glmnet.funs: Elastic net, lasso, ridge regression training and prediction...

In paolo-vergo/conformalInference.multi: Conformal Inference Tools for Regression with Multivariate Response

Elastic net, lasso, ridge regression training and prediction functions.

Description

Construct training and prediction functions for the elastic net, the lasso, or ridge regression, based on the glmnet package, over a sequence of (given or internally computed) lambda values.

Usage

elastic.funs(
  gamma = 0.5,
  standardize = TRUE,
  intercept = TRUE,
  lambda = NULL,
  nlambda = 50,
  lambda.min.ratio = 1e-04,
  cv.rule = c("min", "1se")
)

lasso.funs(
  standardize = TRUE,
  intercept = TRUE,
  lambda = NULL,
  nlambda = 50,
  lambda.min.ratio = 1e-04,
  cv.rule = c("min", "1se")
)

ridge.funs(
  standardize = TRUE,
  intercept = TRUE,
  lambda = NULL,
  nlambda = 50,
  lambda.min.ratio = 1e-04,
  cv.rule = c("min", "1se")
)

Arguments

gamma	Mixing parameter (between 0 and 1) for the elastic net, where 0 corresponds to ridge regression, and 1 to the lasso. Default is 0.5.
standardize, intercept	Should the data be standardized, and should an intercept be included? Default for both is TRUE.
lambda	Sequence of lambda values over which training is performed. This must be in decreasing order, and — this argument should be used with caution! When used, it is usually best to grab the sequence constructed by one initial call to glmnet (see examples). Default is NULL, which means that the nlambda, lambda.min.ratio arguments will define the lambda sequence (see next).
nlambda	Number of lambda values over which training is performed. In particular, the lambda sequence is defined by nlambda log-spaced values between lambda.max and lambda.min.ratio * lambda.max, where lambda.max is the smallest value of lambda at which the solution has all zero components, and lambda.min.ratio is a small fraction (see next). Default is 50.
lambda.min.ratio	Small fraction that gets used in conjunction with nlambda to specify a lambda sequence (see above). Default is 1e-4.
cv.rule	If the cv argument is TRUE, then cv.rule determines which rule should be used for the predict function, either "min" (the usual rule) or "1se" (the one-standard-error rule). See the glmnet help files for details. Default is "min".

Details

This function is based on the packages glmnet. Notice that Cross Validation to select the best lambda value is compulsory! The functions lasso.funs and ridge.funs are convenience functions, they simply call elastic.funs with gamma = 1 and gamma = 0, respectively.

Value

A list with three components: train.fun, predict.fun, active.fun. The third function is designed to take the output of train.fun, and reports which features are active for each fitted model contained in this output.

paolo-vergo/conformalInference.multi documentation built on July 4, 2023, 9:50 a.m.