darts
darts copied to clipboard
typical final accuracy for 1st & 2nd derivative search step?
What accuracy should I expect at the end of a typical architecture search for both the 1st and 2nd derivative cases on CIFAR-10?
To be specific, I mean the actual search numbers rather than the final training run numbers.
Found one answer in #7, it looks like 87%-89% final accuracy should be expected in some search configuration, perhaps the one in the README.md, but it doesn't make clear the difference in results between expectations for 1st and 2nd derivative searches, so this question should remain open.
@quark0