Preferential Bayesian Optimization

Instead of observing a numeric outcome, you simply choose which of two options you prefer. The model learns a latent utility function from these pairwise comparisons and suggests increasingly better options.

How it works: A Pairwise GP models your preferences using Laplace approximation. The EUBO (Expected Utility of Best Option) acquisition function picks the next pair to show you — the pair most likely to reveal new information about what you like.

Interactivity: Click the pattern you prefer. After a few rounds, the model converges on your taste. Switch to Auto Mode to let a simulated user make choices, or try different visual styles from the dropdown.