integratedBrierScore: Integrated Brier Score
In jatherrien/largeRCRF: Large Random Competing Risks Forests

Description Usage Arguments Value References Examples

View source: R/cr_integratedBrierScore.R

Used to calculate the Integrated Brier Score, which for the competing risks setting is the integral of the squared difference between each observed cumulative incidence function (CIF) for each observation and the corresponding predicted CIF. If the survivor function (1 - CDF) of the censoring distribution is provided, weights can be calculated to account for the censoring.

1 2	integratedBrierScore(responses, predictions, event, time, censoringDistribution = NULL, parallel = TRUE)

`responses`	A list of responses corresponding to the provided mortalities; use `CR_Response`.
`predictions`	The predictions to be tested against.
`event`	The event type for the error to be calculated on.
`time`	`time` specifies the upper bound of the integral.
`censoringDistribution`	Optional; if provided then weights are calculated on the errors. There are three ways to provide it - If you have all the censor times and just want to use a simple empirical estimate of the distribution, just provide a numeric vector of all of the censor times and it will be automatically calculated. You can directly specify the survivor function by providing a list with two numeric vectors called `x` and `y`. They should be of the same length and correspond to each point. It is assumed that previous to the first value in `y` the `y` value is 1.0; and that the function you provide is a right-continuous step function. You can provide a function from `stepfun`. Note that this only supports functions where `right = FALSE` (default), and that the first y value (corresponding to y before the first x value) will be to set to 1.0 regardless of what is specified.
`parallel`	A logical indicating whether multiple cores should be utilized when calculating the error. Available as an option because it's been observed that using Java's `parallelStream` can be unstable on some systems. Default value is `TRUE`; only set to `FALSE` if you get strange errors while predicting.

A numeric vector of the Integrated Brier Score for each prediction.

Section 4.2 of Ishwaran H, Gerds TA, Kogalur UB, Moore RD, Gange SJ, Lau BM (2014). “Random Survival Forests for Competing Risks.” Biostatistics, 15(4), 757–773. doi:10.1093/ biostatistics/kxu010.

data <- data.frame(delta=c(1,1,0,0,2,2), T=1:6, x=1:6)

model <- train(CR_Response(delta, T) ~ x, data, ntree=100, numberOfSplits=0, mtry=1, nodeSize=1)

newData <- data.frame(delta=c(1,0,2,1,0,2), T=1:6, x=1:6)
predictions <- predict(model, newData)

scores <- integratedBrierScore(CR_Response(data$delta, data$T), predictions, 1, 6.0)