Is normalisation essential for NBEATS and NHITS, I have seen Darts has normalisation functionality but have not seen this mentioned in NeuralForecast.

check out

scaler_type

and just be aware there are two "flavours" of scaling, i'll link a discussion from a couple weeks ago

Yes I have seen it before @Chris Gervais my question is more on whether it adds value

our experience with scaling in NF is a bit mixed but in general yes and we've found

minmax

tends to do a bit better than

robust

Useful materials thanks @Chris Gervais

we're still trying to diagnose it but sometimes we have issues with scalers on live inference data where the scale is completely off