Merge pull request #4678 from lamblin/fix_advincsubtensor1_reshape

theano/tensor/subtensor.py

@@ -1170,7 +1170,7 @@ def inc_subtensor(x, y, inplace=False, set_instead_of_inc=False,
             # This if is needed to prevent some useless warning about
             # old code bug.
             expanded_y = alloc(y, *[x.shape[i] for i in xrange(x.ndim)])
-            flattened_y = expanded_y.flatten(inner_x.ndim)
+            flattened_y = expanded_y.reshape(inner_x.shape)
         else:
             flattened_y = y
 

theano/tensor/tests/test_subtensor.py

@@ -1221,6 +1221,8 @@ class TestIncSubtensor1(unittest.TestCase):
     # also tests set_subtensor
 
     def setUp(self):
+        self.rng = numpy.random.RandomState(seed=utt.fetch_seed())
+
         self.s = tensor.iscalar()
         self.v = tensor.fvector()
         self.m = tensor.dmatrix()
@@ -1267,6 +1269,21 @@ class TestIncSubtensor1(unittest.TestCase):
         self.assertRaises(TypeError,
                           lambda: inc_subtensor(self.v[self.adv1q], fmatrix()))
 
+    def test_matrix_idx(self):
+        idx = tensor.lmatrix()
+        a = self.m[idx]
+        a2 = inc_subtensor(a, a)
+        f = theano.function([self.m, idx], a2)
+
+        mval = self.rng.random_sample((4, 10))
+        idxval = numpy.array([[1, 2], [3, 2]])
+        a2val = f(mval, idxval)
+
+        utt.assert_allclose(a2val[0], mval[0])
+        utt.assert_allclose(a2val[1], mval[1] * 2)
+        utt.assert_allclose(a2val[2], mval[2] * 3)
+        utt.assert_allclose(a2val[3], mval[3] * 2)
+
 
 inplace_increment_missing = SkipTest(
     "inc_subtensor with advanced indexing not enabled. "