parse_alphalist: parse_alphalist

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

parse_alphalist

Description

Check / standardise a numeric vector of alpha values for a multi-alpha glmnet model.

Usage

parse_alphalist(alphalist, stripNA = FALSE, dedupe = FALSE)

Arguments

alphalist	a numeric of alpha values.
stripNA	missing values either throw an error or are discarded.
dedupe	duplicate alpha values either throw an error or are discarded.

Details

• Very small alphas are typically slow to fit.

• At last time of testing alpha=0.0 (i.e. a pure L2 / ridge penalty) did not work due to a bug related to using fixed folds in glmnet. As a workaround Zeros are replaced with a small non-zero value. In effect different software is required to get results for a 'pure' L2/ridge regression model.

Value

a named numeric vector of, de-duplicated, alpha values between (0, 1] - i.e. excluding zero.

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