Overview
VarianceThreshold removes features whose variance doesn’t meet a threshold. Features with low variance are often not useful for machine learning models.
Constructor
new VarianceThreshold(options?)
Parameters
Variance threshold. Features with variance below this value are removed.
Methods
fit()
Compute variances for each feature.
transform(X: Matrix): Matrix
fitTransform(X: Matrix): Matrix
Properties
Variance of each feature in the training set.
Number of features seen during fit.
Indices of features that passed the variance threshold.
Examples
Remove zero-variance features
import { VarianceThreshold } from "bun-scikit";
const X = [
[0, 2, 0, 3],
[0, 1, 4, 3],
[0, 1, 1, 3]
];
const selector = new VarianceThreshold({ threshold: 0 });
const XNew = selector.fitTransform(X);
// Removes first and last columns (variance = 0)
console.log(XNew);
// [[2, 0], [1, 4], [1, 1]]
Remove low-variance features
import { VarianceThreshold } from "bun-scikit";
const X = [
[0, 0, 1],
[0, 1, 0],
[1, 0, 0],
[0, 1, 1],
[0, 1, 0],
[0, 1, 1]
];
const selector = new VarianceThreshold({ threshold: 0.2 });
const XNew = selector.fitTransform(X);
console.log("Selected features:", selector.selectedFeatureIndices_);
console.log("Variances:", selector.variances_);
Use in pipeline
import { Pipeline, VarianceThreshold, StandardScaler, LogisticRegression } from "bun-scikit";
const pipeline = new Pipeline([
["variance", new VarianceThreshold({ threshold: 0.1 })],
["scaler", new StandardScaler()],
["classifier", new LogisticRegression()]
]);
pipeline.fit(XTrain, yTrain);
const predictions = pipeline.predict(XTest);
Notes
- This is an unsupervised feature selection method
- Does not consider the relationship between features and target
- Useful as a preprocessing step to remove constant or near-constant features
- Throws an error if all features are below the threshold