CoxBoost (version 1.4)

predict.iCoxBoost: Predict method for iCoxBoost fits

Description

Obtains predictions at specified boosting steps from a iCoxBoost object fitted by iCoxBoost.

Usage

"predict"(object,newdata=NULL, subset=NULL,at.step=NULL,times=NULL, type=c("lp","logplik","risk","CIF"),...)

Arguments

object

fitted CoxBoost object from a CoxBoost call.

newdata

data frame with new covariate values (for n.new observations). If just prediction for the training data is wanted, it can be omitted. If the predictive partial log-likelihood is wanted (type=logplik), this frame also has to contain the response information.

subset

an optional vector specifying a subset of observations to be used for evaluation.

at.step

scalar or vector of boosting step(s) at which prediction is wanted. If type="risk" is used, only one step is admissible. If no step is given, the final boosting step is used.

times

vector with T time points where prediction is wanted. Only needed for type="risk"

type

type of prediction to be returned: "lp" gives the linear predictor, "logplik" the partial log-likelihood, "risk" the predicted probability of not yet having had the event at the time points given in times, and "CIF" the predicted cumulative incidence function, i.e., the predicted probability of having had the event of interest.

...

miscellaneous arguments, none of which is used at the moment.

Value

For type="lp" and type="logplik" a vector of length n.new (at.step being a scalar) or a n.new * length(at.step) matrix (at.step being a vector) with predictions is returned. For type="risk" or type="CIF" a n.new * T matrix with predicted probabilities at the specific time points is returned.

Examples

n <- 200; p <- 100
beta <- c(rep(1,2),rep(0,p-2))
x <- matrix(rnorm(n*p),n,p)
actual.data <- as.data.frame(x)
real.time <- -(log(runif(n)))/(10*exp(drop(x %*% beta)))
cens.time <- rexp(n,rate=1/10)
actual.data$status <- ifelse(real.time <= cens.time,1,0)
actual.data$time <- ifelse(real.time <= cens.time,real.time,cens.time)

#   define training and test set

train.index <- 1:100
test.index <- 101:200

#   Fit a Cox proportional hazards model by iCoxBoost

cbfit <- iCoxBoost(Surv(time,status) ~ .,data=actual.data[train.index,],
				   stepno=300,cv=FALSE)

#   mean partial log-likelihood for test set in every boosting step

step.logplik <- predict(cbfit,newdata=actual.data[test.index,],
                        at.step=0:300,type="logplik")

plot(step.logplik)

#   names of covariates with non-zero coefficients at boosting step
#   with maximal test set partial log-likelihood

print(coef(cbfit,at.step=which.max(step.logplik)-1))