Calibration for ML at Etsy - apply() special // Erica Greene and Seoyoon Park // MLOps Coffee Sessions #78
MLOps.community
00:00
Virtuing Binary Prediction
When evaluating binary prediction, click, not clik, is a u c. So if the order of the results based ino the model score, ordered by the model prediction, doesn't change, then tha u c shouldn't change. The second metric that we looked at is what called the calebr calebration which is really simple. It's just the predictedte percent of e true, like clik ones in in the true eval set, divided by the predictedo true label from the model. And with this, it's kind of more granular than the calebration metric, because you will see where the performance starts to go wrong as well.
Play episode from 31:42
Transcript
