ROC Curve widget sets a wrong prior probability

[janezd]

With apologies to self for writing such a bad bug report: I have no time to properly explore it now, but I have to write this down lest I forget.

I encountered a situation (on Pima diabetes data) in which the ROC widget's target was set to 1, but the prior probability was that for class 0. I changed the target to 0 and back to 1, and the prior probabilty was reset properly. My hunch is that if the widget was loaded in the workflow, the target is retrieved from settings, but the prior probability is set before and disregarding the target. Changing the target back and forth calls the necessary callbacks and updates the prior probability. This is just a hypothesis, I don't have time to actually reproduce the bug and check the code.