FeatureUnion combines several transformer objects into a single transformer that concatenates their outputs.
Constructor
import { FeatureUnion } from 'scikitjs';
const union = new FeatureUnion(specs, options);
specs
[string, transformer][]
required
List of (name, transformer) tuples:
- name: string identifier
- transformer: transformer object, ‘drop’, or ‘passthrough’
options.transformerWeights
Multiplicative weights for features from each transformer. Keys are transformer names.
Methods
fit
Fit all transformers.
fit(X: number[][], y?: number[], sampleWeight?: number[]): FeatureUnion
transform(X: number[][]): number[][]
fitTransform(X: number[][], y?: number[], sampleWeight?: number[]): number[][]
getParams / setParams
getParams(deep?: boolean): Record<string, unknown>
setParams(params: Record<string, unknown>): FeatureUnion
Properties
transformerList_
readonly [string, transformer][]
List of (name, transformer) tuples
namedTransformers_
Record<string, transformer>
Access transformers by name
Examples
import {
FeatureUnion,
PCA,
SelectKBest,
f_classif
} from 'scikitjs';
const X = [
[1, 2, 3, 4],
[5, 6, 7, 8],
[9, 10, 11, 12],
[13, 14, 15, 16]
];
const y = [0, 0, 1, 1];
// Combine PCA and univariate selection
const union = new FeatureUnion([
['pca', new PCA({ nComponents: 2 })],
['select', new SelectKBest({ scoreFunc: f_classif, k: 2 })]
]);
const XTransformed = union.fitTransform(X, y);
console.log(XTransformed[0].length); // 4 features (2 from PCA + 2 from SelectKBest)
Feature Weighting
import { FeatureUnion, PCA, StandardScaler } from 'scikitjs';
const X = [
[1, 2, 3],
[4, 5, 6],
[7, 8, 9]
];
// Weight PCA features more heavily
const union = new FeatureUnion(
[
['pca', new PCA({ nComponents: 2 })],
['original', 'passthrough']
],
{
transformerWeights: {
'pca': 2.0,
'original': 1.0
}
}
);
const XTransformed = union.fitTransform(X);
// PCA features are multiplied by 2.0
Multiple Feature Representations
import {
FeatureUnion,
PolynomialFeatures,
StandardScaler
} from 'scikitjs';
const X = [
[1, 2],
[3, 4],
[5, 6]
];
// Create both original and polynomial features
const union = new FeatureUnion([
['original', 'passthrough'],
['poly', new PolynomialFeatures({ degree: 2, includeBias: false })]
]);
const XTransformed = union.fitTransform(X);
console.log(XTransformed[0]);
// [original features + polynomial features]
In a Pipeline
import {
Pipeline,
FeatureUnion,
PCA,
TruncatedSVD,
LogisticRegression
} from 'scikitjs';
const X = [
[1, 2, 3, 4],
[5, 6, 7, 8],
[9, 10, 11, 12],
[13, 14, 15, 16]
];
const y = [0, 0, 1, 1];
// Combine different dimensionality reduction techniques
const union = new FeatureUnion([
['pca', new PCA({ nComponents: 2 })],
['svd', new TruncatedSVD({ nComponents: 2 })]
]);
const pipeline = new Pipeline([
['features', union],
['classifier', new LogisticRegression()]
]);
pipeline.fit(X, y);
const predictions = pipeline.predict(X);
import { FeatureUnion, PCA } from 'scikitjs';
const X = [[1, 2], [3, 4], [5, 6]];
// Conditionally include transformers
const union = new FeatureUnion([
['pca', new PCA({ nComponents: 1 })],
['skip', 'drop'], // This transformer is dropped
['original', 'passthrough']
]);
const XTransformed = union.fitTransform(X);
// Only includes PCA and passthrough features
import { FeatureUnion } from 'scikitjs';
class LogTransformer {
fit(X, y) {
return this;
}
transform(X) {
return X.map(row => row.map(val => Math.log(val + 1)));
}
}
const union = new FeatureUnion([
['log', new LogTransformer()],
['original', 'passthrough']
]);
const X = [[1, 2], [3, 4], [5, 6]];
const XTransformed = union.fitTransform(X);
// Concatenates log-transformed and original features
Ensemble Feature Engineering
import {
Pipeline,
FeatureUnion,
StandardScaler,
PCA,
SelectKBest,
PolynomialFeatures,
RandomForestClassifier
} from 'scikitjs';
const X = [
[1, 2, 3],
[4, 5, 6],
[7, 8, 9],
[10, 11, 12]
];
const y = [0, 0, 1, 1];
// Combine multiple feature engineering strategies
const featureEngineering = new FeatureUnion([
['pca', new PCA({ nComponents: 2 })],
['poly', new PolynomialFeatures({ degree: 2 })],
['select', new SelectKBest({ k: 2 })]
]);
const pipeline = new Pipeline([
['scaler', new StandardScaler()],
['features', featureEngineering],
['classifier', new RandomForestClassifier({ nEstimators: 10 })]
]);
pipeline.fit(X, y);