glmauc.greed: GAS

In BioStaCs/BSCgas: Greedy AUC stepwise

Description

A novel Greedy AUC-maximize Strategy

Usage

glmauc.greed(data.train, data.test = NA, varlist, terms, MaxVar = 8, auc.inf0 = rep(0, MaxVar), varlist.in = NULL, trace.info = T)

Arguments

data.train	Train Data which is needed to be a data-frame.
data.test	If NA, the output is trained data, otherwise, one should offer test data, which is of the same form of train data.
varlist	Features that are used in this model.
terms	Features that are used in this model.
MaxVar	Max number of features that are allowed.
auc.inf0	trace info.
varlist.in	trace info.
trace.info	trace info.

Details

GAS

Value

term
model
auc.train

Note

The license is BSD

Author(s)

Yifan Yang

References

http://sweb.uky.edu/~yya234/wordpress

See Also

GAS

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 (data.train, data.test = NA, varlist, terms, MaxVar = 8, 
    auc.inf0 = rep(0, MaxVar), varlist.in = NULL, trace.info = T) 
{
    tic <- proc.time()
    L <- length(varlist)
    trace.matrix <- matrix(0, 8, L)
    ii = 1
    while (ii <= MaxVar) {
        auc.list <- rep(0, length(varlist))
        for (i in varlist) {
            model.term <- paste.my(terms[c(i, varlist.in)])
            model.parse <- paste("glm(lable~", model.term, ",data=data.train,family=binomial('probit'))", 
                sep = " ")
            model.tmp <- eval(parse(text = model.parse))
            pred <- logit(predict.glm(model.tmp, data.train[, 
                -16], type = "link"))
            auc.list[i] <- auc(pred, data.train$lable)
            if (trace.info) 
                cat(ii, "-th Turn, testing var+=", terms[i], 
                  "    AUC=", auc.list[i], "\n")
        }
        auc.inf0[ii] <- max(auc.list)
        if (ii > 1) {
            if (auc.inf0[ii] > auc.inf0[ii - 1]) {
                varlist.in <- c(varlist.in, varlist[which.max(auc.list)])
                trace.matrix[ii, ] <- sort(auc.list, index.return = T)$ix
                varlist <- varlist[!(varlist == which.max(auc.list))]
                ii = ii + 1
            }
            else {
                if (trace.info) 
                  cat(ii, "-th Turn, testing var --\n")
                ii = ii - 1
                pop.ind <- trace.matrix[ii, 1]
                trace.matrix[ii, ] = c(trace.matrix[ii, 2:L], 
                  0)
                if (trace.matrix[ii, 1] == 0) 
                  break
                varlist.in <- c(varlist.in[1:ii], trace.matrix[ii, 
                  1])
                varlist <- c(varlist, pop.ind)
                ii = ii + 1
            }
        }
        else {
            varlist.in <- c(varlist.in, varlist[which.max(auc.list)])
            trace.matrix[ii, 1:(L + 1 - ii)] <- sort(auc.list, 
                index.return = T)$ix
            varlist <- varlist[!(varlist == which.max(auc.list))]
            ii = ii + 1
        }
    }
    model.term <- paste.my(terms[varlist.in])
    model.parse <- paste("glm(lable~", model.term, ",data=data.train,family=binomial('probit'))", 
        sep = " ")
    model.final <- eval(parse(text = model.parse))
    if (!is.na(data.test)) {
        pred <- logit(predict.glm(model.final, data.test[, -16], 
            type = "link"))
        auc.test <- auc(pred, data.test$lable, T)
        cat("AUC on test set is", auc.inf0[MaxVar], "\nAUC on test set is:", 
            auc.test, "\n")
        toc <- proc.time()
        toc - tic
        return(list(term = model.term, model = model.final, prediction = pred, 
            auc.train = auc.inf0, auc.test = auc.test))
    }
    else {
        return(list(term = model.term, model = model.final, prediction = pred, 
            auc.train = auc.inf0))
    }
  }

BioStaCs/BSCgas documentation built on May 6, 2019, 12:06 a.m.