fix error_clip errors and add unit test

python/paddle/v2/fluid/clip.py

@@ -3,7 +3,8 @@ import layers
 from . import core
 
 __all__ = [
-    'GradientClipByValue', 'append_gradient_clip_ops', 'error_clip_callback'
+    'GradientClipByValue', 'ErrorClipByValue', 'append_gradient_clip_ops',
+    'error_clip_callback'
 ]
 
 
@@ -23,12 +24,12 @@ class ErrorClipByValue(BaseErrorClipAttr):
         self.min = min
 
     def append_clip_op(self, block, grad_name):
-        block.append_op(
-            type="clip",
-            inputs={"X": grad_name},
-            outputs={"Out": grad_name},
-            attrs={"min": self.min,
-                   "max": self.max})
+        clip_op_desc = block.desc.append_op()
+        clip_op_desc.set_type("clip")
+        clip_op_desc.set_input("X", [grad_name])
+        clip_op_desc.set_output("Out", [grad_name])
+        clip_op_desc.set_attr("min", self.min)
+        clip_op_desc.set_attr("max", self.max)
 
 
 def error_clip_callback(block, context):

python/paddle/v2/fluid/framework.py

@@ -271,6 +271,9 @@ class Variable(object):
         uid = core.unique_integer(prefix)  # unique during whole process.
         return "_".join([prefix, str(uid)])
 
+    def set_error_clip(self, error_clip):
+        self.error_clip = error_clip
+
 
 def get_all_op_protos():
     """

python/paddle/v2/fluid/tests/test_clip.py

@@ -0,0 +1,59 @@
+from __future__ import print_function
+import numpy as np
+import paddle.v2 as paddle
+import paddle.v2.fluid as fluid
+
+BATCH_SIZE = 128
+CLIP_MAX = 2e-6
+CLIP_MIN = -1e-6
+
+prog = fluid.framework.Program()
+
+with fluid.program_guard(main_program=prog):
+    image = fluid.layers.data(name='x', shape=[784], dtype='float32')
+
+    hidden1 = fluid.layers.fc(input=image, size=128, act='relu')
+    hidden2 = fluid.layers.fc(input=hidden1, size=64, act='relu')
+    predict = fluid.layers.fc(input=hidden2, size=10, act='softmax')
+
+    label = fluid.layers.data(name='y', shape=[1], dtype='int64')
+
+    cost = fluid.layers.cross_entropy(input=predict, label=label)
+    avg_cost = fluid.layers.mean(x=cost)
+
+prog_clip = prog.clone()
+prog_clip.block(0).var(hidden1.name).set_error_clip(
+    fluid.clip.ErrorClipByValue(
+        max=CLIP_MAX, min=CLIP_MIN))
+
+avg_cost_clip = prog_clip.block(0).var(avg_cost.name)
+fluid.backward.append_backward(loss=avg_cost)
+fluid.backward.append_backward(
+    loss=avg_cost_clip, callback=fluid.clip.error_clip_callback)
+
+hidden1_grad = prog.block(0).var(hidden1.name + "@GRAD")
+hidden1_grad_clip = prog_clip.block(0).var(hidden1.name + "@GRAD")
+
+train_reader = paddle.batch(
+    paddle.reader.shuffle(
+        paddle.dataset.mnist.train(), buf_size=8192),
+    batch_size=BATCH_SIZE)
+
+place = fluid.CPUPlace()
+exe = fluid.Executor(place)
+feeder = fluid.DataFeeder(feed_list=[image, label], place=place)
+exe.run(fluid.default_startup_program())
+
+count = 0
+for data in train_reader():
+    count += 1
+    if count > 5:
+        break
+    out = exe.run(prog, feed=feeder.feed(data), fetch_list=[hidden1_grad])
+    out_clip = exe.run(prog_clip,
+                       feed=feeder.feed(data),
+                       fetch_list=[hidden1_grad_clip])
+    if not (out[0].clip(min=CLIP_MIN, max=CLIP_MAX) == out_clip[0]).all():
+        exit(1)
+
+exit(0)