Nov 26, 2021
Hello. Excellent question.
In theory, we change the decision function parameters and estimate again the probabilities. And we do that for different values in order to create the ROC. This can be done for example by varying the bias in the SVM equation.
Programmatically, we do it like this: https://github.com/scikit-learn/scikit-learn/blob/0d378913be6d7e485b792ea36e9268be31ed52d0/sklearn/svm/_base.py#L499
and here is an example i made for you:
