Problems using TVals parameters in perfcurve function

Hi,

I’m failing to replicate the ROC curves produced by the “Classification Learner” app. My problem is a multiclass classification problem with 5 classes. My final goal is to optimize the choice of a specific classifier based on the false negative rate of a particular class (-2). Using KNNs as a first choice, I’m trying to replicate the ROC curve produced for positive class -2 in the app, but if fails as my ROC curve seems to have much less points than the one from the app.

I tried to set manually the thresholds using the TVals parameter of the perfcurve function, but cannot understand the size of the X, Y, T outputs of the function:

thres = 0:0.001:1;

[X,Y,T,AUC] = perfcurve(partitionedModel.Y,posClassPosteriors,-2, 'TVals',thres);

length(T)

ans=

10

Can anyone explain this to me?

Thank you very much in advance !