search

business-science / modeltime.gluonts

GluonTS Deep Learning with Modeltime
https://business-science.github.io/modeltime.gluonts/
Other
38 stars 11 forks source link

DeepAR: Student t vs. Negative Binomial #44

Open PaulSpringerMI4People opened 2 years ago

PaulSpringerMI4People commented 2 years ago

Hello Matt,

I'm very new to your great package and have a question with regard to the distribution used by deepAR-model. As far I understood it correctly, it uses Student t-distribution by default. For my data, I'm interested in negative binomial distribution. Based on the basic example:

model_fit_deepar <- deep_ar( id = "id", freq = "M", prediction_length = 24, lookback_length = 48, epochs = 5 ) %>% set_engine("gluonts_deepar") %>% fit(value ~ ., training(m750_splits))

could you explain how I can adjust it to negative binomial distribution.

Many Thanks in advance!

leonhGeis commented 1 year ago

I am currently trying to implement the negative binomial distribution, as it is in many cases a better choice, especially if there are a lot of zero values in your times series.

Any news or success, concerning the issue?