mc.roc: Generate ROC curves from several classifiers

In tonedivad/FIEmspro: Flow Injection Electrospray Mass Spectrometry Processing: \\ data processing, classification modelling and variable selection in metabolite fingerprinting

Description

Convenience function to generate a ROC curve from several runs and iterations for one model or a selection of models contained in a mc.acc object (see details mc.agg). This function allows the averaging of several ROC curves produced on each data partitioning of the resampling strategy (i.e cross-validation runs repeated or not several times). Three methods are available to perform the averaging: "horiz" (horizontal), "vert" (vertical) and "thres" (thresholding). (see reference for further details). The default value ("all") means that each data points from individual ROC curves are strictly concatenated into a single ROC curve.

Usage

mc.roc(mc.obj, lmod = 1, method = "all")
## Default S3 method:
mc.roc(mc.obj, lmod = 1, method = "all")

Arguments

mc.obj	mc.agg object - See details mc.agg
lmod	List of models to be considered - Default: all of them
method	Aggregation method ("all", "thres", "vert", "horiz")

Value

roc.list object is a list of two components:

roc	List of ROC curves of length equal to the number of models. For each curve true positive (tpr), false positive rate (fpr), decision boundary threshold (thres) and type (type) of ROC aggregation are given.
cldef	Identical to cldef in mc.agg.

Author(s)

David Enot [e-mail]

References

Fawcett, T. (2004). ROC graphs: notes and practical considerations for researchers. Technical Report HPL-2003-4

See Also

plot.mc.roc,mc.agg

Examples

data(iris)
dat=as.matrix(iris[,1:4])
cl=as.factor(iris[,5])
lrnd=sample(1:150)[1:50]
cl[lrnd]=sample(cl[lrnd])
pars   <- valipars(sampling = "cv",niter = 2, nreps=10)
dat1=dat.sel1(dat,cl,pwise="virginica",mclass=NULL,pars=pars)

res1=lapply(dat1,function(x) accest(x,clmeth="lda"))
res2=lapply(dat1,function(x) accest(x,clmeth="randomForest",ntree=50))
mc=mc.agg(res1,res2)

roc.sv=mc.roc(mc,lmod=1:4,method="thres")
print(roc.sv)

tonedivad/FIEmspro documentation built on May 31, 2019, 6:20 p.m.