caTools (version 1.2)

LogitBoost: LogitBoost Classification Algorithm

Description

Train logitboost classification algorithm using decision stumps (one node decision trees) as weak learners.

Usage

LogitBoost(xlearn, ylearn, nIter=ncol(xlearn))

Arguments

xlearn

A matrix or data frame with training data. Rows contain samples and columns contain features

ylearn

Class labels for the training data samples. A response vector with one label for each row/component of xlearn. Can be either a factor, string or a numeric vector.

nIter

An integer, describing the number of iterations for which boosting should be run, or number of decision stumps that will be used.

Value

• An object of class "LogitBoost" including components:
• Stump
  List of decision stumps (one node decision trees) used:

  • column 1: feature numbers or each stump, or which column each stump operates on
  • column 2: threshold to be used for that column
  • column 3: bigger/smaller info: 1 means that if values in the column are above threshold than corresponding samples will be labeled aslablist[1]. Value "-1" means the opposite.

  If there are more than two classes, than several "Stumps" will be cbind'ed
• lablist
  names of each class

Details

The function was adapted from logitboost.R function written by Marcel Dettling. See references and "See Also" section. The code was modified in order to make it much faster for very large data sets. The speed-up was achieved by implementing a internal version of decision stump classifier instead of using calls to rpart. That way, some of the most time consuming operations were precomputed once, instead of performing them at each iteration. Another difference is that training and testing phases of the classification process were split into separate functions.

References

Dettling and Buhlmann (2002), Boosting for Tumor Classification of Gene Expression Data, available on the web page http://stat.ethz.ch/~dettling/boosting.html. http://www.cs.princeton.edu/~schapire/boost.html

See Also

• predict.LogitBoosthas prediction half of LogitBoost code
• logitboostfunction fromboostlibrary
• logitboostfunction fromlogitboostlibrary (not in CRAN or BioConductor but can be found athttp://stat.ethz.ch/~dettling/boosting.html) is very similar but much slower on very large datasets. It also perform optional cross-validation.

Examples

data(iris)
  Data  = iris[,-5]
  Label = iris[, 5]
  
  # basic interface
  model = LogitBoost(Data, Label, nIter=20)
  Lab   = predict(model, Data)
  Prob  = predict(model, Data, type="raw")
  t     = cbind(Lab, Prob)
  t[1:10, ]

  # two alternative call syntax
  p=predict(model,Data)
  q=predict.LogitBoost(model,Data)
  pp=p[!is.na(p)]; qq=q[!is.na(q)]
  stopifnot(pp == qq)

  # accuracy increases with nIter (at least for train set)
  table(predict(model, Data, nIter= 2), Label)
  table(predict(model, Data, nIter=10), Label)
  table(predict(model, Data),           Label)
  
  # example of spliting the data into train and test set
  mask = sample.split(Label)
  model = LogitBoost(Data[mask,], Label[mask], nIter=10)
  table(predict(model, Data[!mask,], nIter=2), Label[!mask])
  table(predict(model, Data[!mask,]),          Label[!mask])