TensorFlow v1到v2版本兼容问题汇总与解决方案

•  首先将TensorFlow Python文件从版本1.x转换为版本2.x

安装了tensorflow后，在命令提示符cmd中使用 tf_upgrade_v2 将TensorFlow Python文件从版本1.x转换为版本2.x，按如下操作： 

tf_upgrade_v2 --infile file_v1.py --outfile file_v2.py

 转换成功后会生成一个report.txt文件，里面记录了做出修改的地方，如下所示：

TensorFlow 2.0 Upgrade Script
-----------------------------
Converted 1 files
Detected 0 issues that require attention
--------------------------------------------------------------------------------
================================================================================
Detailed log follows:

================================================================================
--------------------------------------------------------------------------------
Processing file 'D:\\ProgramData\\Workspace\\bp_network.py'
 outputting to 'D:\\ProgramData\\Workspace\\bp_network_new.py'
--------------------------------------------------------------------------------

14:17: INFO: Renamed 'tf.random_normal' to 'tf.random.normal'
15:17: INFO: Renamed 'tf.random_normal' to 'tf.random.normal'
16:4: INFO: Renamed 'tf.placeholder' to 'tf.compat.v1.placeholder'
17:4: INFO: Renamed 'tf.placeholder' to 'tf.compat.v1.placeholder'
21:17: INFO: Added keywords to args of function 'tf.reduce_mean'
21:37: INFO: Renamed 'tf.log' to 'tf.math.log'
22:13: INFO: Renamed 'tf.train.AdamOptimizer' to 'tf.compat.v1.train.AdamOptimizer'
29:5: INFO: Renamed 'tf.Session' to 'tf.compat.v1.Session'
30:14: INFO: Renamed 'tf.initialize_all_variables' to 'tf.compat.v1.initialize_all_variables'
--------------------------------------------------------------------------------

• RuntimeError: tf.placeholder() is not compatible with eager execution.

从上面可以看到已经将 tf.placeholder() 转换为兼容版本的 tf.compat.v1.placeholder() ，但运行时依然出现了错误 “tf.placeholder() is not compatible with eager execution.”，意思是 placeholder 和 eager execution 不兼容，因为 TensorFlow 2.x 是默认开启 eager execution 的，所以需要禁用该功能，在文件的最前面添加下面这句代码就可以运行了

import tensorflow as tf

tf.compat.v1.disable_eager_execution()  # 禁用tf的eager execution功能

• AttributeError: module 'tensorflow' has no attribute 'contrib'

版本1.x：tf.contrib.layers.l2_regularizer() 转换为版本2.x：tf.keras.regularizers.l2()