Brian Kroth
Brian Kroth
@jsfreischuetz
See Also: #687
I think the whole point is to automate the enumeration of the configs. I'd rather see us implement #687 and then extend it by adding a `shedule_ablation_study(config)` for a given...
> I am not sure that this functionality belongs to the Scheduler. The Scheduler determines _when_ the trial should be executed, but not the _configuration_ of this trial. The latter...
> @bpkroth says: > > > This isn't determining which config might improve confidence or scores, that's still handled by the Optimizer. > > Instead, it's adjusting the method of...
> In all, I am fine with the implementation, but I really don't like the name `ManualOptimizer`. Can we call it `FewShotOptimizer` or `ConstOptimizer` instead? @bpkroth , WDYT? Works for...
> I think we can merge in the PR already. I am not quite happy with the `ManualOptimizer` name (it does not tell exactly what the class does), but I...
I'm not quite satisfied with this one - there's some issues of it not detecting the need to rescan the whole set of files after conda rebuilds for instance -...
Would happily accept a draft PR to help get started @yshady. Thanks!
We'd happily accept a PR!