Reproducing the results

[edixiong]

Hi author,

I am having difficulty reproducing the results on cifar-10. The paper claimed test error of 2.46+-0.03 with 600 epochs, but when I am evaluating with the provided 'DrNAS_cifar10' genotype, I only get accuracy 94.86 with 600 epochs and getting accuracy of 97.44 with 1200 epochs. It seems like the default parameter in the code matches the parameter claimed in the paper, or did I miss something here?

I saw in PC-DARTS github page that there is much randomness in training cifar10 so the result is not stable, is it also the case here?

Thank you for your response.

[xiangning-chen]

Hi,

Can you try the 'DrNAS_imagenet' and see the result?
There are some randomness, but should not be that much.