ORION (version 1.0.3)

dropSets: Filter out Subcascades

Description

dropSets filters out specific cascade sets.

Usage

dropSets(
  subcascades = NULL,
  sets = NULL,
  direction = "sub",
  ordered = FALSE,
  neighborhood = "direct",
  type = "any"
)

Arguments

subcascades

A Subcascades object as it is returned by subcascades-function. The Subcascades object is made up of a list of matrices. Each matrix comprises the evaluation results of cascades of a specific length and is sorted row-wise according to the achieved minimal classwise sensitivities of the cascades (decreasing). The rownames show the class order by a character string of type '1>2>3' and the entries the sensitivity for each position of the cascade.

sets

Contains the set used for filtering. It is either a list of numeric vectors, a numeric vector, or a vector of characters representing a cascade of the following format '1>2>4'. Empty vectors are not allowed.

direction

Either 'sub','super' or 'exact', indicating whether the subset, the superset or the exact set is filtered. The 'exact' case filters exactly the cascades defined in sets. Please refer to the details section for a detailed description.

ordered

Either TRUE or FALSE indicating whether the order of the classes in sets is considered or not.

neighborhood

Either 'direct' or 'indirect' defines whether the given classes have to be direct neighbors ('direct') or there are other classes allowed inbetween ('indirect').

type

If more than one cascade is defined in sets, this parameter defines whether the filtered cascade has to fit to at least one of the sets ('any') or to all of the given sets ('all').

Value

A Subcascades object comprising the evaluated cascades and their performances. The Subcascades object is made up of a list of matrices. Each matrix comprises the evaluation results of cascades of a specific length and is sorted row-wise according to the achieved minimal classwise sensitivities of the cascades (decreasing). The rownames show the class order by a character string of type '1>2>3' and the entries the sensitivity for each position of the cascade.

Details

This function allows for filtering the Subcascades object.

If direction is set to 'exact' the parameter neighborhood is ignored and the parameter type is used as its default and cannot be changed. There has to be an exact match between the classes of the sets parameter cascades and the cascade of the Subcascades object, so not more but also not less classes are allowed. If the ordered parameter is set to TRUE also the order of the classes within the sets parameter and within the Subcascades object has be be exactly the same.

If direction is set to 'sub' the Subcascades object is filtered for subsets of the given cascades. If the type parameter is set to 'any' each cascade is considered individually otherwise all cascades are used as reference for filtering. This means that either for each cascade of the sets parameter individually or for taking all together the Subcascades object is filtered for cascades that are made up of the same classes, a subset of classes (ordered = FALSE) or a cascade, part of a cascade (ordered = TRUE), resulting in a set of cascades that might contain less classes. The parameter neighbourhood is not considered.

If direction is set to 'super' the Subcascades object is filtered for a superset of the given cascades. If the type parameter is set to 'any' each cascade is considered individually otherwise all cascades are used as reference for filtering. A superset are cascades that are made up of (parts) of the given cascades (ordered = TRUE) or classes (ordered=FALSE) but can contain also more classes. Depending on the parameter neighborhood only classes defined in the sets parameter are allowed as neighbors ('direct'). If neighborhood is set to 'indirect' the filtering is less strict and the direct neighborhood defined in sets is not considered.

See Also

dropSize, keepSize, keepSets, dropThreshold, keepThreshold

Examples

# NOT RUN {
library(TunePareto)
data(esl)
data = esl$data
labels = esl$labels
foldList = generateCVRuns(labels  = labels,
                          ntimes      = 2,
                          nfold       = 2,
                          leaveOneOut = FALSE,
                          stratified  = TRUE)
predMap = predictionMap(data, labels, foldList = foldList, 
                       classifier = tunePareto.svm(), kernel='linear')
# generate Subcascades object
subc = subcascades(predMap,thresh=0.7)

#define sets
set1 = list(c(1,2,3),c(2,3,4))
set2 = c('1>2>3','2>3>4')

# filter for the subset cascades that contain either the classes 
# {1,2,3} or {2,3,4} independent of the order, but neighbored
dropSets(subc, sets = set1, direction = 'sub', 
         ordered = FALSE, neighborhood = 'direct')
dropSets(subc, sets = set2, direction = 'sub', 
         ordered = FALSE, neighborhood = 'direct')

# filter for the superset cascades that contain either the classes 
# {1,2,3} or {2,3,4} independent of the order, but neighbored
dropSets(subc, sets = set1, direction = 'super', 
         ordered = FALSE, neighborhood = 'direct')
dropSets(subc, sets = set2, direction = 'super', 
         ordered = FALSE, neighborhood = 'direct')

# filter for the superset cascades that contain both the classes 
# {1,2,3} and {2,3,4} in exactly the given order, but allowing 
# for other classes inbetween
dropSets(subc, sets = set1, direction = 'super', 
        ordered = TRUE, neighborhood = 'indirect', type = 'all')
dropSets(subc, sets = set2, direction = 'super',
        ordered = TRUE, neighborhood = 'indirect', type = 'all')
         
# filter for the exact cascades
# sets can be a numeric list
result <- dropSets(subc, list(c(0,1,2),c(2,3,4,1)),
                   direction = 'exact', ordered=TRUE)
unlist(t(lapply(result,rownames)))
# or sets can be a character vector
result <- dropSets(subc, c('0>1>2','2>3>4>1'),
                   direction = 'exact',ordered=TRUE)
unlist(t(lapply(result,rownames)))         
# }