TrainROC: Generates ROC from train and test sets.

In bvnlab/SCATTome: FluidgmR

Description

Generates an ROC curve based on the train and test set provided in the sample datasets or by another dataset provided by the user.

Usage

TrainROC(train, ImportantGenes, GeneNumbers = 7)

Arguments

train
ImportantGenes
GeneNumbers

Examples

##---- Should be DIRECTLY executable !! ----
##-- ==>  Define data, use random,
##--	or do  help(data=index)  for the standard data sets.

## The function is currently defined as
function (train, ImportantGenes, GeneNumbers = 7) 
{
    index <- sample.int(dim(train)[1], size = floor(0.8 * dim(train)[1]))
    dtrain <- train[index, -1]
    dtest <- train[-index, -c(1, dim(train)[2])]
    TestStatus <- train[-index, dim(train)[2]]
    library(randomForest)
    library(e1071)
    trcol <- colnames(dtrain)
    tecol <- colnames(dtest)
    SelectedSet <- ImportantGenes[1:GeneNumbers]
    trind <- c()
    tsind <- c()
    for (i in 1:length(SelectedSet)) {
        trind <- c(trind, which(trcol == SelectedSet[i]))
        tsind <- c(tsind, which(tecol == SelectedSet[i]))
    }
    dtr <- dtrain[, c(trind, dim(dtrain)[2])]
    dts <- dtest[, tsind]
    rf <- randomForest(as.factor(Status) ~ ., ntree = 1000, data = dtr)
    prf <- predict(rf, newdata = dts, type = "prob")
    prf1 <- prf[, 2]
    threshold <- seq(0, 1, 0.01)
    tp <- c()
    fp <- c()
    for (i in 1:length(threshold)) {
        temp <- ceiling(prf1 - threshold[i])
        conf <- xtabs(~temp + TestStatus)
        if (dim(conf)[1] == 1) {
            if (unique(temp) == 1) {
                tp = c(tp, 1)
                fp = c(fp, 1)
            }
            else {
                tp = c(tp, 0)
                fp = c(fp, 0)
            }
        }
        else {
            tp <- c(tp, conf[2, 2]/(conf[1, 2] + conf[2, 2]))
            fp <- c(fp, conf[2, 1]/(conf[1, 1] + conf[2, 1]))
        }
    }
    AUC_RF <- sum(tp) * 0.01
    pdf("ROC_CURVES_OF_TRAIN_SET.pdf")
    plot(fp, tp, ylab = "True Positives", xlab = "False Positives", 
        main = "Sample ROC of Random Forest", type = "l", col = 4)
    lines(c(0, 1), c(0, 1), lty = 2)
    sv <- svm(as.factor(Status) ~ ., data = dtr, kernel = "radial", 
        type = "C-classification", cross = 10, probability = TRUE)
    psv <- predict(sv, newdata = dts, probability = TRUE)
    prob <- attr(psv, "probabilities")
    prob1 <- prob[, 2]
    tp <- c()
    fp <- c()
    for (i in 1:length(threshold)) {
        temp <- ceiling(prob1 - threshold[i])
        conf <- xtabs(~temp + TestStatus)
        if (dim(conf)[1] == 1) {
            if (unique(temp) == 1) {
                tp = c(tp, 1)
                fp = c(fp, 1)
            }
            else {
                tp = c(tp, 0)
                fp = c(fp, 0)
            }
        }
        else {
            tp <- c(tp, conf[2, 2]/(conf[1, 2] + conf[2, 2]))
            fp <- c(fp, conf[2, 1]/(conf[1, 1] + conf[2, 1]))
        }
    }
    AUC_SVM1 <- sum(tp) * 0.01
    plot(fp, tp, ylab = "True Positives", xlab = "False Positives", 
        main = "Sample ROC of SVM(Gaussian Kernel)", type = "l", 
        col = 4)
    lines(c(0, 1), c(0, 1), lty = 2)
    sv1 <- svm(as.factor(Status) ~ ., data = dtr, kernel = "sigmoid", 
        type = "C-classification", cross = 10, probability = TRUE)
    psv1 <- predict(sv, newdata = dts, probability = TRUE)
    prob <- attr(psv1, "probabilities")
    prob2 <- prob[, 2]
    tp <- c()
    fp <- c()
    for (i in 1:length(threshold)) {
        temp <- ceiling(prob2 - threshold[i])
        conf <- xtabs(~temp + TestStatus)
        if (dim(conf)[1] == 1) {
            if (unique(temp) == 1) {
                tp = c(tp, 1)
                fp = c(fp, 1)
            }
            else {
                tp = c(tp, 0)
                fp = c(fp, 0)
            }
        }
        else {
            tp <- c(tp, conf[2, 2]/(conf[1, 2] + conf[2, 2]))
            fp <- c(fp, conf[2, 1]/(conf[1, 1] + conf[2, 1]))
        }
    }
    AUC_SVM2 <- sum(tp) * 0.01
    plot(fp, tp, ylab = "True Positives", xlab = "False Positives", 
        main = "Sample ROC of SVM (Sigmoidal Kernel)", type = "l", 
        col = 4)
    lines(c(0, 1), c(0, 1), lty = 2)
    ens.p = (prf1 + prob1 + prob2)/3
    tp <- c()
    fp <- c()
    for (i in 1:length(threshold)) {
        temp <- ceiling(ens.p - threshold[i])
        conf <- xtabs(~temp + TestStatus)
        if (dim(conf)[1] == 1) {
            if (unique(temp) == 1) {
                tp = c(tp, 1)
                fp = c(fp, 1)
            }
            else {
                tp = c(tp, 0)
                fp = c(fp, 0)
            }
        }
        else {
            tp <- c(tp, conf[2, 2]/(conf[1, 2] + conf[2, 2]))
            fp <- c(fp, conf[2, 1]/(conf[1, 1] + conf[2, 1]))
        }
    }
    AUC_Ens <- sum(tp) * 0.01
    plot(fp, tp, ylab = "True Positives", xlab = "False Positives", 
        main = "Sample ROC of Ensemble Models", type = "l", col = 4)
    lines(c(0, 1), c(0, 1), lty = 2)
    dev.off()
    Method <- c("Random Forest", "SVM(Gaussian Kernel)", "SVM(Sigmoidal Kernel)", 
        "Ensemble")
    AUCROC <- c(AUC_RF, AUC_SVM1, AUC_SVM2, AUC_Ens)
    AUC <- as.data.frame(cbind(Method, AUCROC))
    write.csv(AUC, "AUCROC_Train.csv", quote = FALSE, row.names = FALSE)
    return(NULL)
  }

bvnlab/SCATTome documentation built on May 13, 2019, 9:05 a.m.