StackingClassifier stacks the output of individual estimators and uses a final estimator to compute predictions.
Constructor
import { StackingClassifier } from 'scikitjs';
const stacking = new StackingClassifier(estimators, finalEstimator, options);
estimators
[string, estimator | () => estimator][]
required
Base estimators to be stacked. Can be instances or factory functions.
finalEstimator
estimator | () => estimator
required
Final estimator trained on the base estimators’ predictions.
Number of cross-validation folds for generating meta-features.
If true, original features are concatenated with base estimator predictions.
options.stackMethod
'auto' | 'predict' | 'predictProba'
default:"'auto'"
Method for base estimators:
'auto': uses predictProba if available, otherwise predict'predict': uses predicted class labels'predictProba': uses predicted probabilities
Random state for cross-validation splitting.
Methods
fit
Fit base estimators and final estimator.
fit(X: number[][], y: number[], sampleWeight?: number[]): StackingClassifier
predict
Predict class labels.
predict(X: number[][]): number[]
predictProba
Predict class probabilities.
predictProba(X: number[][]): number[][]
score
Return accuracy score.
score(X: number[][], y: number[]): number
Properties
The fitted base estimators
The fitted final estimator
Examples
Basic Stacking
import {
StackingClassifier,
LogisticRegression,
DecisionTreeClassifier,
KNeighborsClassifier
} from 'scikitjs';
const X = [
[1, 2], [2, 3], [3, 4], [4, 5],
[5, 6], [6, 7], [7, 8], [8, 9],
[9, 10], [10, 11]
];
const y = [0, 0, 0, 0, 0, 1, 1, 1, 1, 1];
// Stack multiple base estimators
const stacking = new StackingClassifier(
[
['dt', new DecisionTreeClassifier()],
['knn', new KNeighborsClassifier({ nNeighbors: 3 })],
['lr_base', new LogisticRegression()]
],
new LogisticRegression() // Final estimator
);
stacking.fit(X, y);
const predictions = stacking.predict([[3.5, 4.5], [7.5, 8.5]]);
console.log(predictions); // [0, 1]
const probabilities = stacking.predictProba([[3.5, 4.5]]);
console.log(probabilities); // [[0.85, 0.15]]
With Passthrough
import {
StackingClassifier,
RandomForestClassifier,
GradientBoostingClassifier,
LogisticRegression
} from 'scikitjs';
const X = [
[1, 2, 3], [2, 3, 4], [3, 4, 5], [4, 5, 6],
[5, 6, 7], [6, 7, 8], [7, 8, 9], [8, 9, 10]
];
const y = [0, 0, 0, 0, 1, 1, 1, 1];
// Include original features with meta-features
const stacking = new StackingClassifier(
[
['rf', new RandomForestClassifier({ nEstimators: 10 })],
['gb', new GradientBoostingClassifier({ nEstimators: 10 })]
],
new LogisticRegression(),
{ passthrough: true } // Original features included
);
stacking.fit(X, y);
const predictions = stacking.predict(X);
Custom Stack Method
import {
StackingClassifier,
RandomForestClassifier,
SVC,
LogisticRegression
} from 'scikitjs';
// Force use of predict instead of predictProba
const stacking = new StackingClassifier(
[
['rf', new RandomForestClassifier({ nEstimators: 10 })],
['svc', new SVC()]
],
new LogisticRegression(),
{ stackMethod: 'predict' }
);
stacking.fit(X, y);
Cross-Validation Splits
import { StackingClassifier, DecisionTreeClassifier, SVC } from 'scikitjs';
// Use 10-fold CV for meta-feature generation
const stacking = new StackingClassifier(
[
['dt1', () => new DecisionTreeClassifier({ maxDepth: 5 })],
['dt2', () => new DecisionTreeClassifier({ maxDepth: 10 })],
['svc', () => new SVC()]
],
() => new LogisticRegression(),
{ cv: 10, randomState: 42 }
);
stacking.fit(X, y);
Multi-Layer Stacking
import {
StackingClassifier,
RandomForestClassifier,
GradientBoostingClassifier,
LogisticRegression,
SVC
} from 'scikitjs';
// First layer
const layer1 = new StackingClassifier(
[
['rf', new RandomForestClassifier({ nEstimators: 10 })],
['gb', new GradientBoostingClassifier({ nEstimators: 10 })]
],
new LogisticRegression()
);
// Second layer
const layer2 = new StackingClassifier(
[
['layer1', layer1],
['svc', new SVC({ probability: true })]
],
new LogisticRegression()
);
layer2.fit(X, y);
const predictions = layer2.predict(XTest);
Heterogeneous Ensemble
import {
StackingClassifier,
LogisticRegression,
DecisionTreeClassifier,
KNeighborsClassifier,
GaussianNB,
SVC
} from 'scikitjs';
const X = [
[1, 2], [2, 3], [3, 4], [4, 5],
[5, 6], [6, 7], [7, 8], [8, 9]
];
const y = [0, 0, 0, 0, 1, 1, 1, 1];
// Combine diverse base learners
const stacking = new StackingClassifier(
[
['lr', new LogisticRegression()],
['dt', new DecisionTreeClassifier()],
['knn', new KNeighborsClassifier({ nNeighbors: 3 })],
['nb', new GaussianNB()],
['svc', new SVC({ probability: true })]
],
new RandomForestClassifier({ nEstimators: 50 })
);
stacking.fit(X, y);
console.log('Training score:', stacking.score(X, y));
Compare with Voting
import {
StackingClassifier,
VotingClassifier,
RandomForestClassifier,
GradientBoostingClassifier,
LogisticRegression
} from 'scikitjs';
const estimators = [
['rf', () => new RandomForestClassifier({ nEstimators: 10 })],
['gb', () => new GradientBoostingClassifier({ nEstimators: 10 })]
];
// Voting ensemble
const voting = new VotingClassifier(estimators, { voting: 'soft' });
voting.fit(XTrain, yTrain);
console.log('Voting score:', voting.score(XTest, yTest));
// Stacking ensemble
const stacking = new StackingClassifier(
estimators,
new LogisticRegression()
);
stacking.fit(XTrain, yTrain);
console.log('Stacking score:', stacking.score(XTest, yTest));
Feature Importance Analysis
import {
StackingClassifier,
RandomForestClassifier,
GradientBoostingClassifier,
LogisticRegression
} from 'scikitjs';
const stacking = new StackingClassifier(
[
['rf', new RandomForestClassifier({ nEstimators: 10 })],
['gb', new GradientBoostingClassifier({ nEstimators: 10 })]
],
new LogisticRegression(),
{ passthrough: true }
);
stacking.fit(X, y);
// Access base estimators
for (const [name, estimator] of stacking.estimators_) {
if (estimator.featureImportances_) {
console.log(`${name} importance:`, estimator.featureImportances_);
}
}
// Final estimator coefficients
const finalEst = stacking.finalEstimator_;
if (finalEst.coef_) {
console.log('Meta-learner coefficients:', finalEst.coef_);
}