breaking: allowmean and allowdrift True by default in AutoARIMA

[V-nguard]

Hello,

I encountered the same issue as #896, this should fix it.
I couldn't commit without adding the other files; I don't think I missed a step in the contributing guidelines either.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[CLAassistant]

All committers have signed the CLA.

[jmoralez]

Hey. Thanks for your contribution. Can you please add the pin nbdev==2.3.25 to the settings.ini file and restore the other files? Seems like that was added in v2.3.28

[V-nguard]

Hi, I'm not fully fluent in git, so is there another way to restore the files without creating another commit? Not sure what the intended operations are (like I'm assuming I don't commit the settings.ini change either because that should be another pull request)

[jmoralez]

I think you can run git checkout main -- python to restore all the python files and then nbdev_export again.

It's ok to commit the settings.ini file here, it only changes a dev requirement.

[V-nguard]

Thanks for the git command. This should work.

[jmoralez]

@V-nguard seems like this improves the performance on the M3 dataset. Can you please set this line

statsforecast/experiments/m3/src/evaluation.py

Line 53 in 8e16c02

'AutoARIMA': 4.57,

to 4.46 (logs)?

[V-nguard]

I did the change, but I can't commit due to ruff. Something about imported numpy but unused. I ran ruff check --fix *, and now it's saying

experiments/m3/src/evaluation.py:41: error: No overload variant of "__getitem__" of "list" matches argument type "list[str]"  [call-overload]
experiments/m3/src/evaluation.py:41: note: Possible overload variants:
experiments/m3/src/evaluation.py:41: note:     def __getitem__(self, SupportsIndex, /) -> Any
experiments/m3/src/evaluation.py:41: note:     def __getitem__(self, slice, /) -> list[Any]
experiments/m3/src/evaluation.py:42: error: No overload variant of "__getitem__" of "list" matches argument type "str"  [call-overload]
experiments/m3/src/evaluation.py:42: note: Possible overload variants:
experiments/m3/src/evaluation.py:42: note:     def __getitem__(self, SupportsIndex, /) -> Any
experiments/m3/src/evaluation.py:42: note:     def __getitem__(self, slice, /) -> list[Any]
experiments/m3/src/evaluation.py:42: error: No overload variant of "__setitem__" of "list" matches argument types "str", "Any"  [call-overload]
experiments/m3/src/evaluation.py:42: note:     def __setitem__(self, SupportsIndex, Any, /) -> None
experiments/m3/src/evaluation.py:42: note:     def __setitem__(self, slice, Iterable[Any], /) -> None
experiments/m3/src/evaluation.py:43: error: "list[Any]" has no attribute "set_index"  [attr-defined]
experiments/m3/src/evaluation.py:44: error: "list[Any]" has no attribute "columns"  [attr-defined]
experiments/m3/src/evaluation.py:45: error: "list[Any]" has no attribute "set_index"  [attr-defined]
experiments/m3/src/evaluation.py:46: error: "list[Any]" has no attribute "droplevel"  [attr-defined]
experiments/m3/src/evaluation.py:47: error: "list[Any]" has no attribute "to_csv"  [attr-defined]
experiments/m3/src/evaluation.py:48: error: "list[Any]" has no attribute "query"  [attr-defined]
experiments/m3/src/evaluation.py:49: error: "list[Any]" has no attribute "query"  [attr-defined]
experiments/m3/src/evaluation.py:60: error: "dict[str, float]" has no attribute "index"  [attr-defined]
Found 11 errors in 1 file (checked 3 source files)

[jmoralez]

Yeah, sorry that's mypy. We don't really care about running it on the experiments folder, so can you change this

statsforecast/.pre-commit-config.yaml

Line 28 in 8e16c02

exclude: 'setup.py'

to (experiments|setup.py)?

[V-nguard]

Done!

[jmoralez]

Hmm, it seems that the smape for the statistical ensemble in the m3 is increasing in the test, however I'm not able to reproduce that score locally, I've already tried on two different machines and I get what we currently have. I'll keep investigating.

[jmoralez]

Oh it was due to me having the NIXTLA_ID_AS_COL env variable set, which considered the index when creating the ensemble. I'm able to reproduce the score on the test now, can you please update that one as well? i.e. set this:

statsforecast/experiments/m3/src/evaluation.py

Line 57 in 8e16c02

'StatisticalEnsemble': 4.23,

to 4.3.

[jmoralez]

Thanks a lot!

[AzulGarza]

awesome @V-nguard and @jmoralez! lgtm