Auroc function for ROC curve

Hi everyone,

I have a question in regard the aucroc function.
In TCGA example:

auc.splsda = auroc(final.diablo.model, roc.block = "miRNA", 
                   roc.comp = 2, print = FALSE)

In this example, the function is using component 1 and 2 for building the ROC curve.
On the other hand, if I changing roc.comp = 1, then the curve is built by using only component 1. My understanding is that by using any number higher than 1, the ROC curve is being built by adding the components.
Then, I don’t know if this make any sense, is there any way I can build a ROC curve for an individual component other than 1?


Hi @asegurac,

Our PLS models are built iteratively (i.e component 2 depends on component 1) so it would not make sense to plot individual components, except for comp 1.

We prefer to look at the error rate, how it decreases when the number of components are added. You can also have a look from the perf() outputs at how the samples are classified as new components are added. But fundamentally you always take into account what you have learnt previously (i.e the previous component).