measure: Prediction error measures

In icrf: Interval Censored Recursive Forests

Prediction error measures

Description

This function measures the prediction errors including the IMSE (integrated mean squared error) of type 1 and 2, the integrated absolute error, and the supremum absolute error. When the true survival curve is unknown but the observed interval is available, IMSE is used. When the true survival curve is known, the integrated and supremum absolute errors are used.

Usage

measure(
  surv.hat,
  timepoints,
  tau,
  method = c("all", "imse", "int.error"),
  L = NULL,
  R = NULL,
  surv.true = NULL
)

Arguments

surv.hat	the estimated survival curve matrix with rows representing the observations and the columns representing the time points at which the survival curve is estimated.
timepoints	a vector of time points at which the survival curve is estimated.
tau	the study end time. ([0, tau] is the window for the analysis.)
method	Which measure will be used? Either imse, int.error (int.error returns both integrated and supremum absolute errors), or all (both) should be entered.
L, R	the left and right interval endpoints. These are required when method == "imse" or "all".
surv.true	the true survival curve matrix with rows representing the observations and the columns representing the time points at which the survival curve is evaluated. This is required when method == "int.error" or "all".

Details

For details of the error measures, see Cho H., Jewell N. J., and Kosorok M. R. (2020+). "Interval censored recursive forest"

Value

A vector of prediction errors:

• imse.type1 and imse.type2 when method == "imse"

• int.error and sup.error when method == "int.error"

• imse.type1, imse.type2, int.error, and sup.error when method == "all"

Author(s)

Hunyong Cho [e-mail], based on the code and the documents of randomForest by Andy Liaw and Matthew Wiener.

References

Cho H., Jewell N. J., and Kosorok M. R. (2020+). "Interval censored recursive forests"

Examples

# rats data example.
# Note that this is a toy example. Use a larger ntree and nfold in practice.
library(survival)  # for Surv()
data(rat2)
L = ifelse(rat2$tumor, 0, rat2$survtime)
R = ifelse(rat2$tumor, rat2$survtime, Inf)

set.seed(1)
rats.icrf <-
  icrf(Surv(L, R, type = "interval2") ~ dose.lvl + weight + male + cage.no,
       data = rat2, ntree = 10, nfold = 3)

measure(rats.icrf$predicted.Sm, timepoints = rats.icrf$time.points,
        tau = rats.icrf$tau, method = "imse", L = L, R = R)

icrf documentation built on Oct. 30, 2022, 1:05 a.m.