dCVnet-package: dCVnet: Doubly Cross-Validated Elastic-Net Regularised...

In AndrewLawrence/dCVnet: doubly cross-validated elastic-net regularised generalised linear models

dCVnet: Doubly Cross-Validated Elastic-Net Regularised (Generalised) Linear Models

Description

dCVnet fits and cross-validates regression with elastic-net regularisation. Optimal hyperparameters (lambda & alpha) are selected by double (nested) cross-validation with model performance evaluated in an independent outer cross-validation loop.

Details

The lambda and alpha hyperparameters of the elastic-net allow models to freely range from effectively unregularised, to heavily-regularised (via lambda). With a balance of two types of regularisation: L2 (ridge) and L1 (LASSO) (alpha). This regularisation produces dimensionality reduction and variable selection in the predictors.

The values of the lambda and alpha hyperparameters (and so the amount and type of regularisation) are selected on the basis of cross-validated model performance - i.e. a data-driven approach.

Using a fully nested cross-validation (instead of reporting the CV performance of the selected hyperparameters) keeps the cross-validated performance estimates 'honest' by reducing optimism bias related to hyperparameter tuning.

Author(s)

Maintainer: Andrew J. Lawrence [e-mail]

See Also

Useful links:

• https://github.com/AndrewLawrence/dCVnet

• Report bugs at https://github.com/AndrewLawrence/dCVnet/issues

AndrewLawrence/dCVnet documentation built on Sept. 24, 2024, 5:24 a.m.