Merge pull request #5269 from Saizheng/master

theano/sandbox/rng_mrg.py

@@ -317,9 +317,13 @@ class mrg_uniform_base(Op):
         # this op should not be called directly.
         #
         # call through MRG_RandomStreams instead.
+        broad = []
+        for i in range(self.output_type.ndim):
+                broad.append(tensor.extract_constant(size[i]) == 1)
+        output_type = self.output_type.clone(broadcastable=broad)()
         return Apply(self,
                      [rstate, size],
-                     [rstate.type(), self.output_type()])
+                     [rstate.type(), output_type])
 
     def grad(self, inputs, ograd):
         return [gradient.grad_undefined(self, k, inp,

theano/sandbox/tests/test_rng_mrg.py

@@ -627,6 +627,22 @@ def test_uniform():
                   allow_01=True, inputs=input)
 
 
+def test_uniform_broadcastable():
+    x = tensor.matrix()
+    size1 = (10, 1)
+    size2 = (x.shape[0], 1)
+
+    R = MRG_RandomStreams(234, use_cuda=False)
+
+    # check when all dimensions are constant
+    uu = R.uniform(size=size1)
+    assert uu.broadcastable == (False, True)
+
+    # check when some dimensions are theano variables
+    uu = R.uniform(size=size2)
+    assert uu.broadcastable == (False, True)
+
+
 @attr('slow')
 def test_binomial():
     # TODO: test size=None, ndim=X