Classification vs regression

[marl]

Classification vs regression

[justinsalamon]

Yes, we tried this in some initial experiments, classification worked better so moved forward with classification. We use the gaussian trick (cf. paper) instead of 1-hot vectors during training so that close-but-not-perfect predictions aren't penalized as harshly as further off predictions.

[keunwoochoi]

classification worked better

I see. Does it mean that regression also worked but just not as good as the classification model? I'm curious!

[justinsalamon]

Yes

[keunwoochoi]

👍 Ok thanks!