Implementation:Online ml River Ensemble StackingClassifier
| Knowledge Sources | |
|---|---|
| Domains | Online_Learning, Ensemble_Methods, Stacking |
| Last Updated | 2026-02-08 16:00 GMT |
Overview
Stacking classifier uses a meta-classifier to learn optimal combinations of base classifier predictions for improved binary classification performance.
Description
This implementation trains multiple base classifiers and a meta-classifier in parallel. For each instance, base classifiers first make predictions (out-of-fold style), then update themselves. The meta-classifier learns from these predictions, optionally augmented with original features. Predictions from each base model are formatted as "oof_i_k" features where i is the model index and k is the class label. This prevents information leakage while allowing the meta-model to learn which base models to trust in different scenarios.
Usage
Use stacking when you have multiple diverse base classifiers and want to learn their optimal combination. It's particularly effective when base models have complementary strengths. Set include_features=True to give the meta-classifier access to both predictions and original features, or False to rely only on base model outputs. Choose a simple, fast meta-classifier like LogisticRegression to avoid overfitting.
Code Reference
Source Location
- Repository: Online_ml_River
- File: river/ensemble/stacking.py
Signature
class StackingClassifier(base.Ensemble, base.Classifier):
def __init__(
self,
models: list[base.Classifier],
meta_classifier: base.Classifier,
include_features=True,
):
super().__init__(models)
self.meta_classifier = meta_classifier
self.include_features = include_features
Import
from river import ensemble
I/O Contract
Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
| models | list[Classifier] | required | Base classifiers for level 0 |
| meta_classifier | Classifier | required | Meta-classifier for level 1 |
| include_features | bool | True | Pass original features to meta-classifier |
Attributes
| Attribute | Type | Description |
|---|---|---|
| models | list | Base classifier instances |
| meta_classifier | Classifier | Meta-classifier instance |
Input/Output
| Method | Input | Output |
|---|---|---|
| learn_one | x: dict, y: Any | None |
| predict_proba_one | x: dict | dict[Any, float] |
| predict_one | x: dict | Any |
Usage Examples
from river import compose
from river import datasets
from river import ensemble
from river import evaluate
from river import linear_model as lm
from river import metrics
from river import preprocessing as pp
dataset = datasets.Phishing()
model = compose.Pipeline(
('scale', pp.StandardScaler()),
('stack', ensemble.StackingClassifier(
[
lm.LogisticRegression(),
lm.PAClassifier(mode=1, C=0.01),
lm.PAClassifier(mode=2, C=0.01),
],
meta_classifier=lm.LogisticRegression()
))
)
metric = metrics.F1()
evaluate.progressive_val_score(dataset, model, metric)
# F1: 88.14%