使用Tensorflow时出错:没有属性“”variable_scope“”

Question

将Tensorflow 2.x与1.x代码一起使用时，会出现以下错误：

代码：

# load VAE model
import tensorflow.compat.v1 as tf #version - 2.3.0
tf.disable_v2_behavior()
tf.Variable
tf.enable_eager_execution
tf.variable_scope
config = process_config('NAB_config.json')
create_dirs([config['result_dir'], config['checkpoint_dir']])
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
data = DataGenerator(config)
model_vae = VAEmodel(config)
trainer_vae = vaeTrainer(sess, model_vae, data, config)
model_vae.load(sess)

错误：

AttributeError Traceback (most recent call last)
<ipython-input-61-30d2f020e3b6> in <module>
     16 data = DataGenerator(config)
     17 # create a CNN model
---> 18 model_vae = VAEmodel(config)
     19 # create a CNN model
     20 trainer_vae = vaeTrainer(sess, model_vae, data, config)

~\VAE-LSTM-for-anomaly-detection-master\VAE-LSTM-for-anomaly-detection-master\codes\models.py in __init__(self, config)
     12 class VAEmodel(BaseModel):
     13   def __init__(self, config):
---> 14     super(VAEmodel, self).__init__(config)
     15     self.input_dims = self.config['l_win'] * self.config['n_channel']
     16 

~\VAE-LSTM-for-anomaly-detection-master\VAE-LSTM-for-anomaly-detection-master\codes\base.py in __init__(self, config)
     27     self.config = config
     28     # init the global step
---> 29     self.init_global_step()
     30     # init the epoch counter
     31     self.init_cur_epoch()

~\VAE-LSTM-for-anomaly-detection-master\VAE-LSTM-for-anomaly-detection-master\codes\base.py in init_global_step(self)
     60   def init_global_step(self):
     61     # DON'T forget to add the global step tensor to the tensorflow trainer
---> 62     with tf.variable_scope('global_step'):
     63       self.global_step_tensor = tf.Variable(0, trainable=False, name='global_step')
     64       self.increment_global_step_tensor = tf.assign(

Answer 1

TensorFlow 2.0中没有tf.variable_scope()

每个v1.variable_scope都应该转换成一个Python对象。通常，这将是以下之一：

tf.keras.layers.Layer
tf.keras.Model
tf.Module

请按照此处的迁移指南进行操作Migrate your TensorFlow 1 code to TensorFlow 2

Answer 2

已从TensorFlow 2.0中删除了tf.variable_scope()。

official doc

Answer 3

安装v2包后调用v1功能的解决方案如下：

使用tf.compat.v1.variable_scope()而不是tf.variable_scope()

还有一个名为tf_upgrade_v2的工具，可以将任何Tensorflow v1代码迁移到v2。

更多信息：https://www.tensorflow.org/guide/upgrade