Stopping Conditions

By default GFO runs for a fixed number of iterations specified by n_iter. In practice you often want additional safeguards: a time limit to prevent runaway computations, a target score to stop early when good enough results are found, or early stopping to detect when the optimizer has converged and further iterations would not improve the result. All of these can be combined freely. Optimization stops when any condition is met.

n_iter

Fixed iteration count limit.

max_time

Wall-clock time limit in seconds.

max_score

Stop when target score is reached.

early_stopping

Stop after N iterations without improvement.

Available Conditions

Iteration count (n_iter):

opt.search(objective, n_iter=1000)

Time limit (max_time):

opt.search(objective, n_iter=10000, max_time=3600)  # 1 hour max

Target score (max_score):

opt.search(objective, n_iter=1000, max_score=0.99)

Early stopping:

opt.search(
    objective,
    n_iter=1000,
    early_stopping={"n_iter_no_change": 50}
)

Combining Conditions

Use multiple conditions together. Optimization stops when ANY condition is met:

opt.search(
    objective,
    n_iter=5000,              # Max 5000 iterations
    max_time=1800,            # OR max 30 minutes
    max_score=0.999,          # OR target reached
    early_stopping={
        "n_iter_no_change": 100  # OR no improvement for 100 iterations
    }
)

Early Stopping Details

Early stopping monitors the best score and stops if no improvement is found for a specified number of iterations:

early_stopping = {
    "n_iter_no_change": 50,  # Stop after 50 iterations without improvement
}

This is useful when:

• You don’t know how many iterations are needed

• The function converges before the iteration budget

• You want to save computation time

Examples

ML hyperparameter tuning:

opt.search(
    objective,
    n_iter=200,
    max_time=7200,            # 2 hour limit
    max_score=0.98,           # Stop if 98% accuracy reached
    early_stopping={"n_iter_no_change": 30},
)

Quick exploration:

opt.search(
    objective,
    n_iter=100,
    max_time=60,  # Quick 1-minute exploration
)

Long-running optimization:

opt.search(
    objective,
    n_iter=10000,
    max_time=86400,  # 24-hour limit
    early_stopping={"n_iter_no_change": 500},
)

Checking Why Optimization Stopped

After optimization, you can check how many iterations were completed:

opt.search(objective, n_iter=1000, max_score=0.99)

print(f"Completed iterations: {len(opt.search_data)}")
print(f"Best score: {opt.best_score}")

if len(opt.search_data) < 1000:
    if opt.best_score >= 0.99:
        print("Stopped: target score reached")
    else:
        print("Stopped: early stopping or time limit")