Development version

You are reading the latest (dev) docs. For the stable version, see stable.

Estimator Presets

Preset functions return ordinary param_grid dictionaries using Integer, Continuous, and Categorical dimensions. They are meant as strong starting points, not universal best settings.

from sklearn.ensemble import RandomForestClassifier

from sklearn_genetic import GASearchCV, random_forest_classifier_space

search = GASearchCV(
    estimator=RandomForestClassifier(random_state=42),
    param_grid=random_forest_classifier_space(profile="balanced"),
    scoring="roc_auc",
    cv=5,
    random_state=42,
)

Available presets

Function	Estimator
random_forest_classifier_space	RandomForestClassifier
random_forest_regressor_space	RandomForestRegressor
hist_gradient_boosting_classifier_space	HistGradientBoostingClassifier
hist_gradient_boosting_regressor_space	HistGradientBoostingRegressor
logistic_regression_space	LogisticRegression with solver="saga" and penalty="elasticnet"
svc_space	SVC
xgboost_classifier_space	xgboost.XGBClassifier
xgboost_regressor_space	xgboost.XGBRegressor

Profiles

Each preset accepts profile="fast", "balanced", or "wide":

param_grid = random_forest_classifier_space(profile="fast")

Profile	Use when
fast	You want a quick first run or a notebook demo
balanced	You want a practical default for real tuning
wide	You have more budget and want broader exploration

Discovering presets

Two helpers let you list the available presets and profiles from Python — handy in a notebook or for building a UI:

from sklearn_genetic import list_preset_profiles, list_preset_spaces

list_preset_profiles()
# ['balanced', 'fast', 'wide']

list_preset_spaces()
# ['hist_gradient_boosting_classifier_space', ..., 'xgboost_regressor_space']

Both return sorted lists, and every name from list_preset_spaces() is importable from sklearn_genetic.

Pipeline prefixes

Use prefix when tuning an estimator inside a scikit-learn Pipeline:

from sklearn_genetic import svc_space

param_grid = svc_space(prefix="model__")

This returns parameter names such as model__C, model__kernel, and model__gamma.

Full example: tuning a preset inside a Pipeline

The prefix must match the step name of the estimator in the pipeline. If the step is named "model", use prefix="model__" so the keys become model__n_estimators, model__max_depth, and so on:

from sklearn.datasets import load_breast_cancer
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

from sklearn_genetic import GASearchCV
from sklearn_genetic.presets import random_forest_classifier_space

X, y = load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, stratify=y, random_state=0
)

pipe = Pipeline(
    [
        ("scale", StandardScaler()),
        ("model", RandomForestClassifier(random_state=0, n_jobs=1)),
    ]
)

# prefix="model__" matches the "model" step, e.g. model__n_estimators
param_grid = random_forest_classifier_space(prefix="model__")

search = GASearchCV(
    estimator=pipe,
    param_grid=param_grid,
    scoring="roc_auc",
    cv=3,
    population_size=10,
    generations=5,
    random_state=0,
)
search.fit(X_train, y_train)
print("Best ROC AUC:", round(search.best_score_, 4))

TIP

The prefix is just the pipeline step name followed by __. If your step is called "clf", use prefix="clf__". See Tuning scikit-learn Pipelines for more.

XGBoost

The XGBoost presets cover the common nine-parameter tree booster space used in the XGBoost tutorials and recipes:

from xgboost import XGBClassifier

from sklearn_genetic import GASearchCV, RuntimeConfig, xgboost_classifier_space

search = GASearchCV(
    estimator=XGBClassifier(tree_method="hist", eval_metric="logloss", n_jobs=1),
    param_grid=xgboost_classifier_space(profile="balanced"),
    scoring="roc_auc",
    cv=5,
    runtime_config=RuntimeConfig(n_jobs=-1, parallel_backend="cv"),
    random_state=42,
)

XGBoost threading

Set n_jobs=1 on the XGBoost estimator and parallelize with parallel_backend="cv" on GASearchCV to avoid CPU oversubscription.

See Also

• Search Space — native dimensions and sklearn/scipy conversion
• Tuning scikit-learn Pipelines
• Choosing the Right Search Space
• XGBoost Hyperparameter Tuning