preformat: Preformatting for Training with Warm Starts

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

Preformatting for Training with Warm Starts

Description

Presorts and formats training frame into a form suitable for subsequent training by rfArb caller or rfTrain command. Wraps this form to spare unnecessary recomputation when iteratively retraining, for example, under parameter sweep.

Usage

## Default S3 method:
preformat(x,
		   nThread = 0,
                   verbose=FALSE,
                   ...)

Arguments

x	the design frame expressed as either a data.frame object with numeric and/or factor columns or as a numeric or factor-valued matrix.
nThread	number of cores to run in parallel, if available.
verbose	indicates whether to output progress of preformatting.
...	unused.

Value

an object of class Deframe consisting of:

• rleFrame run-length encoded representation of class RLEFrame consisting of:

  • rankedFrame run-length encoded representation of class RankedFrame consisting of:

    • nRow the number of observations encoded.

    • runVal the run-length encoded values.

    • runRow the corresponding row indices.

    • rleHeight the number of encodings, per predictor.

    • topIdx the accumulated end index, per predictor.

  • numRanked packed representation of sorted numerical values of class NumRanked consisting of:

    • numVal distinct numerical values.

    • numHeight value offset per predictor.

  • facRanked packed representation of sorted factor values of class FacRanked consisting of:

    • facVal distinct factor values, zero-based.

    • facHeight value offset per predictor.

• nRow the number of training observations.

• signature an object of type Signature consisting of:

  • predForm predictor class names.

  • level per-predictor levels, regardless whether realized.

  • factor per-predictor realized levels.

  • colNames predictor names.

  • rowNames observation names.

Author(s)

Mark Seligman at Suiji.

Examples

  ## Not run: 
    data(iris)
    pt <- preformat(iris[,-5])

    ppTry <- seq(0.2, 0.5, by= 0.3/10)
    nIter <- length(ppTry)
    rsq <- numeric(nIter)
    for (i in 1:nIter) {
      rb <- Rborist(pt, iris[,5], predProb=ppTry[i])
      rsq[i] = rb$validiation$rsq
    }
  
## End(Not run)

Rborist documentation built on April 3, 2025, 8:04 p.m.