Model Averaging in NBEATS Ensemble

[aamrb96]

Hello everyone,

I am using the NBEATS Ensemble estimator and I am a bit confused as to how the model averaging is done. I found the callback for ModelAveraging(avg_strategy=SelectNBestMean(num_models=1)). Two questions come to mind here:

1. Is this the way model averaging is done in an NBEATS ensemble?
2. Why is this done in the mx.trainer.Trainer class and not in the Predictor or Estimator class?

Any hint is appreciated.

Best,
Melchior

[aamrb96]

@mbohlkeschneider do you know how this works maybe?