[neuralforecast] - Where is Granger Causality Implemented?

[webert6]

Description

Here the documentation mentions Granger causality is implemented with historical exogenous variables, but I could not find in the source code where this happens. I am in the process of scoping a project including additional exog variables and I'm concerned this is not implemented yet . I'm looking to validate that historic exog variables utilize GC as described. Could you help me identify where this takes place? Thanks!

Link

No response

[marcopeix]

Hello! We're just saying that the validity and predictive power of historic exogenous features relies on Granger-causality. There is no deep learning architecture explicitly verifying Granger-causality. Ultimately, you can let the model find the best features combination to get the best forecast possible.

Does that answer your question?

[webert6]

Makes sense! Thanks for the clarification.