Hi, I'm not entirely sure if this is a bug, but noticed it and figured I'd file it and let the maintainers decide.
It looks like there is a bug in the hyperparameter grid for the MixUp model, where the "range" of the random selection is between 10^-1 and 10^-1 (effectively fixing the value of that hyperparameter):
Seems like a bug indeed. The appendix in the DomainBed paper seems wrong as well. Looking at the original mixup paper, a reasonable grid would be 10**r.uniform(-1, +1). Mind sending a PR?
Hi, I'm not entirely sure if this is a bug, but noticed it and figured I'd file it and let the maintainers decide.
It looks like there is a bug in the hyperparameter grid for the MixUp model, where the "range" of the random selection is between 10^-1 and 10^-1 (effectively fixing the value of that hyperparameter):
https://github.com/facebookresearch/DomainBed/blob/2ed9edf781fe4b336c2fb6ffe7ca8a7c6f994422/domainbed/hparams_registry.py#L67
Specifically, I am referring to
10**r.uniform(-1, -1)
.Is this an issue, or am I just misunderstanding how the code works?