You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I was using Metal's LF analysis function, which gives you the "empirical accuracy" of your LFs. It was pretty confusing, I wasn't sure if it was computing its accuracy or its precision. After some experimentation I realized it's computing rule precisions. So, I thought maybe it would be better to call it "precision" to avoid confusions.
The text was updated successfully, but these errors were encountered:
@astarostap thanks for pointing this out! We need to make sure our naming is clear here. As you pointed out, in all but the simplest label model configurations, 'accuracy' is a misnomer. More precisely, the weights are the estimated conditional probabilities. I think this is clear in the label model now but not necessarily all helper functions throughout the repo... we'll do a cleaning pass here!
I was using Metal's LF analysis function, which gives you the "empirical accuracy" of your LFs. It was pretty confusing, I wasn't sure if it was computing its accuracy or its precision. After some experimentation I realized it's computing rule precisions. So, I thought maybe it would be better to call it "precision" to avoid confusions.
The text was updated successfully, but these errors were encountered: