Closed baniasbaabe closed 11 months ago
Hey @baniasbaabe, thanks for raising this. Do you mean aggregate by unique_id first and then overall or just overall from the start?
Yes, just overall from start, so it shouldn‘t perform a groupby. I guess it‘s not possible rn, or?
It's usually not a good idea because the series can have different scales, so big errors in series with small values wouldn't produce a high RMSE for example. Also longer series would weigh more doing that. It's probably better to average the values across all ids, that way at least each serie weighs the same.
I see that the metrics are usually performed for every
unique_id
but is there a possibility to output an overall metric for allunique_ids
?