Olivier Sprangers
Olivier Sprangers
> > @JQGoh 1.7.4 was [already released today,](https://github.com/Nixtla/neuralforecast/releases/tag/v1.7.4) we took out the futr_exogenous functionality of `TimeMixer` for now, but we can make it part of the next release. > >...
@peacocktrain This is a draft PR, which means it doesn't necessarily work. Please don't use draft PRs to work with, as functionality likely doesn't work properly. For now, I'd advise...
The original paper doesn't seem to support exogenous features. Upon learning this I don't see a reason to support this. Note that it's not easy to include exogenous - it...
Is being addressed!
[TimeXer supports (only) future exogenous variables](https://nixtlaverse.nixtla.io/neuralforecast/docs/capabilities/overview.html) Not in the planning to support historical afaik @marcopeix? I also don't really see how the original architecture supports historical exogenous, and we commonly...
> awesome @elephaint! i left a couple of comments. i was also speaking with @cchallu and we think it might be better to make it univariate in this iteration (mainly...
I think we're probably not going to merge this one.
#1059 is a generalization of this PR, right? I.e. the default behaviour of #1059 could be to remove empty windows during all steps of the pipeline - train/val/predict; and have...
Thanks for the suggestion! I understand the request; from my side I need to think about how this could be integrated in NF.
@JoaquinDF-UniLU #1023 should fix this.