https://github.com/nixtla logo
v

virgilio espina

03/27/2023, 5:19 PM
I have negatives AutoNHITS forecast. Any comments and suggestions?
k

Kin Gtz. Olivares

03/27/2023, 5:25 PM
Hey @virgilio espina I would try using probabilities that have a restricted positive domain. We have currently available Poisson, NegativeBinomial and Tweedie. Here is the DistributionLoss catalog: https://nixtla.github.io/neuralforecast/losses.pytorch.html#distributionloss
v

virgilio espina

03/28/2023, 4:58 AM
how can i add weights for the pandemic period to avoid overlearning? is it possible to add weights? (