Skip to contents

Cross-validated MLX elastic net regression

Source: R/mlxs-cv-glmnet.R
mlxs_cv_glmnet.Rd

Cross-validation wrapper around mlxs_glmnet() that mirrors the core glmnet::cv.glmnet() workflow for the families currently supported by mlxs_glmnet().

Usage

mlxs_cv_glmnet(
  x,
  y,
  weights = NULL,
  offset = NULL,
  lambda = NULL,
  type.measure = c("default", "mse", "deviance", "class", "mae", "auc", "C"),
  nfolds = 10,
  foldid = NULL,
  alignment = c("lambda", "fraction"),
  grouped = TRUE,
  keep = FALSE,
  parallel = FALSE,
  gamma = c(0, 0.25, 0.5, 0.75, 1),
  relax = FALSE,
  trace.it = 0,
  family = mlxs_gaussian(),
  ...
)

Arguments

x

Numeric matrix of predictors (observations in rows).

y

Numeric response vector.

weights

Optional observation weights. Currently unsupported.

offset

Optional offset. Currently unsupported.

lambda

Optional decreasing lambda sequence. If NULL, the full-data fit chooses the path and the same path is reused inside each fold.

type.measure

Loss used to score the holdout predictions.

nfolds

Number of folds.

foldid

Optional integer vector giving the fold assignment for each observation.

alignment

Alignment mode. Only "lambda" is currently supported.

grouped

Should cross-validation be aggregated fold-by-fold? Only TRUE is currently supported.

keep

Should out-of-fold predictions be stored?

parallel

Logical. Parallel refits are currently unsupported.

gamma, relax

Relaxed fits are currently unsupported.

trace.it

Progress tracing. Currently unsupported.

family

MLX-aware family object, e.g. mlxs_gaussian() or mlxs_binomial().

...

Additional arguments passed to mlxs_glmnet(), such as alpha, nlambda, lambda_min_ratio, standardize, intercept, maxit, and tol.

Value

An object of class mlxs_cv_glmnet.

Details

The full-data fit defines a master lambda path. Each fold is then refit on the same lambda values and scored on its holdout set.

Current limitations relative to glmnet::cv.glmnet():

  • only Gaussian and binomial families are supported

  • weights, offset, alignment != "lambda", grouped = FALSE, parallel = TRUE, relax = TRUE, and non-zero trace.it are not implemented

  • type.measure = "auc" and type.measure = "C" are not implemented