varSelRFBoot: Bootstrap the variable selection procedure in varSelRF

Description

Use the bootstrap to estimate the prediction error rate (wuth the .632+ rule) and the stability of the variable selection procedure implemented in varSelRF.

Usage

varSelRFBoot(xdata, Class, c.sd = 1,
             mtryFactor = 1, ntree = 5000, ntreeIterat = 2000,
             vars.drop.frac = 0.2, bootnumber = 200,
             whole.range = TRUE,
             recompute.var.imp = FALSE,
             usingCluster = TRUE,
             TheCluster = NULL, srf = NULL, verbose = TRUE, ...)

Arguments

xdata

A data frame or matrix, with subjects/cases in rows and variables in columns. NAs not allowed.

Class

The dependent variable; must be a factor.

c.sd

The factor that multiplies the sd. to decide on stopping the tierations or choosing the final solution. See reference for details.

mtryFactor

The multiplication factor of $\sqrt{number.of.variables}$ for the number of variables to use for the ntry argument of randomForest.

ntree

The number of trees to use for the first forest; same as ntree for randomForest.

ntreeIterat

The number of trees to use (ntree of randomForest) for all additional forests.

vars.drop.frac

The fraction of variables, from those in the previous forest, to exclude at each iteration.

whole.range

If TRUE continue dropping variables until a forest with only two variables is built, and choose the best model from the complete series of models. If FALSE, stop the iterations if the current OOB error becomes larger than the initial OOB e

recompute.var.imp

If TRUE recompute variable importances at each new iteration.

bootnumber

The number of bootstrap samples to draw.

usingCluster

If TRUE use a cluster to parallelize the calculations.

TheCluster

The name of the cluster, if one is used.

srf

An object of class varSelRF. If used, the ntree and mtryFactor parameters are taken from this object, not from the arguments to this function. If used, it allows to skip carrying out a first iteration to build the random forest to the complete

verbose

Give more information about what is being done.

...

Not used.

Value

An object of class varSelRFBoot, which is a list with components:
number.of.bootsamplesThe number of bootstrap replicates.
bootstrap.pred.errorThe .632+ estimate of the prediction error.
leave.one.out.bootstrapThe leave-one-out estimate of the error rate (used when computing the .632+ estimate).
all.data.randomForestA random forest built from all the data, but after the variable selection. Thus, beware because the OOB error rate is severely biased down.
all.data.varsThe variables selected in the run with all the data.
all.data.runAn object of class varSelRF; the one obtained from a run of varSelRF on the original, complete, data set. See varSelRF.
class.predictionsThe out-of-bag predictions from the bootstrap, of type "response".See predict.randomForest. This is an array, with dimensions number of cases by number of bootstrap replicates.
prob.predictionsThe out-of-bag predictions from the bootstrap, of type "class probability". See predict.randomForest. This is a 3-way array; the last dimension is the bootstrap replication; for each bootstrap replication, the 2D array has dimensions case by number of classes, and each value is the probability of belonging to that class.
number.of.varsA vector with the number of variables selected for each bootstrap sample.
overlapThe "overlap" between the variables selected from the run in original sample and the variables returned from a bootstrap sample. Overlap between the sets of variables A and B is defined as $$\frac{|variables.in.A \cap variables.in.B|}{\sqrt{|variables.in.A| |variables.in.B|}}$$ or size (cardinality) of intersection between the two sets / sqrt(product of size of each set).
all.vars.in.solutionsA vector with all the genes selected in the runs on all the bootstrap samples. If the same gene is selected in several bootstrap runs, it appears multiple times in this vector.
all.solutionsEach solutions is a character vector with all the variables in a particular solution concatenated by a "+". Thus, all.solutions is a vector, with length equal to number.of.bootsamples, of the solution from each bootstrap run.
ClassThe original class argument.
allBootRunsA list of length number.of.bootsamples. Each component of this list is an element of class varSelRF and stores the results from the runs on each bootstrap sample.

Details

If a cluster is used for the calculations, it will be used for the embarrisingly parallelizable task of building as many random forests as bootstrap samples.

References

Breiman, L. (2001) Random forests. Machine Learning, 45, 5--32.

Diaz-Uriarte, R. and Alvarez de Andres, S. (2005) Variable selection from random forests: application to gene expression data. Tech. report. http://ligarto.org/rdiaz/Papers/rfVS/randomForestVarSel.html

Efron, B. & Tibshirani, R. J. (1997) Improvements on cross-validation: the .632+ bootstrap method. J. American Statistical Association, 92, 548--560. Svetnik, V., Liaw, A. , Tong, C & Wang, T. (2004) Application of Breiman's random forest to modeling structure-activity relationships of pharmaceutical molecules. Pp. 334-343 in F. Roli, J. Kittler, and T. Windeatt (eds.). Multiple Classier Systems, Fifth International Workshop, MCS 2004, Proceedings, 9-11 June 2004, Cagliari, Italy. Lecture Notes in Computer Science, vol. 3077. Berlin: Springer.

Examples

Run this code

## This is a small example, but can take some time.

x <- matrix(rnorm(25 * 30), ncol = 30)
x[1:10, 1:2] <- x[1:10, 1:2] + 2
cl <- factor(c(rep("A", 10), rep("B", 15)))  

rf.vs1 <- varSelRF(x, cl, ntree = 200, ntreeIterat = 100,
                   vars.drop.frac = 0.2)
rf.vsb <- varSelRFBoot(x, cl,
                       bootnumber = 10,
                       usingCluster = FALSE,
                       srf = rf.vs1)
rf.vsb
summary(rf.vsb)
plot(rf.vsb)

Run the code above in your browser using DataLab