Thompson Sampling (2D) - ScaleKernel(RBF) - GP fitted via MAP
About this demo
A live Bayesian optimization loop on 2D test functions. Each iteration: the GP model fits to all observations so far, the acquisition function (Thompson Sampling) identifies the most promising point, and a new trial is evaluated.
What you see:
Left panel — the true objective function (ground truth). Observed points appear as dots.
Center panel — the GP posterior mean (what the model thinks the function looks like).
Right panel — the acquisition function value, showing where the optimizer wants to sample next.
Interactivity: Choose a test function from the dropdown, then click Iterate to step through the loop one trial at a time, or Auto to run continuously.
Click Iterate to step through, or Run All for animation