methods: (Generic) Functions for cfboost Objects

In CoxFlexBoost: Boosting Flexible Cox Models (with Time-Varying Effects)

Description

Functions to print or plot objects from cfboost. Further functions to extract the coefficients are specified. The extract function allows to reduce the model to an earlier boosting iteration, as e.g., the optimal stopping iteration computed with mstop.

Usage

## S3 method for class 'cfboost'
print(x, ...)
## S3 method for class 'cfboost'
summary(object, ...)
## S3 method for class 'cfboost'
plot(x, which = NULL, ask = TRUE && dev.interactive(),
     type = "b", ylab = expression(f[partial]), add_rug = TRUE,
     color.palette = c("heat.colors", "terrain.colors", "topo.colors",
     "cm.colors", "rainbow", "none"), ...)
## S3 method for class 'cfboost'
coef(object, ...)
## S3 method for class 'cfboost'
object[i, ...]

Arguments

x	an object of class cfboost
object	an object of class cfboost
which	integer vector. Only the plots for the given base-learners are plotted. Per default, only the selected base-learners are plotted.
ask	logic. If true a user input is required between each plot, see par(ask=.).
type	type of plot to be drawn. See plot for more details.
ylab	A title for the y axis.
add_rug	logic. Determines if rugs are added.
color.palette	character. Determines how time-varying effects of non-binary covariates should be ploted. A color palette of "heat.colors", "terrain.colors", "topo.colors", "cm.colors", "rainbow" is specified (as character). See rainbow for details. If "none" is specified, all observations are printed in black.
i	integer. Index specifying the model to extract. See example for more details.
...	additional arguments (not used a.t.m.)

Details

The function print prints basic information about the model specified by object and returns the argument object invisible. summary is a wrapper to the print function put gives additional information on the frequency of selection of the base-learners (see freq.sel).

plot gives a simple plotting interface for the estimated effects of the model object.

The function coef extracts the coefficients from the model object.

object[i] extracts the model from the i-th boosting iteration.

Value

object[i] returns the cfboost model-object from the i-th iteration.

See Also

cfboost for model fitting. freq.sel for the selection frequencies of base-learners.

Examples

## a simple example

#################
# simulate data #
#################

set.seed(4321)

## sample covariates first
X <- matrix(NA, nrow=400, ncol=3)
X[,1] <- runif(400, -1, 1)
X[,2] <- runif(400, -1, 1)
X[,3] <- runif(400, -1, 1)

## time-constant hazard rate
lambda <- function(time, x){
  exp(0.1 * time - 0.7 * x[1] + 0.5 * x[2])
}

## specify censoring function
cens_fct <- function(time, mean_cens){
  censor_time <- rexp(n = length(time), rate = 1/mean_cens)
  event <- (time <= censor_time)
  t_obs <- apply(cbind(time, censor_time), 1, min)
  return(cbind(t_obs, event))
}
data <- rSurvTime(lambda, X, cens_fct, mean_cens = 5)

##################
# estimate model #
##################

ctrl <- boost_control( mstop = 80, risk="oobag")
weights <- c(rep(1, 300), rep(0, 100))

## non-censored observations (in-bag)
sum(data$event[weights==1])/nrow(data[weights==1,])

## fit (a simple) model
model <- cfboost(Surv(time, event) ~ bolsTime(time) + bols(x.1) + bols(x.2) + bols(x.3),
                 control = ctrl, data = data, weights = weights)

#########################
# processing the output #
#########################

## estimate optimal mstop
(stop.opt <- mstop(model))
model_opt <- model[stop.opt]

model
## summary for mstop = 100
summary(model)
## summary (with "optimal" mstop)
summary(model_opt)

## plot of baseline hazard
plot(model_opt, which=1)
## plot of x.1 und x.2
par(mfrow=c(1,2))
plot(model_opt, which = c(2,3), ask = FALSE)

## extract coefficients
coef(model_opt)
## almost correct estimations

CoxFlexBoost documentation built on May 2, 2019, 6:53 p.m.