Correct expected cost values in test_torch_case of test_ctc

theano/tensor/nnet/tests/test_ctc.py

@@ -14,7 +14,7 @@ class TestCTC(unittest.TestCase):
         if not ctc_enabled:
             self.skipTest('Optional library warp-ctc not available')
 
-    def run_ctc(self, activations, labels, input_length, expected_costs):
+    def run_ctc(self, activations, labels, input_length, expected_costs, expected_grads):
         # Check if softmax probabilites are approximately equal to the gradients
         # of the activations, using utt.assert_allclose(a, b)
 
@@ -31,9 +31,12 @@ class TestCTC(unittest.TestCase):
         test = theano.function([], [t_cost])
 
         cost, grad = train()
-        cost, = test()
+        test_cost, = test()
 
-        utt.assert_allclose(cost, expected_costs)
+        print( grad )
+
+        #utt.assert_allclose(grad, expected_grads)
+        utt.assert_allclose( expected_costs, cost )
 
     # Test obtained from Torch tutorial at:
     # https://github.com/baidu-research/warp-ctc/blob/master/torch_binding/TUTORIAL.md
@@ -50,7 +53,10 @@ class TestCTC(unittest.TestCase):
                              [3, 3],
                              [2, 3]], dtype=np.int32)
 
-        expected_costs = np.asarray([3.03655, 7.35574, 4.93884],
+        expected_costs = np.asarray([1.609437943, 7.355742931, 4.938849926],
                                     dtype=np.float32)
 
-        self.run_ctc(activations, labels, activation_times, expected_costs)
+        grads = [0.200000003, -0.8000000119, 0.200000003, 0.200000003, 0.200000003]
+        expected_gradients = np.asarray(grads, dtype=np.float32)
+
+        self.run_ctc(activations, labels, activation_times, expected_costs, expected_gradients)