Merge pull request #94 from sklam/enh/ddof

Add ddof option and make its default to 1 (like pandas)

pygdf/numerical.py

@@ -143,12 +143,13 @@ def sum(self):
     def mean(self):
         return self.sum().astype('f8') / self.valid_count
 
-    def mean_var(self):
+    def mean_var(self, ddof=1):
         x = self.astype('f8')
         mu = x.mean()
         n = x.valid_count
         asum = _gdf.apply_reduce(libgdf.gdf_sum_squared_generic, x)
-        var = asum / n - mu ** 2
+        div = n - ddof
+        var = asum / div - (mu ** 2) * n / div
         return mu, var
 
     def applymap(self, udf, out_dtype=None):

pygdf/series.py

@@ -655,21 +655,21 @@ def mean(self):
         """
         return self._column.mean()
 
-    def std(self):
+    def std(self, ddof=1):
         """Compute the standard deviation of the series
         """
-        return np.sqrt(self.var())
+        return np.sqrt(self.var(ddof=ddof))
 
-    def var(self):
+    def var(self, ddof=1):
         """Compute the variance of the series
         """
-        mu, var = self.mean_var()
+        mu, var = self.mean_var(ddof=ddof)
         return var
 
-    def mean_var(self):
+    def mean_var(self, ddof=1):
         """Compute mean and variance at the same time.
         """
-        mu, var = self._column.mean_var()
+        mu, var = self._column.mean_var(ddof=ddof)
         return mu, var
 
     def unique_k(self, k):

pygdf/tests/test_stats.py

@@ -26,13 +26,27 @@ def test_series_reductions(method, dtype):
 
     def call_test(sr):
         fn = getattr(sr, method)
-        return fn()
+        if method in ['std', 'var']:
+            return fn(ddof=1)
+        else:
+            return fn()
 
     expect, got = call_test(arr2), call_test(sr)
     print(expect, got)
     np.testing.assert_approx_equal(expect, got)
 
 
+@pytest.mark.parametrize('ddof', range(3))
+def test_series_std(ddof):
+    np.random.seed(0)
+    arr = np.random.random(100) - 0.5
+    sr = Series(arr)
+    pd = sr.to_pandas()
+    got = sr.std(ddof=ddof)
+    expect = pd.std(ddof=ddof)
+    np.testing.assert_approx_equal(expect, got)
+
+
 def test_series_unique():
     for size in [10 ** x for x in range(5)]:
         arr = np.random.randint(low=-1, high=10, size=size)