ModuleNotFoundError: No module named 'tensorflow.keras' error message solution

ModuleNotFoundError: No module named 'tensorflow.keras’The solution to the error message

Index 目录索引
错误信息
解决方法
适用于Windows用户的方法
适用于Linux用户的方法

using deep learning methods,Training an artificial neural network model egEfficientNet的时候,由于kerasThe reason for the different installation directories of files such as libraries,Various errors may be reported due to incompatibility,This series focuses on documenting the various error resolutions I encountered during training,供大家参考和学习.
错误信息：
话不多说,The error message is thrown first：

Traceback (most recent call last):
  File "**/training.py", line 4, in <module>
    from tensorflow.keras import models
ModuleNotFoundError: No module named 'tensorflow.keras'

Error messages are a good thing,You have to learn how to unearth useful things from error messages.
The code block corresponding to this error message is as follows：

from tensorflow.keras import models
from tensorflow.keras import layers
from tensorflow.keras import optimizers
from tensorflow.keras.preprocessing.image import ImageDataGenerator

解决方法：
We found that the above error occurred when importing the package,是因为kerasThe reason for the library installation directory incompatibility,We find ourselves firstkeras的本地安装目录,这里分别就Windows用户和LinuxThe user introduction method is as follows：

适用于Windows用户的方法：
Let's create a new one firstpython文件keras_dir.py,用于输出keras的本地安装目录,在新建的python文件中输入如下代码,即可查询keraswhere the local directory is located（It should be noted in the following codepathBoth sides are double underlined）.

from tensorflow import keras
print(keras.__path__)

运行keras_dir.pyThe output is displayed as follows：

D:\Anaconda3-5.1.0\envs\tensorflow-gpu\python.exe  **/keras_dir.py
['D:\\Anaconda3-5.1.0\\envs\\tensorflow-gpu\\lib\\site-packages\\tensorflow\\tools\\api\\generator\\api\\keras']


Process finished with exit code 0

通过输出的信息,我们可以得知kerasis installed on the local diskD盘下的D:\Anaconda3-5.1.0\envs\tensorflow-gpu\Lib\site-packages\tensorflow\tools\api\generator\api\keras这个目录中.Different computers may find different outputkerasThe directory where it is located can be compared

Look closely at this directory path,我们可以发现keras并不是直接在tensorflow目录下的,在tensorflow和kerasThere are several directories in between,So we can pass the correct inputkerasdirectory location to successfully invokekeras库了,文章最开始的kerasThe calling statement is modified as follows：

from tensorflow.tools.api.generator.api.keras import models
from tensorflow.tools.api.generator.api.keras import layers
from tensorflow.tools.api.generator.api.keras import optimizers
from tensorflow.tools.api.generator.api.keras.preprocessing.image import ImageDataGenerator

修改后再次运行,You can find that this error message will no longer appear.

适用于Linux用户的方法
同样的,still create firstpython文件keras_dir.py,运行keras_dir.pyThe output is displayed as follows：

[**@** EfficientNet]# python keras_dir.py
['/usr/local/python3/lib/python3.6/site-packages/tensorflow/_api/v1/keras']

可以看到keras的安装目录,We modify the program calls according to the directorykeras库的语句：

from tensorflow._api.v1.keras import models
from tensorflow._api.v1.keras import layers
from tensorflow._api.v1.keras import optimizers
from tensorflow._api.v1.keras.preprocessing.image import ImageDataGenerator

至此,We successfully solved itModuleNotFoundError: No module named 'tensorflow.keras’这个错误.