Fix validation error type PR02 and check in CI

[MarckK]

Problem description

In order to have a continuous check by the CI on validation error PR02 (Unknown parameters), fixing them in the code base enables the addition to the CI for automated testing in the future.

Please see overview of the errors.

Todo:

• get rid of the errors in the code base
• update the code_check.sh script to take into account the PR02 type of errors

[nickagian]

I'll work on that if no objection!

[MarckK]

Yes, please do @nickagian. :)

[nickagian]

Ok, I'd like some help.

There are many cases that have the same root cause but I don't know how to solve this.

It happens when a method gets delegated from a class to another. Usually by using the pandas.core.accessor.delegate_names decorator.

The problem is that the new class gets a method method_name(self, *args, **kwargs) but the docstring it inherits` has the parameters by their names and this obviously creates validation errors.

Any ideas on how to solve this?

I thought perhaps of passing also the desired docstring to the decorator instead of using the one from the original class, but I am not sure if that is a good idea.

[TomAugspurger]

@nickagian do you have an example of a method that is failing this check?

[TomAugspurger]

Found one, pd.Series.dt.to_period. I think the proper fix is to ensure that the signature of these (generated) methods is correct. This is possible, but not the most straightforward.

[nickagian]

@TomAugspurger Yes exactly, or pandas.CategoricalIndex.rename_categories and many more.

[TomAugspurger]

If you're interested, https://github.com/pandas-dev/pandas/blob/fe1654faa86836a0007bb513504e57c5c9935b8b/pandas/util/_decorators.py#L200-L215 provides an example of rewriting signatures. But I wouldn't include this under "Good first issue" :)

…

On Wed, Feb 27, 2019 at 3:12 PM nickagian ***@***.***> wrote: @TomAugspurger <https://github.com/TomAugspurger> Yes exactly, or pandas.CategoricalIndex.rename_categories and many more. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#25376 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHIrPwKjT-hWiDu7oawsiXQ_rmrNCAks5vRvTbgaJpZM4bDoBQ> .

[nickagian]

I have another question, regarding the pandas.core.groupby.base.whitelist_method_generator function. I think there is a bug here when we check if isinstance(f, types.MethodType):.

This is wrong if applied directly to a class, since f is only a function not bound to any object yet. Thus, the check we do here will give False anyways.

This creates a family of PR02 errors, for example for pandas.core.groupby.DataFrameGroupBy.quantile. This is wrongly treated as property due to the bug (it is actually a method) and this results in this validation error.

My question is: Is there a way to do this check correctly?
I haven't found a way so I think one solution would be to add an additional parameter to the pandas.core.groupby.base.whitelist_method_generator function that defines the type (method or property).

The only drawback is we have to provide the method names in the whitelist parameter each time it is called. If you spend one minute in the code you'll notice that right now the approach is that we basically define this list in the pandas.core.groupby.base module (common_apply_whitelist, dataframe_apply_whitelist etc.). An alternative would be to modify these lists into dictionaries, where also the type (property or method) is provided.

[mroeschke]

Closing as a duplicate of #27976