API References

BCAPSampler(searchspace;rng)

Define a iterator for the BCAP sampler.

BudgetExceeded([msg])

Used to inform if budget has been exceeded.

GridSampler(;npartitions)

Define a iterator for the grid sampler.

GroupedTrial

Trials grouped per instance.

MedianPruner(;start_after, prune_after)

• start_after: Start up pruner after this number (of completed trials).
• prune_after: Prune a trial after this value (considering the median criteria).
• median_vals: median values reported for each instance.

NoMoreTrials()

Used to inform that ampler refuses to suggest trials.

RandomSampler(;seed, rng)

Define a iterator for the random sampler.

Scenario(;
        parameters...,
        sampler    = default_sampler(),
        pruner     = default_pruner(),
        instances  = [1],
        max_trials = :auto,
        max_evals  = :auto,
        max_time   = :auto,
        verbose    = false,
        batch_size = max(nprocs(), Sys.CPU_THREADS),
    )

Define an Scenario with parameters, and budget.

• sampler sampler to be used.
• pruner pruner to reduce computational cost.
• instances array (iterator) containing the problem instances.
• max_trials maximum number of trials to be evaluated on optimize.
• max_evals maximum number of function evaluations.
• max_time maximum execution time on optimize.
• verbose show message during the optimization.
• batch_size number of trials evaluated for each instance for each iteration.

StatusHyperTuning

Current status of the optimize process for scenario.

..(a, b)

Define a interval between a and b (inclusive).

See also BoxConstrainedSpace

all_instances_succeeded(scenario)

Check if all instances are successfully solved.

See also report_success

best_parameters(scenario)

Return best parameters saved in scenario.

budget_exceeded(scenario)

Check whether if budget is not exceeded.

default_rng_ht(seed)

Default random number generator in HyperTuning.

get_convergence(scenario)

Return vector of tuples containing the trial id and objective value.

get_instance(trial)

Get instance problem which trials has to be evaluated.

history(scenario)

Return all evaluated trails.

optimize!(f, scenario)

Perform an iteration of the optimization process.

When this function is called, scenario.batch_size trials are sampled and evaluated, then update the sampler and pruner.

See also optimize

optimize(f, scenario)

Optimize f on provided scenario while the budget is not exceeded (limited by maxtrials, maxevals, etc).

See also optimize!

parameters(hyperparameters...)

Define hyperparameters.

report_success!(trial)

Report that the trial successfully solved the instance.

report_value!(trial, value)

Report (to the pruner) evaluated value at trail.

save_trials!(ungrouped_trials, scenario)

Save evaluated trials into scenario history, then update best trial found so far. Also, report values to the sampler.

should_prune(trial)

Check whether trial should be pruned.

top_parameters(scenario; ignore_pruned)

Return an array of trade-off trials (regarding success, mean objective value, time, etc).

update_best_trial!(scenario, trials)

Update best trial found so far.

@suggest x in trial

Return a value for x stored in the sampled trial.