Merge pull request #4852 from Faruk-Ahmed/add_ddof_to_var_and_std

theano/tensor/basic.py

@@ -3164,7 +3164,7 @@ def mean(input, axis=None, dtype=None, op=False, keepdims=False,
 
 
 @constructor
-def var(input, axis=None, keepdims=False):
+def var(input, axis=None, ddof=0, keepdims=False):
     """
     Computes the variance along the given axis(es) of a tensor `input`.
 
@@ -3173,6 +3173,8 @@ def var(input, axis=None, keepdims=False):
     axis: None or int or (list of int) (see `Sum`)
         Compute the variance along this axis of the tensor.
         None means all axes (like numpy).
+    ddof: Degrees of freedom; 0 would compute the ML estimate, 1 would compute
+        the unbiased estimate.
     keepdims : bool
         If this is set to True, the axes which are reduced are
         left in the result as dimensions with size one. With this option,
@@ -3187,6 +3189,9 @@ def var(input, axis=None, keepdims=False):
 
     """
 
+    if isinstance(ddof, (bool)):
+        raise ValueError('Parameter keepdims is now at index 3: (input, axis=None, ddof=0, keepdims=False)')
+
     input_ndim = input.type.ndim
     if axis is None:
         axis = list(range(input_ndim))
@@ -3204,13 +3209,19 @@ def var(input, axis=None, keepdims=False):
     centered_input = input - mean_input
 
     # return the mean sqr
-    v = mean((centered_input ** 2), axis, keepdims=keepdims)
+    if ddof == 0:
+        v = mean((centered_input ** 2), axis, keepdims=keepdims)
+    else:
+        shp = shape(input) - ddof
+        v = sum((centered_input ** 2), axis=axis, keepdims=keepdims)
+        for i in axis:
+            v = true_div(v, shp[i])
     v.name = 'var'
     return v
 
 
 @constructor
-def std(input, axis=None, keepdims=False):
+def std(input, axis=None, ddof=0, keepdims=False):
     """
     Computes the standard deviation along the given axis(es) of a tensor `input`.
 
@@ -3234,7 +3245,10 @@ def std(input, axis=None, keepdims=False):
 
     """
 
-    ret = sqrt(var(input=input, axis=axis, keepdims=keepdims))
+    if isinstance(ddof, (bool)):
+        raise ValueError('Parameter keepdims is now at index 3: (input, axis=None, ddof=0, keepdims=False)')
+
+    ret = sqrt(var(input=input, axis=axis, ddof=ddof, keepdims=keepdims))
     ret.name = 'std'
     return ret
 

theano/tensor/opt.py

@@ -1970,17 +1970,17 @@ def local_subtensor_make_vector(node):
                     ret = [x.owner.inputs[v]]
                 except IndexError:
                     raise NotScalarConstantError("Bad user graph!")
-
                 return ret
             except NotScalarConstantError:
                 pass
         elif idx.ndim == 1 and isinstance(idx, T.Constant):
             values = list(map(int, list(idx.value)))
-            ret = [make_vector(*[x.owner.inputs[v] for v in values])]
+            ret = make_vector(*[x.owner.inputs[v] for v in values])
 
             # Copy over stack trace from previous output to new output
             copy_stack_trace(node.outputs[0], ret)
-            return ret
+            ret = T.patternbroadcast(ret, node.outputs[0].broadcastable)
+            return [ret]
         else:
             raise TypeError('case not expected')
     elif isinstance(idx, slice):
@@ -1993,6 +1993,7 @@ def local_subtensor_make_vector(node):
             ret = make_vector(*x.owner.inputs[const_slice])
             # Copy over stack trace from previous outputs to new output
             copy_stack_trace(node.outputs, ret)
+            ret = T.patternbroadcast(ret, node.outputs[0].broadcastable)
             return [ret]
         except NotScalarConstantError:
             pass

theano/tensor/tests/test_basic.py

@@ -6365,6 +6365,15 @@ def test_var():
     f = function([a], var(a, axis=2))
     assert numpy.allclose(numpy.var(a_val, axis=2), f(a_val))
 
+    f = function([a], var(a, axis=0, ddof=0))
+    assert numpy.allclose(numpy.var(a_val, axis=0, ddof=0), f(a_val))
+
+    f = function([a], var(a, axis=1,ddof=1))
+    assert numpy.allclose(numpy.var(a_val, axis=1,ddof=1), f(a_val))
+
+    f = function([a], var(a, axis=2, ddof=1))
+    assert numpy.allclose(numpy.var(a_val, axis=2, ddof=1), f(a_val))
+
 
 class T_sum(unittest.TestCase):
     def test_sum_overflow(self):