Overview
RobustScaler scales features using the median and interquartile range (IQR), making it robust to outliers. This scaler removes the median and scales the data according to the quantile range.
Constructor
new RobustScaler(options?: RobustScalerOptions)
If true, center the data before scaling by removing the median.
If true, scale the data to the interquartile range.
options.quantileRange
[number, number]
default:"[25, 75]"
Quantile range used to calculate scale. The default [25, 75] corresponds to the IQR. Values must satisfy 0 ≤ qMin < qMax ≤ 100.
Properties
The median value for each feature in the training set.
The interquartile range (or custom quantile range) for each feature in the training set.
Methods
fit
Compute the median and quantiles to be used for scaling.
Training data matrix where each row is a sample and each column is a feature.
this - The fitted scaler instance.
transform(X: Matrix): Matrix
Data matrix to transform.
Matrix - Transformed data.
fitTransform(X: Matrix): Matrix
fit(X).transform(X).
Training data matrix to fit and transform.
Matrix - Transformed data.
inverseTransform(X: Matrix): Matrix
Transformed data matrix to inverse transform.
Matrix - Original scale data.
Example
import { RobustScaler } from 'bun-scikit';
// Training data with outliers
const X = [
[1, 2],
[2, 4],
[3, 6],
[4, 8],
[100, 200] // Outlier
];
// Create and fit scaler (default: use IQR)
const scaler = new RobustScaler();
const XScaled = scaler.fitTransform(X);
console.log('Center (median):', scaler.center_);
// Output: Center (median): [3, 6]
console.log('Scale (IQR):', scaler.scale_);
// Output: Scale (IQR): [2, 4]
console.log('Scaled:', XScaled);
// Outlier is less extreme in scaled space
// Custom quantile range
const scaler2 = new RobustScaler({ quantileRange: [10, 90] });
const XScaled2 = scaler2.fitTransform(X);
// Only centering (no scaling)
const scaler3 = new RobustScaler({ withScaling: false });
const XCentered = scaler3.fitTransform(X);
console.log('Centered only:', XCentered);
// Inverse transform
const XOriginal = scaler.inverseTransform(XScaled);
console.log('Original:', XOriginal);
Example: Comparison with StandardScaler
import { RobustScaler, StandardScaler } from 'bun-scikit';
const XWithOutliers = [
[1, 1],
[2, 2],
[3, 3],
[1000, 1000] // Extreme outlier
];
// StandardScaler is affected by outliers
const stdScaler = new StandardScaler();
const XStd = stdScaler.fitTransform(XWithOutliers);
console.log('StandardScaler mean:', stdScaler.mean_);
// Output: StandardScaler mean: [251.5, 251.5] (heavily influenced by outlier)
// RobustScaler is resistant to outliers
const robustScaler = new RobustScaler();
const XRobust = robustScaler.fitTransform(XWithOutliers);
console.log('RobustScaler median:', robustScaler.center_);
// Output: RobustScaler median: [2.5, 2.5] (not influenced by outlier)
Notes
- More robust to outliers compared to StandardScaler
- Features with zero IQR (constant within the quantile range) are not scaled
- The scaler must be fitted before calling
transform() or inverseTransform()
- Input data must be finite (no NaN or Infinity values)
- Best choice when data contains outliers that should not influence scaling