Valentin Flunkert
Valentin Flunkert
> I think this is similar to what I had worked on before. > > #650 and #384 > > I would argue for discussing the designs first, before going...
> https://arxiv.org/abs/2103.03230 Cool, thanks for the suggestion @kashif ! I'll try this out.
> @vafl as far as I can see, all the losses can be now endowed with time series augmentations via the `RandomApply` on both representation branches. I still do not...
I think we had this at some point (or at least considered doing it). If I remember correctly, we decided against it, because many IDEs like intellij will show a...
> I really dislike this argument: Some IDE is dictating how to write code, which I feel not comfortable with. > This is a problem of the IDE and not...
Ok, nice. Maybe that works with pycharm, too then. I think we tried with some more naive option.
I'm ok with the syntactic sugar and I also think that we need to rethink the transformations in general.
Hi Sebastian :-) thanks for the pointer! This would fit in well I think. Ideally, we could support any of the existing metrics on a per-group bases. I'll play around...
Thanks. Just to clarify, we can use the aggregator api without any grouping and it will work exactly like the existing metrics that we compute. The problem I mentioned only...