resultBS: Summarise the results of a bootstrapping analysis.

In bootfs: Derive Robust Feature Sets for Two or Multiclass Classification Problems

Description

Get frequencies of features and the importance graph by combining all bootstrapping runs and classification methods.

Usage

resultBS(results, DIR=NULL, vlabel.cex = 3, filter = 10, 
		useresults=1:length(results), plot=TRUE)

Arguments

results	Result object from doBS.
DIR	Output base directory. If not NULL, output is stored in this folder. Creates the following files: importance_graph_combined.pdf, adj_sorted.csv and tophits.csv.
vlabel.cex	Node label expansion factor.
filter	Edge occurrence filter. Only show edges with larger occurrence count than filter.
useresults	Vector of integers. Specify the elements of the results list to use for summarisation. Defaults to all (i.e. use results from all feature selection algorithms).
plot	Boolean. If TRUE, show the importance graph plot.

Details

Derives the ranking of features across all bootstrapping algorithms, according to how often they occur in all bootstrapping runs. resultBS is a list with one element for each classification method. In each of these elements, for every separate classification task (e.g. different outcome groups) one sub list is stored. resultBS extracts each individual classification task and makes the importance graph and adjacency matrix as described below.

For each classification task, the intersection of the selected features across all algorithms in each bootstrapping iteration is created. Occurences of the intersected features are counted to create the final ranking.

Value

A list with one element for each classification task (i.e. group classification). Each element holds the following sub elements:

allsignatures	Intersection feature sets for each bootstrap run.
allprots	Union of all intersection features.
adj	Adjacency matrix holding the occurence counts.
tophits	Top selected features, ranked according to there occurrence.

Author(s)

Christian Bender

See Also

doBS

Examples

## Not run: 
# library(bootfs)
set.seed(1234)
data <- simDataSet(nsam=30, ngen=100, sigma=2, plot=TRUE)
logX <- data$logX
groupings <- data$groupings

params <- control_params(bstr=5, maxiter=100, maxevals=50, maxRuns=30)
	
## run the bootstrapping
retBS <- doBS(logX, groupings, 
	fs.methods=c("pamr","scad","rf_boruta"),
	DIR=NULL, 
	params=params)

## create the importance graph for all methods
## and export the adjacency matrix containing the 
## numbers of occuerrences of the features, as well 
## as the top hits.
res <- resultBS(retBS, DIR=NULL, vlabel.cex = 3, filter = 0, saveres = FALSE)

## remove the created directory
system("rm -rf bs")


## End(Not run)

bootfs documentation built on May 2, 2019, 5:50 p.m.