Wrapper for fitting a lasso using package glmnet.

Compatible learner wrappers for this package should have a specific format. Namely they should take as input a list called train that contains named objects $Y and $X, that contain, respectively, the outcomes and predictors in a particular training fold. Other options may be passed in to the function as well. The function must output a list with the following named objects: test_pred = predictions of test$Y based on the learner fit using train$X; train_pred = prediction of train$Y based on the learner fit using train$X; model = the fitted model (only necessary if you desire to look at this model later, not used for internal computations); train_y = a copy of train$Y; test_y = a copy of test$Y.

glmnet_wrapper(
  train,
  test,
  alpha = 1,
  nfolds = 5,
  nlambda = 100,
  use_min = TRUE,
  loss = "deviance",
  ...
)

Arguments

train

A list with named objects Y and X (see description).
test

A list with named objects Y and X (see description).
alpha

See glmnet for further description.
nfolds

See glmnet for further description.
nlambda

See glmnet for further description.
use_min

See glmnet for further description.
loss

See glmnet for further description.
...

Other options (passed to cv.glmnet)

Value

A list with named objects (see description).

Details

This particular wrapper implements glmnet. We refer readers to the original package's documentation for more details.

Examples

# load super learner package
library(glmnet)
#> Loading required package: Matrix
#> Loading required package: foreach
#> Loaded glmnet 2.0-18
# simulate data
# make list of training data
train_X <- data.frame(x1 = runif(50), x2 = runif(50))
train_Y <- rbinom(50, 1, plogis(train_X$x1))
train <- list(Y = train_Y, X = train_X)
# make list of test data
test_X <- data.frame(x1 = runif(50), x2 = runif(50))
test_Y <- rbinom(50, 1, plogis(train_X$x1))
test <- list(Y = test_Y, X = test_X)
# fit super learner 
glmnet_wrap <- glmnet_wrapper(train = train, test = test)