fix iscan python bug (#35148)

python/paddle/fluid/tests/unittests/test_gradient_clip.py

@@ -56,6 +56,7 @@ class TestGradientClip(unittest.TestCase):
         self.BATCH_SIZE = 2
         reader = fake_imdb_reader(self.word_dict_len, self.BATCH_SIZE * 100)
         self.train_data = paddle.batch(reader, batch_size=self.BATCH_SIZE)
+        self.clip_gradient = lambda x: None
         self.init()
 
     def init(self):
@@ -67,9 +68,6 @@ class TestGradientClip(unittest.TestCase):
             places.append(fluid.CUDAPlace(0))
         return places
 
-    def clip_gradient(self, params_grads):
-        pass
-
     def check_clip_result(self, out, out_clip):
         pass
 
@@ -132,7 +130,6 @@ class TestGradientClip(unittest.TestCase):
         data = next(self.train_data())
         val = exe.run(prog, feed=feeder.feed(data), fetch_list=[cost])[0]
         self.assertEqual((1, ), val.shape)
-        print(val)
         self.assertFalse(np.isnan(val))
 
     def backward_and_optimize(self, cost):
@@ -143,11 +140,6 @@ class TestGradientClipByGlobalNorm(TestGradientClip):
     def init(self):
         self.clip_norm = 0.2
 
-    def clip_gradient(self, params_grads):
-        clip = fluid.clip.GradientClipByGlobalNorm(clip_norm=self.clip_norm)
-        print(clip)
-        return clip(params_grads)
-
     def check_clip_result(self, out, out_clip):
         global_norm = 0
         for v in out:
@@ -179,7 +171,6 @@ class TestGradientClipByGlobalNorm(TestGradientClip):
     def test_new_gradient_clip(self):
         def func(params_grads):
             clip = fluid.clip.GradientClipByGlobalNorm(clip_norm=self.clip_norm)
-            print(clip)
             return clip(params_grads)
 
         self.clip_gradient = func
@@ -236,11 +227,6 @@ class TestGradientClipByNorm(TestGradientClip):
     def init(self):
         self.clip_norm = 0.2
 
-    def clip_gradient(self, params_grads):
-        clip = fluid.clip.GradientClipByNorm(clip_norm=self.clip_norm)
-        print(clip)
-        return clip(params_grads)
-
     def check_clip_result(self, out, out_clip):
         for u, v in zip(out, out_clip):
             norm = np.sqrt(np.sum(np.power(u, 2)))
@@ -253,6 +239,11 @@ class TestGradientClipByNorm(TestGradientClip):
 
     # test whether the ouput is right when use grad_clip
     def test_gradient_clip(self):
+        def func(params_grads):
+            clip = fluid.clip.GradientClipByNorm(clip_norm=self.clip_norm)
+            return clip(params_grads)
+
+        self.clip_gradient = func
         self.check_gradient_clip(fluid.CPUPlace())
 
     # if grad is None or not need clip
@@ -280,11 +271,6 @@ class TestGradientClipByValue(TestGradientClip):
         self.max = 0.2
         self.min = 0.1
 
-    def clip_gradient(self, params_grads):
-        clip = fluid.clip.GradientClipByValue(max=self.max, min=self.min)
-        print(clip)
-        return clip(params_grads)
-
     def check_clip_result(self, out, out_clip):
         for i, v in enumerate(out):
             out[i] = np.clip(v, self.min, self.max)
@@ -297,6 +283,11 @@ class TestGradientClipByValue(TestGradientClip):
 
     # test whether the ouput is right when use grad_clip
     def test_gradient_clip(self):
+        def func(params_grads):
+            clip = fluid.clip.GradientClipByValue(max=self.max, min=self.min)
+            return clip(params_grads)
+
+        self.clip_gradient = func
         self.check_gradient_clip(fluid.CPUPlace())
 
     # if grad is None or not need clip

python/paddle/tensor/to_string.py

@@ -101,14 +101,14 @@ def _to_sumary(var):
         return var
     elif len(var.shape) == 1:
         if var.shape[0] > 2 * edgeitems:
-            return np.concatenate([var[:edgeitems], var[-edgeitems:]])
+            return np.concatenate([var[:edgeitems], var[(-1 * edgeitems):]])
         else:
             return var
     else:
         # recursively handle all dimensions
         if var.shape[0] > 2 * edgeitems:
             begin = [x for x in var[:edgeitems]]
-            end = [x for x in var[-edgeitems:]]
+            end = [x for x in var[(-1 * edgeitems):]]
             return np.stack([_to_sumary(x) for x in (begin + end)])
         else:
             return np.stack([_to_sumary(x) for x in var])
@@ -162,10 +162,10 @@ def _format_tensor(var, sumary, indent=0, max_width=0, signed=False):
         if sumary and var.shape[0] > 2 * edgeitems:
             items = [
                 _format_item(item, max_width, signed)
-                for item in list(var)[:DEFAULT_PRINT_OPTIONS.edgeitems]
+                for item in list(var)[:edgeitems]
             ] + ['...'] + [
                 _format_item(item, max_width, signed)
-                for item in list(var)[-DEFAULT_PRINT_OPTIONS.edgeitems:]
+                for item in list(var)[(-1 * edgeitems):]
             ]
         else:
             items = [
@@ -181,7 +181,7 @@ def _format_tensor(var, sumary, indent=0, max_width=0, signed=False):
                 for x in var[:edgeitems]
             ] + ['...'] + [
                 _format_tensor(x, sumary, indent + 1, max_width, signed)
-                for x in var[-edgeitems:]
+                for x in var[(-1 * edgeitems):]
             ]
         else:
             vars = [