https://eugeneyan.com/writing/evals/ about eugeneyan-comments HOT 4 OPEN

your discussion of precision and recall opened up a new way of thinking about accuracy to me. Kind of a weird way to learn the concept I imagine but highly appreciated nontheless

from eugeneyan-comments.

Thanks for the great post!

Could you share the code you use to calculate and plot of:

1. Precision/Recall vs. Threshold plot
2. Density vs. Prob inconsistent plot (with JS Divergence on top)

Thank you.

from eugeneyan-comments.

Also, I am curious if this is a typo:

Like ROC-AUC, a perfect classifier has PR-AUC = 1.0 while a random classifier has PR-AUC = 0.0

Why random classifier would have PR-AUC 0.0 instead of 0.5?

from eugeneyan-comments.

Could you share the code you use to calculate and plot of...

Added the code for classification metrics here

Why random classifier would have PR-AUC 0.0 instead of 0.5?

Doh, thanks for catching this. Not sure what I was thinking. It should read "a random classifier would have PR-AUC = proportion of positive labels". Thanks for catching it. Corrected!

from eugeneyan-comments.