https://github.com/nixtla logo
r

Raghuvansh

06/01/2023, 11:10 AM
Hi Guys, I am trying to use AutoArima from Statsforecast but I don't want to use any information criteria for the best model but use a loss on a test set for best model, is there a way I could do it ?
f

fede (nixtla) (they/them)

06/12/2023, 7:36 PM
hey @Raghuvansh! Yes! You can select different information criteria, but we must do some tests. Please help us open an issue so we can write a tutorial on the topic. Here’s the link: Thank you! https://github.com/Nixtla/statsforecast/issues/new?assignees=&labels=documentation[…]&title=%5B%3CLibrary+component%3A+Models%7CCore%7Cetc...%3E%5D+
r

Raghuvansh

06/12/2023, 7:37 PM
Yeah sure, also could you add support for the test set ?
f

fede (nixtla) (they/them)

06/12/2023, 7:40 PM
yes, sure! In that case, you could open an issue related to using different criteria and another one to support the test set. 🙌
r

Raghuvansh

06/12/2023, 7:42 PM
Sure, Created this for test set - feature request - https://github.com/Nixtla/statsforecast/issues/525
Created a documentation issue as well for different information criteria https://github.com/Nixtla/statsforecast/issues/526
f

fede (nixtla) (they/them)

06/12/2023, 7:51 PM
Thank you, @Raghuvansh!
2 Views