Hello everyone I m continuing my experiments with DeepAR and

Question

Hello everyone I m continuing my experiments with DeepAR and I have a few questions for you Putting aside my the performance I m getting is still below the GluonTS implementation One possibility would be the absence of DeepAR being autoregressive it s a waste to do without lag 1 for example Or am I missing something Second question GPU predictions with mixtures mech loss like PMM require far too much GPU RAM +90GO in my case I suppose that by playing with `inference windows batch size` I should be able to handle it but does that mean that for inference DeepAR doesn t keep only the last `input size` points as it does for training Thanks in advance

Manuel · Answer

Maybe the DeepAR implementation could be changed to allow for a `lags` parameter as in `mlforecast` or to accept a list of lags as `input size` instead of a single integer so that we can include only some lags as with GluonTS s lags seq parameter instead of the full input size range < Kin Gtz Olivares>

Cristian (Nixtla) · Answer

Hi Thanks for the suggestion We will add the `lags sec` parameter soon