SuperLearner (version 2.0-29)

SL.biglasso: SL wrapper for biglasso

Description

SL wrapper for biglasso

Usage

SL.biglasso(Y, X, newX, family, obsWeights, penalty = "lasso",
  alg.logistic = "Newton", screen = "SSR", alpha = 1, nlambda = 100,
  eval.metric = "default", ncores = 1, nfolds = 5, ...)

Arguments

Y

Outcome variable

X

Training dataframe

newX

Test dataframe

family

Gaussian or binomial

obsWeights

Observation-level weights

penalty

The penalty to be applied to the model. Either "lasso" (default), "ridge", or "enet" (elastic net).

alg.logistic

The algorithm used in logistic regression. If "Newton" then the exact hessian is used (default); if "MM" then a majorization-minimization algorithm is used to set an upper-bound on the hessian matrix. This can be faster, particularly in data-larger-than-RAM case.

screen

"SSR" (default) is the sequential strong rule; "SEDPP" is the (sequential) EDPP rule. "SSR-BEDPP", "SSR-Dome", and "SSR-Slores" are our newly proposed screening rules which combine the strong rule with a safe rule (BEDPP, Dome test, or Slores rule). Among the three, the first two are for lasso-penalized linear regression, and the last one is for lasso-penalized logistic regression. "None" is to not apply a screening rule.

alpha

The elastic-net mixing parameter that controls the relative contribution from the lasso (l1) and the ridge (l2) penalty.

nlambda

The number of lambda values to check. Default is 100.

eval.metric

The evaluation metric for the cross-validated error and for choosing optimal lambda. "default" for linear regression is MSE (mean squared error), for logistic regression is misclassification error. "MAPE", for linear regression only, is the Mean Absolute Percentage Error.

ncores

The number of cores to use for parallel execution across a cluster created by the parallel package.

nfolds

The number of cross-validation folds. Default is 5.

...

Any additional arguments, not currently used.

References

Zeng Y, Breheny P (2017). biglasso: Extending Lasso Model Fitting to Big Data. https://CRAN.R-project.org/package=biglasso.

See Also

predict.SL.biglasso biglasso cv.biglasso predict.biglasso SL.glmnet

Examples

data(Boston, package = "MASS")
Y = Boston$medv
# Remove outcome from covariate dataframe.
X = Boston[, -14]

set.seed(1)

# Sample rows to speed up example.
row_subset = sample(nrow(X), 30)

# Subset rows and columns & use only 2 folds to speed up example.
sl = SuperLearner(Y[row_subset], X[row_subset, 1:2, drop = FALSE],
                  family = gaussian(), cvControl = list(V = 2),
                  SL.library = "SL.biglasso")
sl

pred = predict(sl, X)
summary(pred$pred)