I have made some progress, within the "lab" attached to one of my training sets. I have defined two scoring functions, and they run properly (I return "synthetic" data consistent with the required formats). But then I get the following error:
ValueError: Classification metrics can't handle a mix of multiclass and continuous targets
It is true that my features are a mixture of multiclass and continuous targets. Still, I have two questions:
1) if I am writing custom functions, why should Dataiku care about this mixture?
2) why can't this error be listed before wasting computational resources processing my custom functions? Dataiku must know about this mix immediately at the very first stage when my custom algorithm is trained (which Dataiku had no issues with.)