ENH: use size instead of cythonized count for fallback cases

[cpcloud]

• use size instead of cythonized count for integer case since we cannot have
  nan in that case
• not faster perf wise
• compilation time is shorter because count has no int templates except for
  dates and times
• any precision issues (don't think there were any) with other integer types are gone, since only int64 is compared with dates, whereas lower precision integers use size.

Anywho, here's the vbench results for this PR vs master:

-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------
groupby_int_count                            |   4.0247 |   4.0337 |   0.9978 |
groupby_multi_count                          |   7.5080 |   7.4813 |   1.0036 |
-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------

[cpcloud]

@jreback ok to merge?

[cpcloud]

actually let me add a test in for lower prec ints

[jreback]

looks fine...maybe squash and merge

[jreback]

issues on windows:

• I think the fallback is not working, e.g. should be x.size, not x.size(); so that is untested
• I think their is a casting issue somewhere here (e.g. regular floats are not getting upcasted to float64), default on windows 32 is float32

>>> c:\users\jeff reback\documents\github\pandas\build\lib.win32-2.6\pandas\core\groupby.py(107)f()
-> result = self.aggregate(lambda x: npfunc(x, axis=self.axis))
(Pdb) l
102             try:
103                 return self._cython_agg_general(alias, numeric_only=numeric_only)
104             except AssertionError as e:
105                 raise SpecificationError(str(e))
106             except Exception:
107  ->             result = self.aggregate(lambda x: npfunc(x, axis=self.axis))
108                 if _convert:
109                     result = result.convert_objects()
110                 return result
111
112         f.__doc__ = "Compute %s of group values" % name
(Pdb) p self._cython_agg_general(alias,numeric_only=numeric_only)
*** ValueError: ValueError("Buffer dtype mismatch, expected 'float64_t' but got 'float'",)
(Pdb) d
> c:\users\jeff reback\documents\github\pandas\build\lib.win32-2.6\pandas\core\groupby.py(2405)aggregate()
-> result = self._aggregate_generic(arg, *args, **kwargs)
(Pdb) d
> c:\users\jeff reback\documents\github\pandas\build\lib.win32-2.6\pandas\core\groupby.py(2464)_aggregate_generic()
-> return self._aggregate_item_by_item(func, *args, **kwargs)
(Pdb) d
> c:\users\jeff reback\documents\github\pandas\build\lib.win32-2.6\pandas\core\groupby.py(2508)_aggregate_item_by_item()
-> raise errors

======================================================================
ERROR: test_mixed_type_join_with_suffix (pandas.tools.tests.test_merge.TestMerge)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\tools\tests\test_merge.py", line 686, in test_mixed_type_join_with_suf
fix
    cn = grouped.count()
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 729, in count
    return self._count().astype('int64')
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 107, in f
    result = self.aggregate(lambda x: npfunc(x, axis=self.axis))
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2405, in aggregate
    result = self._aggregate_generic(arg, *args, **kwargs)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2464, in _aggregate_generic
    return self._aggregate_item_by_item(func, *args, **kwargs)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2508, in _aggregate_item_by_item
    raise errors
TypeError: 'int' object is not callable

======================================================================
ERROR: test_count (pandas.tests.test_groupby.TestGroupBy)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\tests\test_groupby.py", line 2012, in test_count
    count_as = df.groupby('A').count()
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 729, in count
    return self._count().astype('int64')
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 107, in f
    result = self.aggregate(lambda x: npfunc(x, axis=self.axis))
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2405, in aggregate
    result = self._aggregate_generic(arg, *args, **kwargs)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2464, in _aggregate_generic
    return self._aggregate_item_by_item(func, *args, **kwargs)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2508, in _aggregate_item_by_item
    raise errors
TypeError: 'int' object is not callable

======================================================================
ERROR: test_count_object (pandas.tests.test_groupby.TestGroupBy)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\tests\test_groupby.py", line 2026, in test_count_object
    result = df.groupby('c').a.count()
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 729, in count
    return self._count().astype('int64')
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 107, in f
    result = self.aggregate(lambda x: npfunc(x, axis=self.axis))
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2063, in aggregate
    result = self._aggregate_named(func_or_funcs, *args, **kwargs)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 2147, in _aggregate_named
    output = func(group, *args, **kwargs)
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 107, in <lambda>
    result = self.aggregate(lambda x: npfunc(x, axis=self.axis))
  File "c:\Users\Jeff Reback\Documents\GitHub\pandas\build\lib.win32-2.6\pandas\core\groupby.py", line 725, in <lambda>
    _count = _groupby_function('_count', 'count', lambda x, axis=0: x.size(),
TypeError: 'int' object is not callable

[cpcloud]

@jreback small favor: do you have a box or something that i can log into and test this? i don't have a copy of windows anywhere that i can test this stuff on

[jreback]

wish I had an easy way for you to do that.

[cpcloud]

ok i'll see if i can repro this on linux

[jreback]

the first one I think you can 'fake' a test (e.g. force the cython agg to fail), then the rest raises

maybe passing in float32 should fail the othe test

lmk

[jreback]

I can test when you have changes

[cpcloud]

@jreback does np.empty(n) default to float32 on Win32?

[jreback]

same on 2.7-64 & numpy 1.8.0

C:\Users\Jeff Reback\Documents\GitHub\pandas>c:\python26-32\python.exe
Python 2.6.6 (r266:84297, Aug 24 2010, 18:46:32) [MSC v.1500 32 bit (Intel)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy as np
>>> np.__version__
'1.6.2'
>>> np.empty((1,1)).dtype
dtype('float64')
>>> np.empty(1).dtype
dtype('float64')
>>> np.arange(1).dtype
dtype('int32')
>>> np.arange(1.).dtype
dtype('float64')
>>>

[cpcloud]

thx