vimp: Variable Importance
In jatherrien/largeRCRF: Large Random Competing Risks Forests

Description Usage Arguments Value Examples

Calculate variable importance by recording the increase in error when a given predictor is randomly permuted. Regression forests uses mean squared error; competing risks uses integrated Brier score.

1
2
3

vimp(forest, newData = NULL, randomSeed = NULL, type = c("mean", "z",
  "raw"), events = NULL, time = NULL, censoringDistribution = NULL,
  eventWeights = NULL)

`forest`	The forest that was trained.
`newData`	A test set of the data if available. If not, then out of bag errors will be attempted on the training set.
`randomSeed`	The source of randomness used to permute the values. Can be left blank.
`events`	If using competing risks forest, the events that the error measure used for VIMP should be calculated on.
`time`	If using competing risks forest, the upper bound of the integrated Brier score.
`censoringDistribution`	(Optional) If using competing risks forest, the censoring distribution. See `integratedBrierScore for details.`
`eventWeights`	(Optional) If using competing risks forest, weights to be applied to the error for each of the `events`.

A named numeric vector of importance values.

data(wihs)

forest <- train(CR_Response(status, time) ~ ., wihs,
 ntree = 100, numberOfSplits = 0, mtry=3, nodeSize = 5)

vimp(forest, events = 1:2, time = 8.0)