forestWeight: Meinshausen forest weights

In Rborist: Extensible, Parallelizable Implementation of the Random Forest Algorithm

Meinshausen forest weights

Description

Normalized observation counts across a prediction set.

Usage

## Default S3 method:
forestWeight(objTrain, prediction, sampler=objTrain$sampler,
nThread=0, verbose = FALSE, ...)

Arguments

objTrain	an object of class rfArb, created from a previous invocation of the command Rborist or rfArb to train.
prediction	an object of class SummaryReg or SummaryCtg obtained from prediction using objTrain and argument indexing=TRUE.
sampler	an object of class Sampler, as documented in command of the same name.
nThread	specifies a prefered thread count.
verbose	whether to output progress of weighting.
...	not currently used.

Value

a numeric matrix having rows equal to the Meinshausen weight of each new datum.

Author(s)

Mark Seligman at Suiji.

See Also

Rborist

Examples

## Not run: 
  # Regression example:
  nRow <- 5000
  x <- data.frame(replicate(6, rnorm(nRow)))
  y <- with(x, X1^2 + sin(X2) + X3 * X4) # courtesy of S. Welling.
  rb <- Rborist(x,y)


  newdata <- data.frame(replace(6, rnorm(nRow)))

  # Performs separate prediction on new data, saving indices:
  pred <- predict(rb, newdata, indexing=TRUE)
  weights <- forestWeight(rb, pred)

  obsIdx <- 215 # Arbitrary observation index (row number)

  # Inner product should equal prediction, modulo numerical vagaries:
  yPredApprox <- weights[obsIdx,] %*% y
  print((yPredApprox - pred$yPred[obsIdx])/yPredApprox) 

## End(Not run)

Rborist documentation built on July 26, 2023, 5:32 p.m.