[pyspark][group]pyspark GroupedData can't apply agg functions on all left numeric columns.

[citoubest]

What changes were proposed in this pull request?

With pyspark dataframe, the agg method just support two ways, one is to give the column and agg method maps and another one is to use agg functions in package functions to apply on specific columns names. The two approach both ask us to asign a method on specific columns names. But if
I want to apply the agg method on all other numeric columns, I should list all method-column combinations. such as, suppose the df has to columns province,age,income, I want to groupby the province and calculate the min, max and average values on age and income, before change I have to approach: df.groupby('province').agg({'age':'max','age':'min','age':'avg','income':'max','income':'min','income':'avg'})
df.groupby('province').agg({F.min('age'),F.max('age'),F.avg('age'),F.min('income'),F.max('income'),F.avg('income')}), which are both redundant.

with this change, we can simply replace the code with df.groupby('province').agg('max','min','avg')

How was this patch tested?

manual tests

[AmplabJenkins]

Can one of the admins verify this patch?

[petermaxlee]

Isn't it as simple as

cols = [x for x in df.columns if x != "key]
df.groupby("key").agg([F.min(x) for x in cols] + [F.max(x) for x in cols])

[citoubest]

@petermaxlee
In my opinion, list comprehension can reduce code length to some extent. It's better if the agg method can support the easy way in api level.

[citoubest]

@rxin @davies @srowen

[rxin]

I understand the reasons why you want to add this -- but I feel this is too esoteric and if we add this one, there are also a lot of other cases that can be added and I don't know where we would stop.

[citoubest]

OK, because pandas dataframe support the added approach to agg, so I suppose maybe spark dataframe should support, but it not. So I have tried to add this patch. If you think this patch is not necessary , I will close this request later. @rxin .

[rxin]

Pandas doesn't support this, does it?

>>> pd.read_csv('test.csv').groupby('a').agg('sum', 'avg')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 3597, in aggregate
    return super(DataFrameGroupBy, self).aggregate(arg, *args, **kwargs)
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 3114, in aggregate
    result, how = self._aggregate(arg, _level=_level, *args, **kwargs)
  File "/Library/Python/2.7/site-packages/pandas/core/base.py", line 428, in _aggregate
    return getattr(self, arg)(*args, **kwargs), None
TypeError: f() takes exactly 1 argument (2 given)
>>> pd.read_csv('test.csv').groupby('a').agg(['sum', 'avg'])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 3597, in aggregate
    return super(DataFrameGroupBy, self).aggregate(arg, *args, **kwargs)
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 3114, in aggregate
    result, how = self._aggregate(arg, _level=_level, *args, **kwargs)
  File "/Library/Python/2.7/site-packages/pandas/core/base.py", line 564, in _aggregate
    return self._aggregate_multiple_funcs(arg, _level=_level), None
  File "/Library/Python/2.7/site-packages/pandas/core/base.py", line 609, in _aggregate_multiple_funcs
    results.append(colg.aggregate(arg))
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 2574, in aggregate
    (_level or 0) + 1)
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 2636, in _aggregate_multiple_funcs
    results[name] = obj.aggregate(func)
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 2570, in aggregate
    return getattr(self, func_or_funcs)(*args, **kwargs)
  File "/Library/Python/2.7/site-packages/pandas/core/groupby.py", line 498, in __getattr__
    (type(self).__name__, attr))
AttributeError: 'SeriesGroupBy' object has no attribute 'avg'

[citoubest]

with pandas, the param for agg is the function not a str (function names). @rxin 。

Will you consider to add this patch, if not maybe I should close it.

In [13]: df
Out[13]: 
          a         b         c  d
0  0.068300  0.263883  0.237335  1
1  0.226992  0.573966  0.954791  2
2  0.907550  0.930591  0.886454  1
3  0.178581  0.440734  0.414763  2

In [14]: df.groupby('d').agg([max,min])
Out[14]: 
          a                   b                   c          
        max       min       max       min       max       min
d                                                            
1  0.907550  0.068300  0.930591  0.263883  0.886454  0.237335
2  0.226992  0.178581  0.573966  0.440734  0.954791  0.414763

[citoubest]

@davies, what do you think about this patch? Can you give me some advice? Thanks