[MNT] address pd.DataFrame.groupby(axis=1) deprecation in EnsembleForecaster

[ninedigits]

Fixes #5701

Changes accommodate deprecation of pandas.DataFrame.groupby(axis=1) introduced in pandas 2.1.0.

References:

• https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.groupby.html and https://github.com/pandas-dev/pandas/blob/d711b1d2ff9ee2c3faf0e118535d3d6218886a9d/pandas/core/frame.py#L9105-L9111

[fkiraly]

You should also update all-contributorsrc with your badges!
maintenance I suppose.

[fkiraly]

Your update is failing linting (code formatting) checks - perhaps you may like to split the one-liner into multiple lines to avoid too many brackets and indents.

[fkiraly]

PS, how about this (different from the other code in the issue):

y_pred = y_pred.T.groupby(level=1).agg( 
     _aggregate, self.aggfunc, self.weights 
 ).T

[ninedigits]

PS, how about this (different from the other code in the issue):

y_pred = y_pred.T.groupby(level=1).agg( 
     _aggregate, self.aggfunc, self.weights 
 ).T

In order to get this to work, the input data into _aggregate also needs to be transformed.

y_pred = y_pred.T.groupby(level=1).agg(
    lambda y, aggfunc, weights: _aggregate(y.T, aggfunc, weights), 
    self.aggfunc, 
    self.weights
).T

This passes the unit tests and avoids the FutureWarning, and it's similar to the original code. We could keep this line the same

lambda y, aggfunc, weights: _aggregate(y.T, aggfunc, weights)

Or define it as a wrapper somewhere else:

def _aggregate_transform(y, aggfunc, weights):
    return _aggregate(y.T, aggfunc, weights)

What do you think?

Edit:
I have a unit test that I wrote to see if the future warning is triggered. Not sure it's appropriate to include in the codebase since a different future warning might be triggered in the future.

@pytest.mark.parametrize(
    "forecasters",
    [
        [("trend", PolynomialTrendForecaster()), ("naive", NaiveForecaster())],
    ],
)
def test_ensemble_pandas_future_warn(forecasters):
    """Raises exception if FutureWarning triggered within EnsembleForecaster"""
    y = make_forecasting_problem()
    with pytest.warns(None) as record:

        forecaster = EnsembleForecaster(forecasters)
        forecaster.fit(y, fh=[1, 2, 3])
        _ = forecaster.predict()
        for warning in record:
            if isinstance(warning.message, FutureWarning):
                raise AssertionError("FutureWarning was triggered")

[fkiraly]

Thanks, that works!

Re warning, we monitor for FutureWarning-s regularly, but addressing them is a matter of maintainer time, so this is very much appreciated! They are listed at the end of diagnostic runs (you can look at "details" and the CRON job)

[ninedigits]

Looks like it still failed the linting steps. Should I revise or are we good to go?

[fkiraly]

Should I revise or are we good to go?

Please revise, the tests should all pass.

Here's a guide on setting up linting checks on your computer:
https://www.sktime.net/en/stable/developer_guide/coding_standards.html#setting-up-local-code-quality-checks

Let me know if it turns out too fiddly, I can do it for you then.

[ninedigits]

Should I revise or are we good to go?

Please revise, the tests should all pass.

Here's a guide on setting up linting checks on your computer: https://www.sktime.net/en/stable/developer_guide/coding_standards.html#setting-up-local-code-quality-checks

Let me know if it turns out too fiddly, I can do it for you then.

Thanks, I submitted the latest revision after running the black code reformatting steps.