hard20 pts · first answer countstraining-serving-skew

Offline and online scores disagree

For the same customer, batch scoring in Python and the online Java service give different predictions from the same model file. What is the best long-term fix?

A.Package preprocessing with the model, or use one shared feature definition, so both paths run identical transformsB.Round all predictions to one decimalC.Retrain the model more oftenD.Turn off the batch job

Discussion 0

Join free to post in the discussion.

Offline and online scores disagree

Discussion 0

Related questions