Merge pull request #4366 from cpcloud/frame-sum-int-perf-reg

BUG: fix sum over integer frames

Author: cpcloud

doc/source/release.rst

@@ -59,6 +59,8 @@ pandas 0.13
     representation of the index (:issue:`4136`)
   - Fix running of stata IO tests. Now uses temporary files to write
     (:issue:`4353`)
+  - Fixed an issue where ``DataFrame.sum`` was slower than ``DataFrame.mean``
+    for integer valued frames (:issue:`4365`)
 
 pandas 0.12
 ===========

doc/source/v0.13.0.txt

@@ -38,6 +38,9 @@ Bug Fixes
   - Fix running of stata IO tests. Now uses temporary files to write
     (:issue:`4353`)
 
+  - Fixed an issue where ``DataFrame.sum`` was slower than ``DataFrame.mean``
+    for integer valued frames (:issue:`4365`)
+
 See the :ref:`full release notes
 <release>` or issue tracker
 on GitHub for a complete list.

pandas/core/nanops.py

@@ -71,7 +71,9 @@ def f(values, axis=None, skipna=True, **kwds):
 
                 if _USE_BOTTLENECK and skipna and _bn_ok_dtype(values.dtype):
                     result = bn_func(values, axis=axis, **kwds)
-                    # prefer to treat inf/-inf as NA
+
+                    # prefer to treat inf/-inf as NA, but must compute the func
+                    # twice :(
                     if _has_infs(result):
                         result = alt(values, axis=axis, skipna=skipna, **kwds)
                 else:
@@ -86,7 +88,8 @@ def f(values, axis=None, skipna=True, **kwds):
 
 def _bn_ok_dtype(dt):
     # Bottleneck chokes on datetime64
-    return dt != np.object_ and not issubclass(dt.type, (np.datetime64,np.timedelta64))
+    time_types = np.datetime64, np.timedelta64
+    return dt != np.object_ and not issubclass(dt.type, time_types)
 
 
 def _has_infs(result):
@@ -95,10 +98,8 @@ def _has_infs(result):
             return lib.has_infs_f8(result)
         elif result.dtype == 'f4':
             return lib.has_infs_f4(result)
-        else:  # pragma: no cover
-            raise TypeError('Only suppose float32/64 here')
-    else:
-        return np.isinf(result) or np.isneginf(result)
+        return False
+    return np.isinf(result) or np.isneginf(result)
 
 def _get_fill_value(dtype, fill_value=None, fill_value_typ=None):
     """ return the correct fill value for the dtype of the values """

vb_suite/stat_ops.py

@@ -43,6 +43,35 @@
     Benchmark("df[1].sum(level=[0, 1])", setup, repeat=1,
               start_date=datetime(2011, 11, 15))
 
+sum_setup = common_setup + """
+df = DataFrame(np.random.randn(100000, 4))
+dfi = DataFrame(np.random.randint(1000, size=df.shape))
+"""
+
+stat_ops_frame_sum_int_axis_0 = \
+    Benchmark("dfi.sum()", sum_setup, start_date=datetime(2013, 7, 25))
+
+stat_ops_frame_sum_float_axis_0 = \
+    Benchmark("df.sum()", sum_setup, start_date=datetime(2013, 7, 25))
+
+stat_ops_frame_mean_int_axis_0 = \
+    Benchmark("dfi.mean()", sum_setup, start_date=datetime(2013, 7, 25))
+
+stat_ops_frame_mean_float_axis_0 = \
+    Benchmark("df.mean()", sum_setup, start_date=datetime(2013, 7, 25))
+
+stat_ops_frame_sum_int_axis_1 = \
+    Benchmark("dfi.sum(1)", sum_setup, start_date=datetime(2013, 7, 25))
+
+stat_ops_frame_sum_float_axis_1 = \
+    Benchmark("df.sum(1)", sum_setup, start_date=datetime(2013, 7, 25))
+
+stat_ops_frame_mean_int_axis_1 = \
+    Benchmark("dfi.mean(1)", sum_setup, start_date=datetime(2013, 7, 25))
+
+stat_ops_frame_mean_float_axis_1 = \
+    Benchmark("df.mean(1)", sum_setup, start_date=datetime(2013, 7, 25))
+
 #----------------------------------------------------------------------
 # rank