Calibration for ML at Etsy - apply() special // Erica Greene and Seoyoon Park // MLOps Coffee Sessions #78
MLOps.community
00:00
The Distributions of Output Scores Shift Wildly When You Changed Models
A. Leokig: A lot of these models the loss function, or something a we mean, the models trained to optimize the loss function. And sometimes these a mild distribution scores shifted wildly after changing from i a treats base model to a t and n legistic regression on the distribution would shift greatly. Since it's an imput to another model, tho, output scores also shift very greatly. Ten, ah, there's a lack of observability in that space, which we can go into. Ah, but yet, those are some of the use cases where talbisn comes very important. Did you figure out why it shifted so much when you changed these different models? Yes
Play episode from 24:21
Transcript
