1
$\begingroup$

I had some old code used to train a Random Forest Classifier (sklearn 0.17.1), for classification on two classes (spam/ham). I ran this in a docker container and sent it some data.

Sklearns predict_proba returns [0.3, 0.7].

I “rewrote” (literally changed a few lines for Python3 compatibility), trained with the same data, and sent it the same data, and now predict_proba returns [0.7, 0.3].

Did something change in sklearn.ensemble.RandomForestClassifier predict_proba or have I messed up somewhere, or does the new one still work?

Improve this question
$\endgroup$
  • 2
    $\begingroup$ Could you share a code that reproduces your error ? $\endgroup$ – Robin Nicole Apr 11 '19 at 20:02
  • $\begingroup$ My first guess would be that the way spam/ham get encoded changed. You should be able to call classes_ to check. $\endgroup$ – Ben Reiniger Apr 18 '19 at 2:31
1
$\begingroup$

Maybe... Have a you seen if one of the predict_proba produces 2 columns?

I think Python 3 produces two columns and the one comparable is the second.

Improve this answer
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.