Olivier Sprangers
Olivier Sprangers
> I ran my own small experiment, and models are slightly faster, with no degradation in performance. The refactoring looks good to me, but a second opinion would be great...
Todo: - [x] Check correct functioning of #1233 and add a test for that functionality in the code - [x] ~~Add #1233 to mixture losses~~ Not in this PR, keep...
Thanks - I can reproduce, it seems to come from the fact that the config hyperparameter isn't properly updated.
Hey, thanks for reporting. I think this issue is in the wrong repo (should be in NeuralForecast). To answer though, the Tweedie loss is a bit finicky. I think the...
Did you update to the latest version of Neuralforecast? I can't reproduce the issue.
- You should always set a scaler when training an RNN. Try with `robust`. - Train more steps, for recurrent easily 2000.
Thanks for creating the issue @Antoine-Schwartz. I'll have a look into it
THanks for raising the issue; I can reproduce. I'll have to think about this a bit; I agree with your point that forecasts should only be required for the Top,...
I'm still trying to solve this, but it's a pain in the *ss. Due to the design that lets a user provide multiple reconcilers at the same time, there's a...
I think you're right - what would such an example look like in your case? Maybe we can start with an existing example and fit it to your needs? (i.e....