Comparing Hyperparameter Optimization Algorithms (2D)

Search setup

Algorithm

Trial budget 60

Landscape ruggedness 45%

Search speed 4

Actions

Live stats

Trials run

Best loss

—

Trials to best

—

FPS

—

Trial sample

Current best

Latest trial

How each strategy searches

Grid — evaluates a fixed evenly-spaced lattice; exhaustive but wastes trials in flat regions.
Random — samples uniformly; often beats grid because it never over-invests in unimportant axes.
Bayesian — explores broadly early, then narrows a shrinking radius around the current best.
Genetic — keeps a population of candidates, breeds the fittest half each generation with mutation.