Package smile.feature.transform

package smile.feature.transform

Feature transformations for preprocessing numeric data before model training.

The package provides the following transformers:

Feature transformers

Class	Technique	Scope	Typical Use Case
Scaler	Min–max scaling to [0, 1]	Column-wise	Bounded features; sensitive to outliers
WinsorScaler	Percentile-clipped min–max scaling to [0, 1]	Column-wise	Outlier-robust bounded scaling (default 5th–95th percentile)
MaxAbsScaler	Divide by maximum absolute value; range [−1, 1]	Column-wise	Preserves sparsity; suitable for sparse data
Standardizer	Zero mean, unit variance (z-score)	Column-wise	Gaussian features; distance-based models (KNN, SVM)
RobustStandardizer	Subtract median, divide by IQR	Column-wise	Outlier-robust standardization
Normalizer	Scale each row to unit L1 / L2 / L∞ norm	Row-wise (stateless)	Text classification, cosine-similarity models

All column-wise transformers implement InvertibleColumnTransform and can be composed into a pipeline via Transform.pipeline(Transform...).

Class	Description
MaxAbsScaler	Scales each feature by its maximum absolute value.
Normalizer	Normalize samples individually to unit norm.
Normalizer.Norm	Vector norm.
RobustStandardizer	Robustly standardizes numeric feature by subtracting the median and dividing by the IQR.
Scaler	Scales the numeric variables into the range [0, 1].
Standardizer	Standardizes numeric feature to 0 mean and unit variance.
WinsorScaler	Scales all numeric variables into the range [0, 1].