
Prediction error estimates
perr.Rd
Extracts prediction error estimates from peperr
objects.
Usage
perr(peperrobject,
type = c("632p", "632", "apparent", "NoInf", "resample", "nullmodel"))
Arguments
- peperrobject
peperr object obtained by call to function
peperr
.- type
"632p"
for the .632+ prediction error estimate (default),"632"
for the .632 prediction error estimate."apparent"
,"NoInf"
,"resample"
and"nullmodel"
return the apparent error, the no-information error, the mean sample error and the nullmodel fit, see Details.
Value
If type="632p"
or type="632"
: Prediction error: Matrix, with one row per complexity value.
If type="apparent"
: Apparent error of the full data set. Matrix: One row per complexity value. In case of survival response, columns correspond to evaluation timepoints, which are given in attribute addattr
.
If type="NoInf"
: No-information error of the full data set, i. e. evaluation in permuted data. Matrix: One row per complexity value. Columns correspond to evaluation timepoints, which are given in attribute addattr
.
If type="resample"
: Matrix. Mean prediction error of resampling test samples, one row per complexity value.
If type="nullmodel"
: Vector or scalar: Null model prediction error, i.e. of fit without information of covariates. In case of survival response Kaplan-Meier estimate at each time point, if response is binary logistic regression model, else not available.
Details
The .632 and the .632+ prediction error estimates are weighted combinations of the apparent error and bootstrap cross-validation error estimate, for survival data at given time points.
References
Binder, H. and Schumacher, M. (2008) Adapting prediction error estimates for biased complexity selection in high-dimensional bootstrap samples. Statistical Applications in Genetics and Molecular Biology, 7:1.
Gerds, T. and Schumacher, M. (2007) Efron-type measures of prediction error for survival analysis. Biometrics, 63, 1283–1287.
Schumacher, M. and Binder, H., and Gerds, T. (2007) Assessment of Survival Prediction Models in High-Dimensional Settings. Bioinformatics, 23, 1768-1774.
Examples
if (FALSE) { # \dontrun{
n <- 200
p <- 100
beta <- c(rep(1,10),rep(0,p-10))
x <- matrix(rnorm(n*p),n,p)
real.time <- -(log(runif(n)))/(10*exp(drop(x %*% beta)))
cens.time <- rexp(n,rate=1/10)
status <- ifelse(real.time <= cens.time,1,0)
time <- ifelse(real.time <= cens.time,real.time,cens.time)
# Example:
# Obtain prediction error estimate fitting a Cox proportional hazards model
# using CoxBoost
# through 10 bootstrap samples
# with fixed complexity 50 and 75
# and aggregate using prediction error curves
peperr.object <- peperr(response=Surv(time, status), x=x,
fit.fun=fit.CoxBoost, complexity=c(50, 75),
indices=resample.indices(n=length(time), method="sub632", sample.n=10))
# 632+ estimate for both complexity values at each time point
perr(peperr.object)
} # }