Guessing correctly the EM_EVENTPROBABILITY in scoring code

When using pzmm.ImportModel.import_model for wirting scoring code for binary/multinomial classification it assumes from result = SKmodel.predict_proba(...) that the result[0] is the target variable probability and assigns it to EM_EVENTPROBABILITY, which may be incorrect, the levels probability may be in a distinct column.

At least for classification models, you can get the probability class labels and orders from SKmodel.classes_. If the user adds something like targetevent = "label", we could use it to assign the correct column of probability to EM_EVENTPROBABILITY.

It is likely that this only works for classification, and should not be confused when the model is a regression.

Commencez par pzmm.ImportModel.import_model et examinez comment SKmodel.predict_proba(...) est mappé vers EM_EVENTPROBABILITY. Comparez les colonnes de probabilités avec SKmodel.classes_, puis déterminez comment une valeur de targetevent pourrait sélectionner la classe correcte pour la classification binaire et multinomiale sans affecter le comportement de la régression. Vérifiez le comportement obtenu avec les tests pertinents du projet.