Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

Hey <@U038ZT9UDGA> any updates on this issue? <https://github.com/Nixtla/neuralforecast/issues/566>

<https://nixtla.github.io/neuralforecast/examples/signal_decomposition.html>

I mean this one:
```Regarding the RuntimeError: maximum size for tensor at dimension 2 is 280, but the size is 336

For the moment we are yet to include the exogenous partial predictions, as the current version of NBEATSx is not performing the projections on the exogenous basis, rather its taking the exogenous data as inputs of the MLPs.
We will get to work on this feature soon.```

looking at the example it doesn't have exo?

No for the moment, we have not implemented the decomposition against exogenous variables in NeuralForecast yet

Here is the original paper code:
<https://github.com/cchallu/nbeatsx>

Any idea on what modifications can be done to add support for exo?

We need to modify this script:
<https://github.com/Nixtla/neuralforecast/blob/main/neuralforecast/models/nbeatsx.py>

1. Add ExogenousBasis to `nbeatsx.py`: <https://github.com/cchallu/nbeatsx/blob/main/src/nbeats/nbeats_model.py#L299>
2. Add the ExogenousBasis option into `nbeatsx.py`'s create_stack, following: <https://github.com/cchallu/nbeatsx/blob/main/src/nbeats/nbeats.py#L212>

If want it in NF, yes. I think it can be useful for many people.
Do you want to try?

Sure, I'll start by reading your paper, maybe we can setup a chat next week or so?

Sounds good.
Here is the NBEATSx paper: <https://arxiv.org/abs/2104.05522>