ENH: consistent types in output of df.groupby

[ikamensh]

Is your feature request related to a problem?

The problem that occurred to me is due to inconsistent types outputted by for values, match_df in df.groupby(names) - if names is a list of length >1, then values can be a tuple. Otherwise it's an element type, which can be e.g. string.

I had a bug where I iterating over values resulted in going over characters of a string, together with zip it just discarded most of the field.

Describe the solution you'd like

Whenever df.groupby is called with a list, output a tuple of values found. Specifically when called with a list of length 1, output a tuple of length 1 instead of unwrapped element.

API breaking implications

This would change how existing code behaves.

Describe alternatives you've considered

Leave things as they are. Unfortunately it means people need to write special case handling ifs around groupby to get consistent types outputted.

Additional context

import pandas as pd


df = pd.DataFrame(columns=['a','b','c'], index=['x','y'])
df.loc['y'] = pd.Series({'a':1, 'b':5, 'c':2})
print(df)

values, _ = next(iter(df.groupby(['a', 'b'])))
print(type(values))  # <class 'tuple'>

values, _ = next(iter(df.groupby(['a'])))
print(type(values))  # <class 'int'>  # IMHO should be 'tuple'!

values, _ = next(iter(df.groupby('a')))
print(type(values))  # <class 'int'>

[rhshadrach]

I haven't yet considered the implementation or all the ramifications of this, but we do similar things in other places, e.g.

df['a']  # Series
df[['a']]  # DataFrame

df.quantile(0.1)  # Series, index is the columns of df
df.quantile([0.1])  # DataFrame, index consists of the value 0.1

My guess is that a tuple may be appropriate in the case of passing a list to groupby, but that the result should still have an Index (not a MultiIndex).

cc @jbrockmendel for any thoughts here.

[jbrockmendel]

a length-1 tuple seems reasonable to me

[rhshadrach]

Will need to deprecate current behavior for this.

[ahmedibrhm]

@rhshadrach should I add a warning or just fix it?

[rhshadrach]

I think this needs to be deprecated before changing.

[ahmedibrhm]

@rhshadrach does that mean raising a deprecation warning for the case where there is only one group name?

[rhshadrach]

Yes, that's correct - something along the lines of "In a future version of pandas, a length 1 tuple will be returned when grouping by a list of length 1. Don't supply a list with a single grouper to avoid this warning."