Eval metrix calculated with model_performance() differ from those with caret::confusionMatrix()

[FrieseWoudloper]

I'd like to compare a decision tree with a random forest, So I first trained a decision tree and calculated some evaluation measures on a test set using caret::confusionMatrix(). I did the same using the DALEX package. Although the trees and predictions are the same, the metrics (precision, recall, F1) calculated by model_performance() differ from those calculated with caret::confusionMatrix(). Why is this? Am I doing something wrong?
See: https://rpubs.com/friesewoudloper/630778

[hbaniecki]

Hi,
I believe that caret uses 0 as the positive class and DALEX uses 1 as the positive class.
https://stackoverflow.com/questions/38263137/set-positive-class-to-1-in-r

[pbiecek]

@maksymiuks can we take care about this in the default predict function?

[maksymiuks]

It is, but the solution would require modification of the input model (adding an attribute in explainer function)

[hbaniecki]

Why not in model_performance?

[pbiecek]

imho either explainer or predict function should know which label is positive

[maksymiuks]

Let's focus on it after DALEX 2.0.0 release.

[maksymiuks]

Solved in #353