New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[MNT] [ENH] Several updates in direct statsforecast interface estimators #5920
Conversation
Apologies, I did a mindless bulk migration - might be my mistake, might be not having been entered properly in the first place. |
Minor note, |
a0c20c0
to
406613a
Compare
Agreed, but The failing tests are all from |
to avoid blocking this PR, can we skip |
v1.0.0 is the first tag supporting `from statsforecast.models import AutoARIMA`
v1.1.0 is the first tag supporting `from statsforecast.models import AutoCES`
v1.3.0 is the first tag supporting `from statsforecast.models import AutoTheta`
v1.3.2 is the first tag supporting `from statsforecast.models import AutoETS`
v1.5.0 is the first tag supporting `from statsforecast.models import (G)ARCH`
v1.2.0 is the first tag supporting `from statsforecast.models import MSTL`
406613a
to
9ebe468
Compare
I thought the same, but wasn't sure if it's an acceptable solution. Since you also suggested, doing it for now by combining both options.
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Makes sense.
Can I ask a quick question: why did test_statsforecast_mstl
previously pass?
Blocking comments:
- the content of
_config
affects only suite tests, in theTestAll
clasess, sotest_statsforecast_mstl
wlil have no effect, so there it should be removed - I worry that we track nowhere that we need to reactivate the test, or what to do when the issue is fixed. Alternatively, we should record in an issue what exacly to do to reverse once the bug is fixed.
- in
test_statsforecast_mstl
, should we maybe instead theck the tag and returnNone
? That would remove one location requiring an explicit reversal.
- in
I do not know. Unchecked guess: specific I checked locally that running following commands works only if I have > pytest sktime/forecasting/tests/test_statsforecast.py
> pytest sktime/forecasting/tests/test_all_forecasters.py -k "StatsForecastMSTL" It fails with |
I see - back-adaptation is quite tricky, perhaps impossible, if the interfaces do not remain stable |
@fkiraly @benHeid @achieveordie please let me know if there are any suggestions for this PR. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nothing more to do - my request (test skip) has been addressed
Per estimator lower bound
I checked the imports in
_get_statsforecast_class
for all direct estimators, and first tag in StatsForecast repository where that class is present. Accordingly, updated thepython_dependencies
tag per concrete estimators as follows:AutoARIMA
:v1.0.0
AutoCES
:v1.1.0
AutoTheta
:v1.3.0
AutoTheta
:v1.3.2
ARCH
andGARCH
:v1.5.0
(FYI @eyjo)MSTL
:v1.2.0
(FYI @luca-miniati)Missing credits
_GeneralisedStatsForecastAdapter
by adding support for argument differences across versions, added in theauthors
.StatsForecastBackAdapter
) were missing from__author__
, added there.StatsForecastBackAdapter
in__all__
.StatsForecastARCH
, added there.Enhancements
v1.7.1
adds a newphi
parameter toAutoETS
, added support for the same.pydocstyle
was failing due to missing docstrings, added one liner summaries (unlikely to be user facing methods).MSTL changes
StatsForecastMSTL
was initialising the attributes after call tosuper().__init__
, which was causingAttributeError
in tests that check for presence of properties. This PR moves those initialisations prior to that call. FYI @luca-miniati.StatsForecastMSTL
, as it was not working as reported by @EliasKng in [ENH] expose fitted params forstatsforecast
forecasters viaget_fitted_params
#5720. FYI @luca-miniati.Update of lower/upper bounds
statsforecast<1.0.0
with their current imports, not evenAutoARIMA
, updated lower bound from0.5.2
to1.0.0
. Previous adapter by @FedericoGarza was importing from a different location.v0.5.2
was released in March 2022, andv1.0.0
was released in August 2022, so I think chances of affecting users is quite low.AutoTBATS
needsv1.7.2
, so updated upper bound from1.7.0
to1.8.0
.