Hi Team, I have multiple time series with different scale and different seasonality. Which neural-forecast will be ideal for this purpose?

Hello! It's hard to suggest models just like that. Keep in mind that neuralforecast trains a global model. So if your series are very different in patterns, it might be better to separate them and train a model for group of series that are similar. For the scale, it shouldn't matter too much, but you can specify the

scaler_type

to tune it to your use case. Otherwise, for univariate, NHITS, TiDE are good starting points. For multivariate, TSMixer, iTransformer usually work well. You can see the list of all models here.

Hi @Marco, thanks for the suggestions. What's your thoughts on ESRNN on this case. Does Nixtla have its own implementation?

We currently don't have ESRNN in neuralforecast.