Hiroyuki Vincent Yamazaki
Hiroyuki Vincent Yamazaki
@not522 do you mind elaborating a bit on what must be discussed given it's labeled `needs-discussion` ?
Ah, thanks for the explanation. Given the documentation, it sounds like bugs but that needs to be fixed. But you mean that it may/may not be more intuitive to keep...
I think you're right. It's clearly manifested when creating the bracket studies (a study which contains a subset of the trials) in https://github.com/optuna/optuna/blob/v2.10.0/optuna/pruners/_hyperband.py. Let me share a somewhat more limited...
Thanks for the prompt response. How about keeping the issue open until it's been fixed?
Using `PYTHONHASHSEED` is maybe a reasonable workaround, in user-land. As for how to address that in Optuna, I suggest we adopt some hashing algorithm in `hashlib` (we don't want to...
You're right, and a simple mod might also suffice. The logical change is probably a matter of a few lines of code (if not one), but we probably want to...
Right, so benchmarking with kurobako in a multi-process setting is not straight forward (cc. @sile is this correct? 🤔 ). One compromise could be to consider this a bug fix...
Thanks for the reference and the careful consideration. As it's also stated there, benchmarking is important, and in my understanding @toshihikoyanase addressed that in https://github.com/optuna/optuna/issues/3083#issuecomment-963833619 so certain degree. I did...
@jrbourbeau, don't mean to stress but do you have any concerns or questions with the review state so far? Just leaving a heads up that we've made some recent changes...
Sure, I'll try to stay updated with this PR. I'd be happy to help. There was by the way another method introduced to the storage just a few days ago,...