preprocomb: the MAIN function of programmatic use.

In preprocomb: Tools for Preprocessing Combinations

Description

preprocomb executes the computation of classification accuracy, hopkins statistic and ORH outlier score. An alternative to preprocomb is to use package 'metaheur' for faster finding of near-optimal combinations.

Usage

preprocomb(models = "rpart", gridclassobject, nholdout = 2,
  searchmethod = "exhaustive", predict = TRUE, cluster = FALSE,
  outlier = FALSE, cores = 1)

Arguments

models	(character) vector of models (names of models as defined in package caret), defaults to "rpart"
gridclassobject	(GridClass) object representing the grid of combinations
nholdout	(integer) number of holdout rounds for predictive classification, must be two or more, defaults to two
searchmethod	(character) defaults to "exhaustive" full blind search, "random" search 20 percent of grid, "grid" grid search 10 percent
predict	(boolean) compute predictions, defaults to TRUE
cluster	(boolean) compute clustering tendency, defaults to FALSE
outlier	(boolean) compute outlier tendency, defaults to FALSE
cores	(integer) number of cores used in parallel processing of holdout rounds, defaults to 1

Details

caret messages will be displayed during processing

Value

a PreProCombClass object

Examples

## modifiediris <- droplevels(iris[-c(1:60),])
## grid <- setgrid(phases=c("outliers", "scaling"), data=modifiediris)
## library(kernlab)
## result <- preprocomb(models=c("svmRadial"), grid=grid, nholdout=1, search="grid")
## result@allclassification
## result@allclustering
## result@alloutliers
## result@rawall
## result@catclassification
##
## newphases <- c("outliers", "smoothing", "scaling", "selection", "sampling")
## newmodels <- c("knn", "rf", "svmRadial")
## grid1 <- setgrid(phases=newphases, data=modifiediris)
## result1 <- preprocomb(models=newmodels, grid=grid1, nholdout=1, search="grid")

preprocomb documentation built on May 2, 2019, 3:23 a.m.