From cdd0bcf91a5703226d9116bcc52c3c50b1c0cf8a Mon Sep 17 00:00:00 2001
From: Lev Givon <lev@columbia.edu>
Date: Tue, 20 Oct 2015 14:02:48 -0400
Subject: [PATCH] Add support for ddof param to skcuda.misc.std().

---
 docs/source/changes.rst |  3 ++-
 skcuda/misc.py          | 11 ++++++++---
 tests/test_misc.py      |  6 ++++++
 3 files changed, 16 insertions(+), 4 deletions(-)

diff --git a/docs/source/changes.rst b/docs/source/changes.rst
index f8fd89b3..c3642df5 100644
--- a/docs/source/changes.rst
+++ b/docs/source/changes.rst
@@ -15,7 +15,8 @@ Release 0.5.1 - (under development)
   Erichson).
 * Randomized linear algebra routines (enh. by N. Ben Erichson).
 * Add triu function (enh. by N. Ben Erichson).
-* Support Bessel correction in computation of variance (#143).
+* Support Bessel correction in computation of variance and standard 
+  deviation (#143).
 
 Release 0.5.0 - (July 14, 2015)
 -------------------------------
diff --git a/skcuda/misc.py b/skcuda/misc.py
index a7bbf7b9..c15f0883 100644
--- a/skcuda/misc.py
+++ b/skcuda/misc.py
@@ -1336,7 +1336,7 @@ def _inplace_pow(x_gpu, p, stream):
     return out
 
 
-def std(x_gpu, axis=None, stream=None, keepdims=False):
+def std(x_gpu, ddof=0, axis=None, stream=None, keepdims=False):
     """
     Compute the standard deviation along the specified axis.
 
@@ -1348,6 +1348,11 @@ def std(x_gpu, axis=None, stream=None, keepdims=False):
     ----------
     x_gpu : pycuda.gpuarray.GPUArray
         Array containing numbers whose std is desired.
+    ddof : int (optional)
+        "Delta Degrees of Freedom": the divisor used in computing the 
+        variance is ``N - ddof``, where ``N`` is the number of elements.
+        Setting ``ddof = 1`` is equivalent to applying Bessel's
+        correction.
     axis : int (optional)
         Axis along which the std are computed. The default is to
         compute the std of the flattened array.
@@ -1368,9 +1373,9 @@ def _inplace_pow(x_gpu, p, stream):
                     p, x_gpu.gpudata, x_gpu.gpudata, x_gpu.mem_size)
 
     if axis is None:
-        return var(x_gpu, stream=stream, keepdims=keepdims) ** 0.5
+        return var(x_gpu, ddof=ddof, stream=stream, keepdims=keepdims) ** 0.5
     else:
-        out = var(x_gpu, axis=axis, stream=stream, keepdims=keepdims)
+        out = var(x_gpu, ddof=ddof, axis=axis, stream=stream, keepdims=keepdims)
         _inplace_pow(out, 0.5, stream)
     return out
 
diff --git a/tests/test_misc.py b/tests/test_misc.py
index b1975464..ca36ded7 100644
--- a/tests/test_misc.py
+++ b/tests/test_misc.py
@@ -307,6 +307,12 @@ def impl_test_std(self, dtype):
         assert np.allclose(misc.std(x_gpu).get(), x.std())
         assert np.allclose(misc.std(x_gpu, axis=0).get(), x.std(axis=0))
         assert np.allclose(misc.std(x_gpu, axis=1).get(), x.std(axis=1))
+
+        assert np.allclose(misc.std(x_gpu, ddof=1).get(), x.std(ddof=1))
+        assert np.allclose(misc.std(x_gpu, ddof=1, axis=0).get(),
+                           x.std(ddof=1, axis=0))
+        assert np.allclose(misc.std(x_gpu, ddof=1, axis=1).get(),
+                           x.std(ddof=1, axis=1))
         # Currently not working due to a bug in PyCUDA, see Issue #92
         #x = x.astype(dtype=dtype, order='F')
         #x_gpu = gpuarray.to_gpu(x)