isle_post: Importance sampled learning ensemble

Description

Uses glmnet or cv.glmnet to fit the entire LASSO path for post-processing the individual trees of a tree-based ensemble (e.g., a random forest).

isle_post(
  X,
  y,
  newX = NULL,
  newy = NULL,
  cv = FALSE,
  nfolds = 5,
  family = NULL,
  loss = "default",
  offset = NULL,
  ...
)

A list with two components:

results: A data frame with one row for each value of lambda in the coefficient path and columns giving the corresponding number of trees/non-zero coefficients, error metric(s), and the corresponding value of lambda.
lasso.fit: The fitted glmnet or cv.glmnet object.

X: A matrix of training predictions, one column for each tree in the ensemble.
y: Vector of training response values. See glmnet for acceptable values (e.g., numeric for family = "gaussian").
newX: Same as argument X, but should correspond to an independent test set. (Required whenever cv = FALSE.)
newy: Same as argument y, but should correspond to an independent test set. (Required whenever cv = FALSE.)
cv: Logical indicating whether or not to use n-fold cross-validation. Default is FALSE (Must be TRUE whenever newX = NULL and newy = NULL.)
nfolds: Integer specifying the number of folds to use for cross-validation (i.e., whenever cv = TRUE). Default is FALSE.
family: The model fitting family (e.g., family = "binomial" for binary outcomes); see glmnet for details on acceptable values.
loss: Optional character string specifying the loss to use for n-fold cross-validation. Default is "default"; see cv.glmnet for details. (Only used when cv = TRUE.)
offset: Optional value for the offset. Default is NULL, which corresponds to no offset.
...: Additional (optional) arguments to be passed on to glmnet (e.g., intercept = FALSE).