Screen variable before penalized regression

Source: R/screener.R

Expands a contingency table to a data frame where each observation in the table becomes a single observation in the data frame with corresponding information for each for each combination of the table dimensions.

screen_variables(x, y, lambda = 0.1, method = c("global-strong", "global-DPP"))

Arguments

x

A table or matrix

y

A vector of outcomes

lambda

a vector of positive values used for the penalization parameter.

method

a string giving the method used for screening. Two possibilities are "global-strong" and "global-DPP"

Value

A list with three elements: lambda which contains the lambda values, selected which contains the indices of the selected variables, and method a string listing the method used.

Details

Note that no standardization is done (not necessary?)

References

Hastie, Tibshirani and Wainwright (2015). "Statistical Learning with Sparsity". CRC Press.

Author

Claus Ekstrom claus@rprimer.dk

Examples


x <- matrix(rnorm(50*100), nrow=50)
y <- rnorm(50, mean=x[,1])
screen_variables(x, y, lambda=c(.1, 1, 2))
#> $lambda
#> [1] 0.1 1.0 2.0
#> 
#> $selected
#> integer(0)
#> 
#> $method
#> [1] "global-strong"
#>