medium10 pts · first answer countsml-pipelinesdata-versioning

Label logic changed quietly

The definition of "churned" changed from 30 to 60 days of inactivity in the label SQL, but the dataset version did not change. What is the risk?

A.The model will refuse to load the new labelsB.Training will run twice as fast as beforeC.Nothing, since labels are not part of the dataD.Runs look comparable but were trained on different targets

Discussion 0

Join free to post in the discussion.

Label logic changed quietly

Discussion 0

Related questions