autogridsearch_wrapper

pybear.model_selection.autogridsearch_wrapper(GridSearchParent)

Wrap a scikit-learn, pybear, or dask_ml GridSearchCV class with a class that overwrites the fit method of that GridSearchCV.

The superseding fit method automates multiple calls to the parent fit method with progressively more precise search grids based on previous search results. See the scikit, pybear, and dask_ml documentation for more information about the available GridSearchCV modules.

Parameters:

GridSearchParentobject

Sci-kit, pybear, or dask_ml GridSearchCV CLASS, not instance.

Returns:

AutoGridSearchobject

Wrapped GridSearchCV class. The original fit method is replaced with a new fit method that can make multiple calls to the original fit method with increasingly precise search grids.

See also

sklearn.model_selection.GridSearchCV

pybear.model_selection.GSTCV

pybear-dask.model_selection.GSTCVDask

Examples

>>> from pybear.model_selection import autogridsearch_wrapper
>>> from sklearn.model_selection import GridSearchCV as sk_GSCV
>>> from sklearn.linear_model import Ridge as sk_RR
>>> from sklearn.datasets import make_regression
>>>
>>> AGSCV = autogridsearch_wrapper(sk_GSCV)
>>> _params = {
...     'alpha': [[0, 0.1, 0.2], 3, 'soft_float'],
...     'fit_intercept': [[True, False], [2, 1, 1], 'fixed_bool']
... }
>>> agscv = AGSCV(estimator=sk_RR(), params=_params, total_passes=3)
>>> X, y = make_regression(n_samples=20, n_features=2, n_informative=2)
>>> agscv.fit(X, y)
AutoGridSearch(
    estimator=Ridge(),
    params={
        'alpha': [[0.0, 0.1, 0.2], [3, 3, 3], 'soft_float'],
        'fit_intercept': [[True, False], [2, 1, 1], 'fixed_bool']
    },
    total_passes=3
)
>>> print(agscv.RESULTS_[1])
{'alpha': 0.025, 'fit_intercept': True}