Module preprocess 

Data preprocessing for regularized regression.

This module provides standardization utilities that match glmnet output behavior:

• Predictors are centered and scaled (if enabled)
• The intercept column is not penalized, so it’s handled specially
• Coefficients can be unstandardized back to the original scale
• Observation weights are supported for weighted regression

Standardization Convention

The scaling factor used is sqrt(sum(x²) / n), which gives unit variance under the 1/n convention (matching the glmnet paper).

Weighted Standardization

When weights are provided, they are first normalized to sum to 1: weights_normalized = w / sum(w). Then weighted means and variances are computed.

Structs

StandardizationInfo

Information stored during standardization, used to unstandardize coefficients.

StandardizeOptions

Options for standardization.

Functions

predict

Computes predictions using unstandardized coefficients.

standardize_xy

Standardizes X and y for regularized regression (glmnet-compatible).

unstandardize_coefficients

Unstandardizes coefficients from the standardized space back to original scale.