Add auto_arima model from pyramid.arima

[alexcombessie]

Description

I want to benchmark GluonTS models with the auto ARIMA model implementation from pyramid-arima in Python.

References

• https://pypi.org/project/pyramid-arima/

[timoschowski]

That may be a good contribution. Can you follow the examples from here: https://github.com/awslabs/gluon-ts/tree/master/examples/dockerfiles where we have "R Forecast" Package included.

Alternatively, you could also try the auto ARIMA implementation from that very package, if it helps your purpose.

Additional documentation is here: https://gluon-ts.mxnet.io/api/gluonts/gluonts.model.r_forecast.html https://gluon-ts.mxnet.io/api/gluonts/gluonts.model.prophet.html

[alexcombessie]

Hi,

Thanks, I am indeed familiar with R forecast::auto.arima. However, in my case, I am specifically looking for a pure Python alternative. Our customers do not always have Docker and R installed.

Can you confirm that it would be possible to integrate Pyramid-ARIMA to GluonTS without going through Docker?

Thanks,

Alex

[timoschowski]

Understood, Alex. Yes, you should be able to integrate Pyramid-ARIMA directly (should be in the sense that there are no dependency conflicts or something else that comes up practically). Can you try this locally?

[alexcombessie]

I'll have to find some time for myself or someone from my team @dataiku to try it. I'll let you know if we succeed :)

[alexcombessie]

Hi @timoschowski,

FYI we've implemented a custom auto-arima model for our need: https://github.com/dataiku/dss-plugin-timeseries-forecast/blob/main/python-lib/gluonts_forecasts/custom_models/autoarima.py

Would you be interested in a PR to add it to your repo?

Cheers,

Alex