swag {swag}

Spare Wrapper AlGorithm (swag)

Description

swag is used to trains a meta-learning procedure that combines screening and wrapper methods to find a set of extremely low-dimensional attribute combinations. swag works on top of the caret package and proceeds in a forward-step manner.

Usage

swag(
  x,
  y,
  control = swagControl(),
  auto_control = TRUE,
  caret_args_dyn = NULL,
  ...
)

Arguments

x

A matrix or data.frame of attributes

y

A vector of binary response variable.

control

see swagControl

auto_control

A boolean, whether some control parameters are adjusted depending on x and y (see swagControl).

caret_args_dyn

If not null, a function that can modify arguments for train dynamically (see the details).

...

Arguments to be passed to train functions (see the details).

Details

Currently we expect the user to replace ... with the arguments one would use for train. This requires to know how to use train function. If ... is left unspecified, default values of train are used. But this might lead to unexpected results.

The function caret_args_dyn is expected to take as a first argument a list with all arguments for train and as a second argument the number of attributes (see examples in the vignette).

More specifically, swag builds and tests learners starting from very few attributes until it includes a maximal number of attributes by increasing the number of attributes at each step. Hence, for each fixed number of attributes, the algorithm tests various (randomly selected) learners and picks those with the best performance in terms of training error. Throughout, the algorithm uses the information coming from the best learners at the previous step to build and test learners in the following step. In the end, it outputs a set of strong low-dimensional learners. See Molinari et al. (2020) for more details.

Value

swag returns an object of class "swag". It is a list with the following components:

x	same as x input
y	same as y input
control	the control used (see swagControl)
CVs	a list containing cross-validation errors from all trained models
VarMat	a list containing information about which models are trained
cv_alpha	a vector of size pmax containing the cross-validation error at alpha (see swagControl)
IDs	a list containing information about trained model that performs better than corresponding cv_alpha error
args_caret	arguments used for train
args_caret_dyn	same as args_caret_dyn input

Author(s)

Gaetan Bakalli, Samuel Orso and Cesare Miglioli

References

Molinari R, Bakalli G, Guerrier S, Miglioli C, Orso S, Scaillet O (2020). “SWAG: A Wrapper Method for Sparse Learning.” Version 1: 23 June 2020, https://arxiv.org/pdf/2006.12837.pdf.