4 ms·
Very cool, thanks for doing/sharing. I am curious why he says "non linear SVM are the most accurate but can't scale". I've used the good old C svmlight on prett
by thomasrossi 11y ago
Very cool, thanks for doing/sharing. I am curious why he says "non linear SVM are the most accurate but can't scale". I've used the good old C svmlight on pretty large datasets, it was not real time but..
- chestervonwinch 11y agoYou have ~n^2 exp operations (assuming rbf kernel) that weren't present before. This becomes extremely slow for big n in my experience.
- pmelendez 11y ago> I've used the good old C svmlight on pretty large datasets When the dataset size is in Gbs, Non-lineal kernels might not even finish. Kaggle has good sized datasets that you could use to see the difference, it is insane. That said, I would love to see it included in this kind of benchmark to see how ACU would compare.