Model's predicted values are not the same when using Dataiku API to retrieve the model and use it in

minliang1215
minliang1215 Registered Posts: 4 ✭✭✭✭
edited June 28 in Using Dataiku

I built an XGBoost model in DSS, and retrieved it using the Dataiku API inside python recipe. However, when I use the predict() function on the retrieved model object, the predicted values are not the same as they appear in the original model.

true value and predicted value in the original model object



Answers

  • UserBird
    UserBird Dataiker, Alpha Tester Posts: 535 Dataiker
    edited June 28

    Hi,

    The original model apply a set of preprocessing on data before doing the prediction. If you want to reproduce the result, you can use directly `my_predictor.predict()`.

    Another thing to be aware is that in the scoring recipe, for reproducebility, you should check the Force backend engine option

    Cheers,


Setup Info
    Tags
      Help me…