https://github.com/nixtla logo
#mlforecast
Title
# mlforecast
k

Kyle Schmaus

05/31/2023, 9:20 PM
๐Ÿ‘‹ Hello
mlforecast
world ๐ŸŒŽ ! A question for the package maintainers - would you every be interested in adding support for fraction differencing, analogous to ARFIMA models? I'm going to implement something like this for an existing
mlforecast
project, and I'd be happy to create a ticket and/or PR if it's a direction you'd be interested in.
f

fede (nixtla) (they/them)

05/31/2023, 9:52 PM
hey @Kyle Schmaus! That would be awesome! If I understand correctly, ARFIMA models are local (one model per time series). If thatโ€™s correct, maybe we could include them in the
statsforecast
library, because
mlforecast
is designed to train machine learning models globally (one model for all the time series). We would be delighted if you could collaborate on this new feature by sending us a PR. Let me know what you think! We can also guide you on how to collaborate on our repos. ๐Ÿ™‚
k

Kyle Schmaus

05/31/2023, 10:02 PM
@fede (nixtla) (they/them) Thank you for being open to the idea ๐Ÿ’ก ! I think one could do ARFIMA in
statsforecast
, but I was more interested in enabling non-integer differencing in
mlforecast
. So right now I think if I wanted to difference hourly data once I might use
differences=[24]
(I might pick another interger too ) or if I wanted to difference twice I might do
differences=[24, 24]
. There are some use cases where I want to the difference order to be a number between 1 and 2.
I can put together a worked example over the next few days. What's the best method of communication: sharing via your slack channel or through github issues?
f

fede (nixtla) (they/them)

06/01/2023, 6:32 AM
Oh, that makes total sense for the
mlforecast
library. Letโ€™s use github issues to communicate more visibility for users about the new feature. Feel free to ping me on Github so I can help you with whatever you need. Thank you, Kyle! This is exciting :)
๐ŸŽ‰ 1
k

Kyle Schmaus

06/01/2023, 4:42 PM
Sounds great - I'll try to drop something by early next week (spinning plates till then)
๐Ÿ™Œ 1