Global Search#

Global search optimizers explore the entire search space more thoroughly. They aim to find good solutions without getting trapped in local optima.

Random Search#

Samples random points from the search space. Simple but surprisingly effective baseline.

from hyperactive.opt.gfo import RandomSearch

optimizer = RandomSearch(
    search_space=search_space,
    n_iter=100,
    experiment=objective,
)

Grid Search#

Evaluates all combinations systematically. Only practical for small search spaces.

from hyperactive.opt.gfo import GridSearch

optimizer = GridSearch(
    search_space=search_space,
    experiment=objective,
)

Random Restart Hill Climbing#

Runs hill climbing from multiple random starting points.

from hyperactive.opt.gfo import RandomRestartHillClimbing

optimizer = RandomRestartHillClimbing(
    search_space=search_space,
    n_iter=100,
    experiment=objective,
)

Powell’s Method and Pattern Search#

Classical derivative-free optimization methods.

from hyperactive.opt.gfo import PowellsMethod, PatternSearch

When to Use Global Search#

Global search optimizers are best suited for:

Establishing baselines: Random search is a strong baseline for any problem
Small search spaces: Grid search provides exhaustive coverage
Unknown landscapes: When you don’t know the structure of your objective
Simple problems: When more sophisticated methods aren’t necessary

Consider using population-based or model-based methods if:

Random search isn’t finding good solutions
You have expensive evaluations and need smarter exploration
Your search space is too large for grid search