Merge pull request #870 from pascanur/maxArgmax_bug

theano/tensor/basic.py

@@ -2300,6 +2300,9 @@ class MaxAndArgmax(Op):
         x, axis = inp
         g_max, g_max_idx = grads
 
+        # Check to see if the gradient on max is None
+        if g_max is None:
+            return None, None
         xmax = max(x, axis)
 
         # Raise the g_max and xmax to the same number of dim as the input.

theano/tensor/tests/test_basic.py

@@ -2020,6 +2020,27 @@ class T_max_and_argmax(unittest.TestCase):
             v = eval_outputs(max_and_argmax(n, axis)[0].shape)
             assert tuple(v) == numpy.max(data, np_axis).shape
 
+    def test_arg_grad(self):
+        """
+        The test checks if computing the gradient of argmax(x).sum() fails
+        because there is no differentiable path from cost to the input and
+        not because of an error of the grad method of the op
+        """
+        x = matrix()
+        cost = argmax(x, axis=0).sum()
+        value_error_raised = False
+        try:
+            gx = grad(cost, x)
+        except ValueError:
+            # It is the error saying there is no differentiable path to the
+            # input
+            value_error_raised = True
+        if not value_error_raised:
+            raise ValueError(('Test failed because exception saying '
+                              'no differentiable path found was not '
+                              'raised'))
+
+
     def test_grad(self):
         data = rand(2, 3)
         n = as_tensor_variable(data)