Problems using TVals parameters in perfcurve function

Hi,

I'm trying to replicate the ROC curves produced by the "classification learner" app. My problem is a multiclass classification problem,. Im' trying to optimize the choice of a specific classifier (for the moment, I tried KNN and Trees) by looking at the ROC curves for a specific class (-2) that should get as less false negatives as possible (yet a correct false positive rate). Using KNNs, I'm willing to compute a ROC curve using perfcurve but cannot manage to get the same ROC curve as the one produced by the app.

One of the issue might be that I don't get enough values of X and Y when using perfcurve, even when setting the TVals parameters.

For example, my thres array is

thres = 0:0.001:1;

[X,Y,T,AUC] = perfcurve(partitionedModel.Y, validationScores(:,1), -2, 'TVals',thres);

but X and Y have only 10 elements. I don't understand why.

Thank you very much for your answers.