Performance improvements for nunique method.

pandas/core/base.py

@@ -288,21 +288,29 @@ def value_counts(self, normalize=False, sort=True, ascending=False,
                                             tz=getattr(self, 'tz', None))
         return result
 
-    def unique(self):
+    def unique(self, dropna=False):
         """
         Return array of unique values in the object. Significantly faster than
         numpy.unique. Includes NA values.
 
+        Parameters
+        ----------
+        dropna : boolean, default False
+            Don't include NaN in the result.
+
         Returns
         -------
         uniques : ndarray
         """
-        from pandas.core.nanops import unique1d
-        values = self.values
-        if hasattr(values,'unique'):
-            return values.unique()
-
-        return unique1d(values)
+        if dropna:
+            return self.dropna().unique()
+        else:
+            if hasattr(self.values, 'unique'):
+                # Categorical Series not supported by unique1d
+                return self.values.unique()
+            else:
+                from pandas.core.nanops import unique1d
+                return unique1d(self.values)
 
     def nunique(self, dropna=True):
         """
@@ -319,7 +327,7 @@ def nunique(self, dropna=True):
         -------
         nunique : int
         """
-        return len(self.value_counts(dropna=dropna))
+        return len(self.unique(dropna=dropna))
 
     def factorize(self, sort=False, na_sentinel=-1):
         """

pandas/core/categorical.py

@@ -869,7 +869,7 @@ def mode(self):
                              fastpath=True)
         return result
 
-    def unique(self):
+    def unique(self, **kwargs):
         """
         Return the unique values.
 

pandas/core/index.py

@@ -215,6 +215,16 @@ def is_(self, other):
         # use something other than None to be clearer
         return self._id is getattr(other, '_id', Ellipsis)
 
+    def dropna(self):
+        """
+        Return Index without null values
+
+        Returns
+        -------
+        dropped : Index
+        """
+        return self[~isnull(self.values)]
+
     def _reset_identity(self):
         """Initializes or resets ``_id`` attribute with new object"""
         self._id = _Identity()

pandas/tests/test_index.py

@@ -907,6 +907,13 @@ def test_nan_first_take_datetime(self):
         exp = Index([idx[-1], idx[0], idx[1]])
         tm.assert_index_equal(res, exp)
 
+    def test_dropna(self):
+        idx = Index([np.nan, 'a', np.nan, np.nan, 'b', 'c', np.nan],
+                    name='idx')
+        expected = Index(['a', 'b', 'c'], name='idx')
+        result = idx.dropna()
+        tm.assert_index_equal(result, expected)
+
 
 class TestFloat64Index(tm.TestCase):
     _multiprocess_can_split_ = True
@@ -1051,6 +1058,12 @@ def test_astype_from_object(self):
         tm.assert_equal(result.dtype, expected.dtype)
         tm.assert_index_equal(result, expected)
 
+    def test_dropna(self):
+        idx = Float64Index([np.nan, 1.0, np.nan, np.nan, 2.0, 3.0, np.nan])
+        expected = Float64Index([1.0, 2.0, 3.0])
+        result = idx.dropna()
+        tm.assert_index_equal(result, expected)
+
 
 class TestInt64Index(tm.TestCase):
     _multiprocess_can_split_ = True
@@ -1476,6 +1489,12 @@ def test_slice_keep_name(self):
         idx = Int64Index([1, 2], name='asdf')
         self.assertEqual(idx.name, idx[1:].name)
 
+    def test_dropna_does_nothing(self):
+        idx = Int64Index([1, 2, 3], name='idx')
+        expected = Int64Index([1, 2, 3], name='idx')
+        result = idx.dropna()
+        tm.assert_index_equal(result, expected)
+
 
 class TestMultiIndex(tm.TestCase):
     _multiprocess_can_split_ = True
@@ -2948,6 +2967,12 @@ def test_level_setting_resets_attributes(self):
         # if this fails, probably didn't reset the cache correctly.
         assert not ind.is_monotonic
 
+    def test_dropna_does_nothing(self):
+        idx = MultiIndex.from_tuples([('bar', 'two')])
+        expected = idx
+        result = idx.dropna()
+        tm.assert_index_equal(result, expected)
+
 
 def test_get_combined_index():
     from pandas.core.index import _get_combined_index

pandas/tseries/index.py

@@ -852,6 +852,23 @@ def take(self, indices, axis=0):
             return self[maybe_slice]
         return super(DatetimeIndex, self).take(indices, axis)
 
+    def unique(self, dropna=False):
+        """
+        Index.unique with handling for DatetimeIndex metadata
+
+        Parameters
+        ----------
+        dropna : boolean, default False
+            Don't include NaN in the result.
+
+        Returns
+        -------
+        result : DatetimeIndex
+        """
+        result = Int64Index.unique(self, dropna=dropna)
+        return DatetimeIndex._simple_new(result, tz=self.tz,
+                                         name=self.name)
+
     def union(self, other):
         """
         Specialized union for DatetimeIndex objects. If combine

vb_suite/series_methods.py

@@ -27,3 +27,11 @@
                               's2.nsmallest(3, take_last=False)',
                               setup,
                               start_date=datetime(2014, 1, 25))
+
+series_nunique1 = Benchmark('s1.nunique()',
+                            setup,
+                            start_date=datetime(2014, 1, 25))
+
+series_nunique2 = Benchmark('s2.nunique()',
+                            setup,
+                            start_date=datetime(2014, 1, 25))

vb_suite/suite.py

@@ -23,6 +23,7 @@
            'plotting',
            'reindex',
            'replace',
+           'series_methods',
            'sparse',
            'strings',
            'reshape',