RobustScaler

Centers and scales features using statistics that are robust to outliers: the median for centering and the interquartile range (IQR) for scaling.

Formula

$$Z_{ij} = \frac{X_{ij} - \text{median}_j}{\text{IQR}_j}$$

where:

• $\text{median}_j$ is the median of feature $j$
• $\text{IQR}_j = Q_{75}(X_j) - Q_{25}(X_j)$ is the interquartile range

Unlike StandardScaler, the median and IQR are not influenced by extreme values, so the scaling remains stable even with outliers in the data.

When to Use

• Data with outliers: When the data contains extreme values that would distort mean/variance-based scaling.
• Robust pipelines: As a preprocessing step before models sensitive to feature scales.
• The quantile_range parameter allows customizing the centering quantiles (e.g., using the 5th and 95th percentiles).

Mirrors sklearn.preprocessing.RobustScaler.

Constructor

Skigen::RobustScaler<Scalar> scaler(bool with_centering = true,
                                     bool with_scaling = true,
                                     {q_min, q_max} = {25, 75});

Parameter	Default	Description
with_centering	true	Center by median
with_scaling	true	Scale by IQR
quantile_range	{25, 75}	Quantile range for scaling

Methods

Method	Description
fit(X)	Compute median and IQR
transform(X)	Scale using median and IQR
inverse_transform(Z)	Recover original scale
transform_inplace(X)	Scale in-place

Fitted Attributes

Accessor	Type	Description
center()	RowVectorType	Per-feature median
scale()	RowVectorType	Per-feature IQR

Example

#include <Skigen/Preprocessing>

Skigen::RobustScaler scaler;
scaler.fit(X);
auto X_scaled = scaler.transform(X);