Model-based proposals
TPE models parameter distributions associated with objective outcomes. GP sampling uses a surrogate model to guide proposals. Their behavior depends on the objective, budget, and sampler settings.
Watch the search learn.
Explore how TPE, Gaussian-process optimization, and random search sample the same objective. Step through the recorded observations and compare what each strategy has found.
Loading the bundled dataset…
What changes after observation?
A useful baseline
Random search samples without learning from previous objective values. It can still perform well. This small, seeded demonstration is not evidence that one strategy is best for inference tuning.