How to Test Your Wrapper¶

This guide shows you how to verify that a wrapper class is correctly implemented and fully compatible with Scikit-Learn's ecosystem.

Prerequisites¶

A working wrapper class (Getting Started)
pytest installed

Verify Basic Fit/Predict¶

Start with the simplest possible test: fit, predict, and check the output shape.

import numpy as np
import pytest

def test_fit_predict(wrapper_class, estimator_class):
    X = np.random.randn(50, 3)
    y = np.random.randn(50)

    wrapper = wrapper_class(**{wrapper_class._estimator_name: estimator_class})
    wrapper.fit(X, y)
    predictions = wrapper.predict(X)

    assert predictions.shape == (50,)

Verify Cloning¶

Scikit-Learn's clone() creates a new unfitted copy from get_params(). If this fails, GridSearchCV and other meta-estimators will not work.

from sklearn.base import clone

def test_clone(fitted_wrapper):
    cloned = clone(fitted_wrapper)

    # Clone should have same params
    assert cloned.get_params() == fitted_wrapper.get_params()

    # Clone should NOT be fitted
    assert not hasattr(cloned, "instance_")

Verify Parameter Round-Trip¶

Parameters set via set_params() should be reflected in get_params():

def test_param_roundtrip(wrapper):
    original = wrapper.get_params()
    wrapper.set_params(**original)
    assert wrapper.get_params() == original

Verify Fitted State¶

After calling fit(), the wrapper should pass check_is_fitted():

from sklearn.utils.validation import check_is_fitted

def test_fitted_state(wrapper, X, y):
    with pytest.raises(Exception):
        check_is_fitted(wrapper)

    wrapper.fit(X, y)
    check_is_fitted(wrapper)  # Should not raise

Verify GridSearchCV Compatibility¶

This is the most important integration test: does the wrapper work with GridSearchCV?

from sklearn.model_selection import GridSearchCV

def test_gridsearch(wrapper, X, y, param_grid):
    search = GridSearchCV(wrapper, param_grid, cv=2, error_score="raise")
    search.fit(X, y)

    assert search.best_params_ is not None
    assert hasattr(search, "best_score_")

Use error_score="raise" during testing so failures surface immediately rather than being silently replaced with NaN.

Verify Pipeline Compatibility¶

If your wrapper will be used inside pipelines:

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

def test_pipeline(wrapper, X, y):
    pipe = Pipeline([("scaler", StandardScaler()), ("model", wrapper)])
    pipe.fit(X, y)
    predictions = pipe.predict(X)

    assert predictions.shape == (y.shape[0],)

Organize Tests with Fixtures¶

Use pytest fixtures to avoid repeating setup across tests:

import pytest
import numpy as np

@pytest.fixture
def sample_data():
    X = np.random.randn(100, 5)
    y = np.random.randn(100)
    return X, y

@pytest.fixture
def wrapper():
    return MyWrapper(regressor=MyEstimator, alpha=1.0)

@pytest.fixture
def fitted_wrapper(wrapper, sample_data):
    X, y = sample_data
    wrapper.fit(X, y)
    return wrapper

Common Test Failures and Fixes¶

clone() raises TypeError: The wrapper's __init__ signature does not match get_params(). Ensure all constructor parameters are passed through to BaseClassWrapper and not stored separately.
check_is_fitted() raises after fit(): The @_fit_context decorator was not applied to the fit() method, so self.instance_ was never created.
GridSearchCV fails with parameter errors: A parameter name contains __ (reserved for nested access) or the parameter grid references names not in get_params().