How to Use with GridSearchCV¶

This guide shows you how to tune hyperparameters of wrapped estimators using Scikit-Learn's GridSearchCV, Pipeline, and cross_val_score.

Interactive notebook available

Try this guide as an interactive notebook: Grid Search Integration

Prerequisites¶

sklearn-wrap installed (Getting Started)
A working wrapper class (How to Wrap a Class)

Define a Parameter Grid¶

Wrapped estimators expose parameters through the standard get_params() / set_params() interface. Build your parameter grid using those parameter names:

from sklearn.model_selection import GridSearchCV

wrapper = MyWrapper(model=MyClass, alpha=1.0, n_iters=100)

param_grid = {
    "alpha": [0.01, 0.1, 1.0, 10.0],
    "n_iters": [50, 100, 200],
}

search = GridSearchCV(wrapper, param_grid, cv=5, scoring="neg_mean_squared_error")
search.fit(X, y)

print(search.best_params_)
print(search.best_score_)

Tune Nested Parameters¶

For wrappers that contain nested wrappers or estimators, use the double-underscore (__) syntax to reach into nested levels:

param_grid = {
    "estimator1__scale": [0.5, 1.0, 1.5],
    "estimator2__scale": [0.8, 1.2],
    "blend": [0.3, 0.5, 0.7],
}

search = GridSearchCV(ensemble_wrapper, param_grid, cv=3)
search.fit(X, y)

The __ notation traverses the parameter hierarchy. estimator1__scale calls set_params(estimator1__scale=value) on the outer wrapper, which delegates to the inner wrapper's set_params(scale=value).

Use in a Pipeline¶

Wrapped estimators work in Pipeline like any other estimator. Prefix parameter names with the pipeline step name:

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

pipe = Pipeline([
    ("scaler", StandardScaler()),
    ("model", MyWrapper(model=MyClass, alpha=1.0)),
])

param_grid = {
    "model__alpha": [0.01, 0.1, 1.0],
}

search = GridSearchCV(pipe, param_grid, cv=5)
search.fit(X, y)

Use cross_val_score¶

For quick evaluation without parameter search:

from sklearn.model_selection import cross_val_score

wrapper = MyWrapper(model=MyClass, alpha=1.0)
scores = cross_val_score(wrapper, X, y, cv=5, scoring="neg_mean_squared_error")
print(f"Mean score: {scores.mean():.4f} (+/- {scores.std():.4f})")