Hello,
I am not a machine-learning expert, so my question might be trivial and am having issues with the MLSeq package when trying to create classifiers for my RNA-seq data.
Everything runs fine when I ran code similar to the documentation, but I was trying to get the ROC curves of my models, and for that, I need the class probabilities. It works fine using methods MLSeq imported from the caret package (svm, PAM, rf, etc.).
For specific models (NBLDA, PLDA,PLDA2, voomDLDA and voomLSC), I cannot obtain the probabilities out of the function predict, the code runs but ignores the type="prob" or states that it is ignoring the 'type=prob' part. Those models are described as 'discrete' or 'voom-based' by the package, so maybe it's linked to that. Seems to be using non-linear prediction functions.
Either it's inherent to those models, in that case, my question is : is it even possible to get ROC curves from those models ? Or I'm doing something wrong and I'd like to understand what it is.
Thank you for listening, don't hesitate to ask for clarifications.