diff --git a/python/paddle/fluid/regularizer.py b/python/paddle/fluid/regularizer.py
index dc641cdd1afbecfc9122c9f2e8ce6fac77b53f21..029db7d2dd4b7def8cea374e3f2ed31226f2bc18 100644
--- a/python/paddle/fluid/regularizer.py
+++ b/python/paddle/fluid/regularizer.py
@@ -176,7 +176,6 @@ class L1DecayRegularizer(WeightDecayRegularizer):
             dtype="float32", shape=param.shape, lod_level=param.lod_level)
 
         if grad.type == core.VarDesc.VarType.SELECTED_ROWS:
-            # add concat_rows
             decay = block.create_var(
                 dtype="float32",
                 shape=param.shape,
diff --git a/python/paddle/fluid/tests/book/test_machine_translation.py b/python/paddle/fluid/tests/book/test_machine_translation.py
index caa9596a100de4f9364467690db1e80ee227c3c1..fa38bd3762423497b82c3b421b3a1db4cd87525b 100644
--- a/python/paddle/fluid/tests/book/test_machine_translation.py
+++ b/python/paddle/fluid/tests/book/test_machine_translation.py
@@ -181,7 +181,10 @@ def train_main(use_cuda, is_sparse, is_local=True):
     cost = pd.cross_entropy(input=rnn_out, label=label)
     avg_cost = pd.mean(cost)
 
-    optimizer = fluid.optimizer.Adagrad(learning_rate=1e-4)
+    optimizer = fluid.optimizer.Adagrad(
+        learning_rate=1e-4,
+        regularization=fluid.regularizer.L2DecayRegularizer(
+            regularization_coeff=0.1))
     optimize_ops, params_grads = optimizer.minimize(avg_cost)
 
     train_data = paddle.batch(