diff --git a/fluid/PaddleRec/din/README.md b/fluid/PaddleRec/din/README.md
index 3a1285b420c91e3e9aa00ce56cad23496276bb81..3538ba760ff9b80807a6a56aed4b75400c97ae03 100644
--- a/fluid/PaddleRec/din/README.md
+++ b/fluid/PaddleRec/din/README.md
@@ -36,6 +36,10 @@ DIN通过一个兴趣激活模块(Activation Unit)，用预估目标Candidate AD
 ```
 cd data && sh data_process.sh && cd ..
 ```
+如果执行过程中遇到找不到某个包（例如pandas包）的报错，使用如下命令安装对应的包即可。
+```
+pip install pandas
+```
 
 * Step 2: 产生训练集、测试集和config文件
 ```
@@ -103,6 +107,12 @@ model saved in din_amazon/global_step_50000
 ...
 ```
 
+提示：
+
+* 在单机条件下，使用代码中默认的超参数运行时，产生最优auc的global step大致在440000到500000之间
+
+* 训练超出一定的epoch后会稍稍出现过拟合
+
 ## 预测
 参考如下命令，开始预测.
 
diff --git a/fluid/PaddleRec/din/data/build_dataset.py b/fluid/PaddleRec/din/data/build_dataset.py
index d2d93cbf73c5e25ef527c54df7b8b95367edd484..34c053ccdb2686c10875740f72f1e0abf3cb4f10 100644
--- a/fluid/PaddleRec/din/data/build_dataset.py
+++ b/fluid/PaddleRec/din/data/build_dataset.py
@@ -1,6 +1,6 @@
+from __future__ import print_function
 import random
 import pickle
-from __future__ import print_function
 
 random.seed(1234)
 
diff --git a/fluid/PaddleRec/din/data/convert_pd.py b/fluid/PaddleRec/din/data/convert_pd.py
index e019a42a27a03a615cd7c7c8256c6dd864a50649..d7927c7ef1a9da28732cad9c44be24e72095983a 100644
--- a/fluid/PaddleRec/din/data/convert_pd.py
+++ b/fluid/PaddleRec/din/data/convert_pd.py
@@ -1,3 +1,4 @@
+from __future__ import print_function
 import pickle
 import pandas as pd
 
@@ -13,10 +14,12 @@ def to_df(file_path):
         return df
 
 
+print("start to analyse reviews_Electronics_5.json")
 reviews_df = to_df('./raw_data/reviews_Electronics_5.json')
 with open('./raw_data/reviews.pkl', 'wb') as f:
     pickle.dump(reviews_df, f, pickle.HIGHEST_PROTOCOL)
 
+print("start to analyse meta_Electronics.json")
 meta_df = to_df('./raw_data/meta_Electronics.json')
 meta_df = meta_df[meta_df['asin'].isin(reviews_df['asin'].unique())]
 meta_df = meta_df.reset_index(drop=True)
diff --git a/fluid/PaddleRec/din/data/remap_id.py b/fluid/PaddleRec/din/data/remap_id.py
index bb6d36b7b5a48370d4733c4007442cef28b7274f..b110dac54de8f8d201ede7248d6a2844ac350c90 100644
--- a/fluid/PaddleRec/din/data/remap_id.py
+++ b/fluid/PaddleRec/din/data/remap_id.py
@@ -1,7 +1,7 @@
+from __future__ import print_function
 import random
 import pickle
 import numpy as np
-from __future__ import print_function
 
 random.seed(1234)
 
diff --git a/fluid/PaddleRec/din/train.py b/fluid/PaddleRec/din/train.py
index a8c2b752a0e8657482c318f39c3e318e145efb94..db3ac0b178fd77db88ee12af6faaeed742ad2bcf 100644
--- a/fluid/PaddleRec/din/train.py
+++ b/fluid/PaddleRec/din/train.py
@@ -121,7 +121,7 @@ def train():
                 loss_sum = 0.0
 
                 if (global_step > 400000 and global_step % PRINT_STEP == 0) or (
-                        global_step < 400000 and global_step % 50000 == 0):
+                        global_step <= 400000 and global_step % 50000 == 0):
                     save_dir = args.model_dir + "/global_step_" + str(
                         global_step)
                     feed_var_name = [