ENH: groupby().apply(f) accepts combine=0 arg, to return results unmolested #3241

pandas/core/groupby.py

@@ -287,7 +287,7 @@ def apply(self, func, *args, **kwargs):
         """
         Apply function and combine results together in an intelligent way. The
         split-apply-combine combination rules attempt to be as common sense
-        based as possible. For example:
+        based as possible. For example (overridable with combine=False):
 
         case 1:
         group DataFrame
@@ -307,6 +307,9 @@ def apply(self, func, *args, **kwargs):
         Parameters
         ----------
         func : function
+        combine : (default: True), You may pass in a combine=True argument to get back
+                 the values exactly as returned by func, as long as func doesn't itself
+                 use a `combine` keyword  or capture all kwd args using **kwds.
 
         Notes
         -----
@@ -320,14 +323,34 @@ def apply(self, func, *args, **kwargs):
         -------
         applied : type depending on grouped object and function
         """
+        import inspect
+
         func = _intercept_function(func)
+
+        # make sure f doesn't expect a "combine" keyword
+        # and if not, hijack it if specified
+        combine = True
+        try:
+            fargs=inspect.getargspec(func)
+            if not fargs.keywords and 'combine' not in fargs.args[len(fargs.defaults or []):]:
+                combine = kwargs.pop('combine',True)
+        except TypeError: # func is not a python function?
+            pass
+
         f = lambda g: func(g, *args, **kwargs)
-        return self._python_apply_general(f)
 
-    def _python_apply_general(self, f):
+        return self._python_apply_general(f,combine=combine)
+
+    def _python_apply_general(self, f,combine=True):
         keys, values, mutated = self.grouper.apply(f, self.obj, self.axis)
 
-        return self._wrap_applied_output(keys, values,
+        if not combine:
+            if len(keys) == 0:
+                return Series([])
+            else:
+                return zip(keys,values)
+        else:
+            return self._wrap_applied_output(keys, values,
                                          not_indexed_same=mutated)
 
     def aggregate(self, func, *args, **kwargs):

pandas/tests/test_groupby.py

@@ -11,7 +11,8 @@
 from pandas.core.groupby import GroupByError, SpecificationError, DataError
 from pandas.core.series import Series
 from pandas.util.testing import (assert_panel_equal, assert_frame_equal,
-                                 assert_series_equal, assert_almost_equal)
+                                 assert_series_equal, assert_almost_equal,
+                                 makeCustomDataframe as mkdf)
 from pandas.core.panel import Panel
 from pandas.tools.merge import concat
 from collections import defaultdict
@@ -2369,6 +2370,35 @@ def noddy(value, weight):
         # don't die
         no_toes = df_grouped.apply(lambda x: noddy(x.value, x.weight ))
 
+    def test_groupby_apply_raw(self):
+        from random import randint
+        df=mkdf(10,2,data_gen_f=lambda x,y: randint(1,10))
+        df
+        def f1(g):
+            return g.sort('C_l0_g0')
+        def f2(g,combine=None):
+            return g.sort('C_l0_g0')
+        def f3(g,**kwds):
+            return g.sort('C_l0_g0')
+
+        g=df.groupby(lambda key: int(key.split("g")[-1]) >= 5)
+        r=g.apply(f1) # default result without  using combine
+
+        r1=g.apply(f1,combine=False)
+        r2=g.apply(f2,combine=False)
+        r3=g.apply(f3,combine=False)
+
+        # if the combine keyword is in the transformer signature, don't mess with it
+        assert_frame_equal(r,r2)
+        # if  the transformer catches all keywords, don't mess with it
+        assert_frame_equal(r,r3)
+
+        # else, make sure we get a kv pair, with the values
+        # being exactly what the transformer returns
+        self.assertEqual(len(r1),2)
+        assert_frame_equal(r1[0][1],f1(list(g)[0][1]))
+        assert_frame_equal(r1[1][1],f1(list(g)[1][1]))
+
 def assert_fp_equal(a, b):
     assert((np.abs(a - b) < 1e-12).all())