grpss (version 3.0.1)

predict.grpreg: Model predictions for a fitted grpreg object

Description

Similar to usual predict methods and predict.grprep in grpreg package.

Usage

"predict"(object, newdata, lambda = NULL, type = c("response", "class", "probability"), ...)

Arguments

object

A fitted "grpreg" object from grpss, or grpreg function.

newdata

Optionally, a matrix or data frame where to predict. If omits, the fitted predictors are used.

lambda

Value of the regularization parameter lambda at which predictions are requested. See details for the default.

type

The type of prediction: "response" gives the fitted values; "class" returns the predicted class for the binomial outcome; "probability" returns the predicted probabilities for the logistic regression.

...

Not used.

Value

The predicted values depending on the type.

Details

This function gives the predictions at newdata or all predictors if the argument newdata is not supplied. The default lambda for "grpreg" object is the one at which we obtain the minimum loss value, i.e., negative log-likelihood value. Typically, type = "response" is used for linear or poisson regression, and type = "class" or type = "probability" is used for logistic regression.

See Also

grpss

Examples

library(MASS)
set.seed(23)
n <- 30 # sample size
p <- 3  # number of predictors in each group
J <- 50  # group size
group <- rep(1:J,each = 3)  # group indices
X <- mvrnorm(n,seq(0,5,length.out = p*J),diag(p*J))
beta <- runif(12,-2,5)
mu <- X%*%matrix(c(beta,rep(0,p*J-12)),ncol = 1)

# linear regression with family = "gaussian"
y <-  mu + rnorm(n)

## without cross-validation
gss12 <- grpss(X,y,ncut = 10,group,select = TRUE)
predict(gss12) # fitted values
predict(gss12,lambda = 0.2) # fitted values at lambda = 0.2

# logistic regression with family = "binomial"
set.seed(23)
y1 <- rbinom(n,1,1/(1 + exp(-mu)))
gss21 <- grpss(X,y1,group, criterion = "gDC",select = TRUE,
               family = "binomial")
predict(gss21)