[DOC] BaseForecaster score function uses MAPE instead of sMAPE as metric

[MBristle]

Describe the issue linked to the documentation

The score function of the BaseForecaster suggests the metric that is used for evaluation is sMAPE:

Returns

        -------

        score : float
            sMAPE loss of self.predict(fh, X) with respect to y_test.

The code calls the mean_absolute_percentage_error function without any further parameters.

from sktime.performance_metrics.forecasting import (
            mean_absolute_percentage_error,
        )

        return mean_absolute_percentage_error(y, self.predict(fh, X))

But the default values of the mean_absolute_percentage_error function are:

def mean_absolute_percentage_error(
    y_true,
    y_pred,
    horizon_weight=None,
    multioutput="uniform_average",
    symmetric=False,
    **kwargs,
):

and the doc states:

... 
    symmetric : bool, default=False
        Calculates symmetric version of metric if True.
...

Thus I conclude that instead of the sMAPE, MAPE is returned.

Suggest a potential alternative/fix

Either adjust the documentation such that it states that the MAPE metric is used or adjust the implementation by adding the symmetric parameter. I did not investigate if there are other places with the documentation error. ;) Many thanks!!

[fkiraly]

Good spot! I thik the default for MAPE (non-symmetric) was changed a while ago, which affected the score function.

In order to not break functionality downstream, it's better to change the docstring than the code logic.

Would you like to make a PR to change the docstring? Would be much appreciated!