Hi everyone, I'm training NHITS and NBEATSx for a demand prediction task and I have some futr_exog_list features. I'm looking at 208 unique ids over a period of two and a half year and I'm predicting hourly values four weeks out. Both models perform pretty well on this task, but not during easter. Demand has peaks during easter and I have an easter feature, but I don't see much forecast improvement from that feature. I'm wondering whether moving holidays like easter is going to be hard to fit to a recurring pattern that these models can discover. Some unique ids have triple the normal demand during easter, so on the one hand it should be a robust signal, but on the other hand, the models have only seen two easters of the past. Is there anything I can do to help the model learn, or should I try some other models?

Maybe add demand of last year during the same event. Might still be difficult to pick up if there's few years of data.

Yes, true. I have also tried to add static exogenous features, but there was very little response to see from that too. I'm pondering an idea of adding a per unique id embedding to N-HITS. Is this something that the community would be interested in? I guess I can just create my own fork and see if I can get some improvements from it and take if from there. Any advice regarding implementing this? Should I write an issue?

Easter is a holiday that moves around the calendar very wildly (2025-04-20, 2024-03-31, 2023-04-09) so if there are other features that have higher correlations to the target variable it is unlikely any exogenous feature you add will have the desired effect. I would suggest targeted pre/post processing to adjust for the holiday.