R/buildBalancedModelData.R

In adamkc/DIMEChelper: Companion functions for DIMEC

#' Build Balanced Model Data
#'
#' This function takes a starting folder "Original" and splits the classes
#'  within that folder into Training, Validation, and Testing data.  The data is
#'  balanced so that the smallestclass dictates how much training data is added
#'   for each class.  The remainders are added to the validation dataset.
#'
#' @param data.folder Top directory that contains "Original" folder.
#'  "Model Data" will be added to this directory as well.
#' @param trainpercent Percent of data to add to the training dataset. These
#'  three need to add to 1.
#' @param validatepercent Percent of data to add to the validation dataset.
#'  These three need to add to 1.
#' @param testpercent Percent of data to add to the test dataset. These three
#'  need to add to 1.
#'
#' @return Builds a balanced dataset from the Original folder.  To be replaced
#'  eventually by buildModelData with balancing logical argument.
#' @export
buildBalancedModelData <- function(data.folder,
                                   trainpercent = 0.80,
                                   validatepercent = 0.10,
                                   testpercent = 0.10){
  # oldwd <- data.folder
  # setwd(paste0(data.folder,"/", data.folder))
  sapply(c("Model Data","Model Data/training","Model Data/validation",
           "Model Data/testing","Model Data/testingExtra",
           "Model Data/testing/Unclassified",
           "Model Data/testingExtra/Unclassified")
         ,dir.create)
  classlist <- list.files(path = paste0(data.folder, "/Original"))
  minClass <- min(sapply(paste0(data.folder, "/Original/",classlist),function(x)
    length(list.files(x))))
  sapply(classlist,function(x) dir.create(paste0("Model Data/training/",x)))
  sapply(classlist,function(x) dir.create(paste0("Model Data/Validation/",x)))
  labelandmoveoneclass <- function(targetclass = classlist[4]){
    classFolder <- paste0(data.folder,"/Original/",targetclass)
    classImages <- list.files(classFolder)
    trainingImages <- sample(classImages,
                             size = round(minClass*trainpercent,0),
                             replace = FALSE)
    valPool <- classImages[!(classImages %in% trainingImages)]
    validationImages <- sample(valPool,
                               size = round(minClass*validatepercent,0),
                               replace = FALSE)
    testPool <- valPool[!(valPool %in% validationImages)]
    testImages <- sample(testPool,
                         size=round(minClass*testpercent,0),
                         replace = FALSE)
    testExtraImages <- testPool[!(testPool %in% testImages)]

    file.copy(paste0(classFolder,"/",trainingImages),
              paste0(data.folder,"/Model Data/training/",targetclass))
    file.copy(paste0(classFolder,"/",validationImages),
              paste0(data.folder,"/Model Data/validation/",targetclass))
    file.copy(paste0(classFolder,"/",testImages),
              paste0(data.folder,"/Model Data/testing/Unclassified"))
    file.copy(paste0(classFolder,"/",testExtraImages),
              paste0(data.folder,"/Model Data/testingExtra/Unclassified"))
    print(paste0(targetclass, " copying is complete with ",
                 length(trainingImages),
                 " training images and ",length(testImages),
                 " test images"))
  }
  sapply(classlist,labelandmoveoneclass)
  # file.rename(paste0(data.folder,"Model Data"),
  #             paste0(data.folder,"Model Data/",Sys.Date()))
  # setwd(oldwd)
}