REGR: Series.groupby([]) raises

[TomAugspurger]

Pandas 0.24.2

In [14]: s = pd.Series([])

In [15]: s.groupby([])
Out[15]: <pandas.core.groupby.groupby.SeriesGroupBy object at 0x11f4d8748>

master

In [8]: s = pd.Series([])

In [9]: s.groupby([])
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-9-cd9ba1ec9bdd> in <module>
----> 1 s.groupby([])

~/sandbox/pandas/pandas/core/generic.py in groupby(self, by, axis, level, as_index, sort, group_keys, squeeze, observed, **kwargs)
   7496         return groupby(self, by=by, axis=axis, level=level, as_index=as_index,
   7497                        sort=sort, group_keys=group_keys, squeeze=squeeze,
-> 7498                        observed=observed, **kwargs)
   7499
   7500     def asfreq(self, freq, method=None, how=None, normalize=False,

~/sandbox/pandas/pandas/core/groupby/groupby.py in groupby(obj, by, **kwds)
   2402         raise TypeError('invalid type: {}'.format(obj))
   2403
-> 2404     return klass(obj, by, **kwds)

~/sandbox/pandas/pandas/core/groupby/groupby.py in __init__(self, obj, keys, axis, level, grouper, exclusions, selection, as_index, sort, group_keys, squeeze, observed, **kwargs)
    362                                                     sort=sort,
    363                                                     observed=observed,
--> 364                                                     mutated=self.mutated)
    365
    366         self.obj = obj

~/sandbox/pandas/pandas/core/groupby/grouper.py in _get_grouper(obj, key, axis, level, sort, observed, mutated, validate)
    608
    609     if len(groupings) == 0:
--> 610         raise ValueError('No group keys passed!')
    611
    612     # create the internals grouper

ValueError: No group keys passed!

I think this should work. If not, we'll want to deprecate it properly.

[TomAugspurger]

@WillAyd do you have a good feeling for whether this should raise? It seems like a length-0 DataFrame raises for length-0 grouper.

Is there a problem with length-0 groupers, when the obj is also length-0?

[WillAyd]

Raising seems more intuitive - is there a particular use for the empty grouper?

…

Sent from my iPhone

On Jul 2, 2019, at 4:04 PM, Tom Augspurger ***@***.***> wrote: @WillAyd do you have a good feeling for whether this should raise? It seems like a length-0 DataFrame raises for length-0 grouper. Is there a problem with length-0 groupers, when the obj is also length-0? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[TomAugspurger]

This raising broke all of dask.dataframe's groupby :)

Dask uses an empty series / Frame in many places to infer the output shape / dtypes of an operation.

So I'll defer to you on this raising being appropriate, but I'll make a small PR that deprecates this.

[jreback]

hmm no this is valid i think

[TomAugspurger]

In general, it's nice when ops on a length-0 object match the behavior of a non-empty object (but that's not always possible). FYI, here are the groupby properties

In [24]: gr = s.groupby([])

In [25]: gr.grouper.groupings[0]
Out[25]: Grouping(None)

In [26]: gr.grouper.group_info
Out[26]: (array([], dtype=int64), array([], dtype=int64), 0)

[TomAugspurger]

And FWIW, when grouping by level we don't raise on master

In [1]: import pandas as pd
s
In [2]: s = pd.Series([], index=pd.MultiIndex.from_product([[], []], names=['a', 'b']))

In [3]: s.groupby(level=0).grouper.groupings
Out[3]: [Grouping(a)]

[WillAyd]

Inconsistency there is definitely a bug so I guess the question is which way to go. Jeff - not entirely clear on your response - do you think the empty grouping is valid? IMO a GroupBy is different than Series or Dataframes because it’s not a mutable container. There’s not much you can do with an empty GroupBy at instatiation or later.

…

Sent from my iPhone

On Jul 2, 2019, at 5:19 PM, Tom Augspurger ***@***.***> wrote: And FWIW, when grouping by level we don't raise on master In [1]: import pandas as pd s In [2]: s = pd.Series([], index=pd.MultiIndex.from_product([[], []], names=['a', 'b'])) In [3]: s.groupby(level=0).grouper.groupings Out[3]: [Grouping(a)] — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[jreback]

@WillAyd empty aggregators are valid, I agree with @TomAugspurger here; this is a regression which we should patch (not sure where it came up).