the contour map, toggle covariance/step-size adaptation on or off, and watch the sampling ellipse stretch, rotate and shrink toward the optimum generation by generation.