R/makeSubClassifier.R

In pcahan1/cancerCellNet: CellNet, for cancer

#' @title
#' Make Classifier tailored for subclassifier
#' @description
#' Create a random forest classifier with the transformed training data from \code{\link{query_transform}}.
#'
#' @param expTrain transformed training data from \code{\link{query_transform}}
#' @param genes vector of gene pairs from \code{\link{ptGetTop}} used as predictors
#' @param groups named vector of cells to cancer categories
#' @param nRand number of randomized profiles to make
#' @param ntrees number of trees to build
#' @param classMatrix_rand the broad classification for the random category
#' @param pairTransformedMatrix the pairtransformed matrix of the samples
#' @param stratify TRUE for stratified sampling
#' @param sampsize the stratified sample size for each category
#' @importFrom randomForest randomForest
#'
#' @return Random Forest Classifier object for sub-classifier
makeSubClassifier<-function(expTrain, genes, groups, nRand, ntrees=2000, classMatrix_rand, pairTransformedMatrix, stratify=FALSE, sampsize=40){

  randDat = ModifiedRandomize(pairTransformedMatrix, num=ncol(classMatrix_rand))
  randDat = rbind(randDat, classMatrix_rand)
  expTrain = cbind(expTrain, randDat)

  allgenes = rownames(expTrain)
  missingGenes = setdiff(unique(genes), allgenes)
  cat("Number of missing genes ", length(missingGenes),"\n")
  ggenes = intersect(unique(genes), allgenes)

  if(!stratify){
    rf = randomForest::randomForest(t(expTrain[ggenes,]), as.factor(c(groups, rep("rand", ncol(randDat)))), ntree=ntrees)
  }
  else{
    rf = randomForest::randomForest(t(expTrain[ggenes,]), as.factor(c(groups, rep("rand", ncol(randDat)))), ntree=ntrees, strata = as.factor(c(groups, rep("rand", ncol(randDat)))), sampsize=rep(sampsize, length(unique(c(groups, "rand")))))
  }

  returnList = list(tspRF = rf, namedVector = as.factor(c(groups, rep("rand", ncol(randDat)))))

  return(returnList)
}