Implementation:Scikit learn Scikit learn MetadataRouting

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Machine Learning, API Design
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for routing metadata (such as sample weights) within scikit-learn estimator pipelines, provided by scikit-learn.

Description

The sklearn.utils.metadata_routing module provides utilities to route metadata within scikit-learn estimators. It re-exports key classes and functions from the internal _metadata_requests module, including MetadataRouter, MetadataRequest, MethodMapping, process_routing, and get_routing_for_object. It also provides sentinel values UNCHANGED, UNUSED, and WARN for controlling metadata request behavior.

Usage

Use this module when building custom estimators that need to pass metadata (like sample_weight or groups) through pipelines and composite estimators. It enables fine-grained control over which metadata is forwarded to which sub-estimator method.

Code Reference

Source Location

• Repository: scikit-learn
• File: sklearn/utils/metadata_routing.py

Signature

from sklearn.utils._metadata_requests import (
    UNCHANGED,
    UNUSED,
    WARN,
    MetadataRequest,
    MetadataRouter,
    MethodMapping,
    _MetadataRequester,
    _raise_for_params,
    _raise_for_unsupported_routing,
    _routing_enabled,
    _RoutingNotSupportedMixin,
    get_routing_for_object,
    process_routing,
)

Import

from sklearn.utils.metadata_routing import MetadataRouter, process_routing

I/O Contract

Inputs

Name	Type	Required	Description
obj	estimator	Yes	The estimator or meta-estimator to get routing information for
params	dict	Yes	Metadata parameters to be routed to sub-estimators

Outputs

Name	Type	Description
routing	MetadataRouter	Router object containing routing configuration for the estimator
routed_params	dict	Metadata parameters routed to the appropriate sub-estimator methods

Usage Examples

Basic Usage

from sklearn.utils.metadata_routing import MetadataRouter, MethodMapping

router = MetadataRouter(owner="MyEstimator")
# Configure routing for sub-estimators in composite estimators
# This is primarily used internally by scikit-learn's meta-estimators

Related Pages

• Principle:Scikit_learn_Scikit_learn_Pipeline_Execution