Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

NeuralForecast |TimeMixer Model - exogeneous variables support #1137

Open
MLfreakPy opened this issue Sep 2, 2024 · 3 comments · May be fixed by #1088
Open

NeuralForecast |TimeMixer Model - exogeneous variables support #1137

MLfreakPy opened this issue Sep 2, 2024 · 3 comments · May be fixed by #1088

Comments

@MLfreakPy
Copy link

What happened + What you expected to happen

Using exogenous variables in TimeMixer model, whether historical or future, leads to an error message. As a result, the model can currently not be used with exogeneous variables, however, these are very crucial for good forecasts in many applications.

The issue has been raised also in Github Issue 1016.

@marcopeix, @kwuking: Do you have a timeline of when this issue can be fixed?

Versions / Dependencies

Name: neuralforecast
Version: 1.7.4

Reproduction script

see above

Issue Severity

High: It blocks me from completing my task.

@MLfreakPy MLfreakPy added the bug label Sep 2, 2024
@marcopeix marcopeix added enhancement and removed bug labels Sep 3, 2024
@marcopeix marcopeix linked a pull request Sep 4, 2024 that will close this issue
@marcopeix
Copy link
Contributor

We have a draft PR trying to support exogenous feature for TimeMixer. We welcome any contributions from the community! If you want to get give it try, let us know and we're here to answer your questions!

@elephaint
Copy link
Contributor

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 insanely hard to do this right, so it seems a limitation of the architecture that we are not placed to solve.

I'd vote for closing unless I'm misunderstanding here.

@carusyte
Copy link

carusyte commented Sep 7, 2024

I'm really keen to harness this model's potentiality especially on multivariate / exogenous features, too.
TBH I'm not a data scientist, so I could be missing the nuances of multivariate time series and exogenous features. I read in the original paper that TimeMixer supports multivariate, though. How can I feed additional covariates into this model other than 'y' feature?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging a pull request may close this issue.

4 participants