Skip to contents

softridge 0.3.0

  • Added opt-in training least-squares rescaling to hr() and cv_hr(). Numerical path, scale and CV score calculations use RcppArmadillo.
  • Added fixed positive scale arguments and oracle scale information to exact risk and SURE routines. These exact formulas apply to fixed tuning and a fixed scale, not a scale estimated from the same response.
  • Preserved the default unscaled estimator and fold-local threshold fractions.
  • Retained rank-aware minimum-norm least-squares refitting and documented explicit absolute grids for independent implementation comparison.

Shared GitHub documentation (2026-10-25)

  • Added a common entry point for softridge and scaleCorrection, installation instructions, naming map, worked examples and reproduction guidance.
  • Added software citations and package-specific reference documentation.
  • Preserved numerical functions, package APIs and historical reproduction engines.