predict.grpreg: Model predictions based on a fitted `grpreg` object

Description

Similar to other predict methods, this function returns predictions from a fitted "grpreg" object.

Usage

# S3 method for grpreg
predict(object, X, type=c("link", "response", "class",
"coefficients", "vars", "groups", "nvars", "ngroups", "norm"), lambda,
which=1:length(object$lambda), ...) 
# S3 method for grpreg
coef(object, lambda, which=1:length(object$lambda),
drop=TRUE, ...)
# S3 method for cv.grpreg
predict(object, X, lambda=object$lambda.min,
which=object$min, type=c("link", "response", "class", "coefficients",
"vars", "groups", "nvars", "ngroups", "norm"), ...) 
# S3 method for cv.grpreg
coef(object, lambda=object$lambda.min,
which=object$min, ...)

Arguments

object

Fitted "grpreg" or "cv.grpreg" model object.

Matrix of values at which predictions are to be made. Not used for type="coefficients"

lambda

Values of the regularization parameter lambda at which predictions are requested. For values of lambda not in the sequence of fitted models, linear interpolation is used.

which

Indices of the penalty parameter lambda at which predictions are required. By default, all indices are returned. If lambda is specified, this will override which.

type

Type of prediction: "link" returns the linear predictors; "response" gives the fitted values; "class" returns the binomial outcome with the highest probability; "coefficients" returns the coefficients; "vars" returns the indices for the nonzero coefficients; "groups" returns the indices for the groups with at least one nonzero coefficient; "nvars" returns the number of nonzero coefficients; "ngroups" returns the number of groups with at least one nonzero coefficient; "norm" returns the L2 norm of the coefficients in each group.

drop

By default, if a single value of lambda is supplied, a vector of coefficients is returned. Set drop=FALSE if you wish to have coef always return a matrix (see drop).

…

Not used.

Value

The object returned depends on type.

Details

coef and predict methods are provided for "cv.grpreg" options as a convenience. They simply call coef.grpreg and predict.grpreg with lambda set to the value that minimizes the cross-validation error.

Examples

Run this code

# NOT RUN {
# Fit penalized logistic regression model to birthweight data
data(Birthwt)
X <- Birthwt$X
y <- Birthwt$low
group <- Birthwt$group
fit <- grpreg(X, y, group, penalty="grLasso", family="binomial")

# Coef and predict methods
coef(fit, lambda=.001)
predict(fit, X, type="link", lambda=.07)[1:10]
predict(fit, X, type="response", lambda=.07)[1:10]
predict(fit, X, type="class", lambda=.01)[1:15]
predict(fit, type="vars", lambda=.07)
predict(fit, type="groups", lambda=.07)
predict(fit, type="norm", lambda=.07)

# Coef and predict methods for cross-validation
cvfit <- cv.grpreg(X, y, group, family="binomial", penalty="grMCP")
coef(cvfit)
predict(cvfit, X)[1:10]
predict(cvfit, X, type="response")[1:10]
predict(cvfit, type="groups")
# }