Fix Reshape.infer_shape when one of the given shapes is -1

theano/tensor/basic.py

@@ -3788,9 +3788,18 @@ class Reshape(Op):
         g_out, = grads
         return [reshape(g_out, shape(x), ndim=x.ndim), None]
     def infer_shape(self, node, ishapes):
-        #we can't just put node.inputs[1] as not all op support interation
-        #and this is needed in the ShapeOptimizer
-        return [tuple([node.inputs[1][i] for i in range(self.ndim)])]
+        # inputs[1] can contain at most one value of '-1', meaning the actual
+        # shape of the output will be automatically computed by reshape, so
+        # that the total number of elements stays the same.
+        # TODO: Maybe put that formula here?
+        # It's not trivial, because we would have to check if the product of
+        # all the non-minus-one shapes is a divisor of the product of the
+        # original shapes.
+        return [tuple([switch(eq(node.inputs[1][i], -1),
+                                 theano.tensor.opt.Shape_i(i)(node.outputs[0]),
+                                 node.inputs[1][i])
+                            for i in range(self.ndim)]
+            )]
 
 def reshape(x, newshape, ndim=None, name=None):
     if ndim is None:

theano/tensor/tests/test_basic.py

@@ -3290,9 +3290,11 @@ class T_op_cache(unittest.TestCase):
         a = numpy.random.rand(5,2).astype(config.floatX)
         self.assertTrue(numpy.all(fn_py(a) == fn_c_or_py(a)))
 
+class T_reshape(unittest.TestCase):
+    def setUp(self):
+        utt.seed_rng()
 
-def test_reshape():
-
+    def test_reshape(self):
         a = dvector()
         b = dmatrix()
         d = dmatrix()
@@ -3361,9 +3363,30 @@ def test_reshape():
         assert numpy.all(f(numpy.asarray([[0,1,2],[3,4,5]])) == numpy.asarray([[[0],[1],[2]],[[3],[4],[5]]]))
         assert f.maker.env.toposort()[-2].outputs[0].type.broadcastable==(False, False, True)
 
-
         assert numpy.all(f_sub(a_val,b_val)==[2,3])
 
+    def test_infer_shape(self):
+        a = matrix('a')
+        shapes = ivector('shapes')
+        ndim = 2
+
+        r = a.reshape(shapes, ndim=2)
+        z = zeros_like(r)
+
+        f = function([a, shapes], z.shape)
+
+        rng = numpy.random.RandomState(seed=utt.fetch_seed())
+        a_val = rng.uniform(size=(3,4)).astype(config.floatX)
+
+        self.assertTrue((f(a_val, [4, 3]) == [4, 3]).all())
+        self.assertTrue((f(a_val, [-1, 3]) == [4, 3]).all())
+        self.assertTrue((f(a_val, [4, -1]) == [4, 3]).all())
+        self.assertRaises(ValueError, f, a_val, [-1, 5])
+        self.assertRaises(ValueError, f, a_val, [7, -1])
+        self.assertRaises(ValueError, f, a_val, [7, 5])
+        self.assertRaises(ValueError, f, a_val, [-1, -1])
+
+
 def test_make_column_matrix_broadcastable():
     # The goal of the operation made by `b` is to ensure the second dimension
     # of the column matrix is broadcastable.