Implementation:Online ml River Compat RiverToSklearn

Knowledge Sources	Online_ml_River
Domains	Online_Learning, Scikit_Learn, Compatibility, Machine_Learning
Last Updated	2026-02-08 16:00 GMT

Overview

Compatibility wrappers that convert River online learning models to scikit-learn compatible estimators with batch processing interfaces.

Description

The river-to-sklearn compatibility layer provides wrapper classes that adapt River's online learning estimators to work with scikit-learn's batch learning API. The module includes four main wrapper classes: River2SKLRegressor, River2SKLClassifier, River2SKLTransformer, and River2SKLClusterer.

Each wrapper maintains a deep copy of the River estimator and implements scikit-learn's standard interface methods (fit, partial_fit, predict, transform). The wrappers handle data validation, feature consistency checks, and proper streaming of batch data through the River model's learn_one/predict_one interface.

For classifiers, the module handles both binary and multiclass cases, applying label encoding when necessary for binary classifiers. For transformers, it distinguishes between supervised and unsupervised transformers. The convert_river_to_sklearn function automatically detects the estimator type and applies the appropriate wrapper.

Usage

Use these wrappers when you need to integrate River's online learning models into scikit-learn pipelines or workflows that expect the standard scikit-learn API. This is particularly useful for comparing online learning models with batch models, or for using River models in existing scikit-learn-based systems.

Code Reference

Source Location

Repository: Online_ml_River
File: river/compat/river_to_sklearn.py

Signature

def convert_river_to_sklearn(estimator: base.Estimator):
    ...

class River2SKLRegressor(River2SKLBase, sklearn_base.RegressorMixin):
    def __init__(self, river_estimator: base.Regressor):
        ...

class River2SKLClassifier(River2SKLBase, sklearn_base.ClassifierMixin):
    def __init__(self, river_estimator: base.Classifier):
        ...

class River2SKLTransformer(River2SKLBase, sklearn_base.TransformerMixin):
    def __init__(self, river_estimator: base.Transformer):
        ...

class River2SKLClusterer(River2SKLBase, sklearn_base.ClusterMixin):
    def __init__(self, river_estimator: base.Clusterer):
        ...

Import

from river import compat

I/O Contract

Input (convert_river_to_sklearn)
Parameter	Type	Description
estimator	base.Estimator	River estimator to wrap

Output
Method	Return Type	Description
convert_river_to_sklearn(estimator)	sklearn wrapper	Returns appropriate scikit-learn compatible wrapper

Common Methods (all wrappers)
Method	Parameters	Return Type	Description
fit(X, y)	X: array-like, y: array-like	self	Fits model on entire dataset
partial_fit(X, y)	X: array-like, y: array-like	self	Incrementally fits model on data portion
predict(X)	X: array-like	ndarray	Makes predictions on dataset

Additional Methods (River2SKLClassifier)
Method	Parameters	Return Type	Description
predict_proba(X)	X: array-like	ndarray	Returns class probabilities

Additional Methods (River2SKLTransformer)
Method	Parameters	Return Type	Description
transform(X)	X: array-like	ndarray	Transforms dataset

Usage Examples

from river import compat
from river import linear_model
from river import preprocessing
from sklearn.model_selection import cross_val_score
import numpy as np

# Convert River regressor to sklearn
river_reg = preprocessing.StandardScaler() | linear_model.LinearRegression()
sklearn_reg = compat.convert_river_to_sklearn(river_reg)

# Now use it like any sklearn estimator
X = np.random.randn(100, 5)
y = np.random.randn(100)

# Batch training
sklearn_reg.fit(X, y)
predictions = sklearn_reg.predict(X)

# Incremental training
X_batch = np.random.randn(20, 5)
y_batch = np.random.randn(20)
sklearn_reg.partial_fit(X_batch, y_batch)

# Example with classifier
from river import naive_bayes

river_clf = naive_bayes.GaussianNB()
sklearn_clf = compat.convert_river_to_sklearn(river_clf)

X_clf = np.random.randn(100, 5)
y_clf = np.random.randint(0, 2, 100)

sklearn_clf.fit(X_clf, y_clf)
probas = sklearn_clf.predict_proba(X_clf)
predictions = sklearn_clf.predict(X_clf)

# Works with sklearn cross-validation
scores = cross_val_score(sklearn_clf, X_clf, y_clf, cv=5)
print(f"Cross-validation scores: {scores}")

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment