Added logic to preserve broadcastable flag if repeats=1. Added unit test.

theano/tensor/extra_ops.py

@@ -261,8 +261,15 @@ class RepeatOp(theano.Op):
         if self.axis is None:
             broadcastable=[False]
         else:
-            broadcastable = list(x.broadcastable)
-            broadcastable[self.axis] = False
+            try:
+                const_reps = basic.get_constant_value(repeats)
+            except TypeError:
+                const_reps = None
+            if const_reps == 1:
+                broadcastable = x.broadcastable
+            else:
+                broadcastable = list(x.broadcastable)
+                broadcastable[self.axis] = False
             
         out_type = theano.tensor.TensorType(x.dtype, broadcastable)
 

theano/tensor/tests/test_extra_ops.py

@@ -282,6 +282,14 @@ class TestRepeatOp(utt.InferShapeTester):
             for axis in self._possible_axis(ndim):
                 utt.verify_grad(lambda x: RepeatOp(axis=axis)(x, 3), [a])
 
+    def test_broadcastable(self):
+        x = T.TensorType(config.floatX, [False, True, False])()
+        r = RepeatOp(axis=1)(x, 2)
+        self.assertEqual(r.broadcastable, (False, False, False))
+        r =  RepeatOp(axis=1)(x, 1)
+        self.assertEqual(r.broadcastable, (False, True, False))
+        r =  RepeatOp(axis=0)(x, 2)
+        self.assertEqual(r.broadcastable, (False, True, False))
 
 class TestBartlett(utt.InferShapeTester):