Mar 7Liked by Avi Chawla

Log loss is the gold standard in this case. The top k accuracy is still discontinuous with respect to the predicted probabilities, so it still obscures the changes in model performance. However, the log loss varies continuously as the predicted probabilities change.

Expand full comment

Yes, totally, Joe :)

Loss is like the most granular you can get. The only reason I tend to prefer top-k accuracy is because it has a higher interpretability in comparison to loss, and loss is not something you would typically measure on a test set :)

Expand full comment