How to Nest Wrappers¶

This guide shows you how to compose wrappers that contain other wrappers, and how to control parameters at any nesting depth using the double-underscore (__) syntax.

Interactive notebook available

Try this guide as an interactive notebook: Nested Wrapper Parameters

Prerequisites¶

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

Create a Nested Structure¶

Pass inner wrappers as constructor parameters to an outer wrapper:

from sklearn_wrap.base import BaseClassWrapper, _fit_context
from sklearn.base import RegressorMixin

class InnerWrapper(BaseClassWrapper, RegressorMixin):
    _estimator_name = "regressor"
    _estimator_base_class = object

    @_fit_context(prefer_skip_nested_validation=True)
    def fit(self, X, y):
        self.instance_.fit(X, y)
        return self

    def predict(self, X):
        return self.instance_.predict(X)


class OuterWrapper(BaseClassWrapper, RegressorMixin):
    _estimator_name = "ensemble"
    _estimator_base_class = object

    @_fit_context(prefer_skip_nested_validation=False)  # Validate nested
    def fit(self, X, y):
        self.instance_.fit(X, y)
        return self

    def predict(self, X):
        return self.instance_.predict(X)


inner = InnerWrapper(regressor=SomeRegressor, scale=0.8)
outer = OuterWrapper(ensemble=SomeEnsemble, estimator1=inner, blend=0.5)

Inspect Nested Parameters¶

Use get_params(deep=True) to see the full parameter hierarchy:

params = outer.get_params(deep=True)
# {'ensemble': SomeEnsemble,
#  'estimator1': InnerWrapper(...),
#  'estimator1__regressor': SomeRegressor,
#  'estimator1__scale': 0.8,
#  'blend': 0.5}

Set Nested Parameters¶

Use the __ syntax to modify parameters at any depth:

outer.set_params(estimator1__scale=1.5)

This calls set_params(scale=1.5) on the estimator1 inner wrapper.

Use prefer_skip_nested_validation Correctly¶

For outer wrappers that contain nested wrappers, set prefer_skip_nested_validation=False so that nested wrappers also validate their parameters:

class OuterWrapper(BaseClassWrapper, RegressorMixin):
    _estimator_name = "ensemble"
    _estimator_base_class = object

    @_fit_context(prefer_skip_nested_validation=False)
    def fit(self, X, y):
        self.instance_.fit(X, y)
        return self

For leaf wrappers (no nested wrappers), use prefer_skip_nested_validation=True to avoid redundant validation.

Validate Nested Wrapper Types¶

Use the wrapper_base_class constraint to ensure a parameter is a wrapper wrapping a specific class:

class OuterWrapper(BaseClassWrapper, RegressorMixin):
    _estimator_name = "ensemble"
    _estimator_base_class = object
    _parameter_constraints = {
        "base_estimator": [{"wrapper_base_class": SomeBaseClass}],
    }

This verifies at validation time that base_estimator is a BaseClassWrapper whose _estimator_base_class matches SomeBaseClass.