Tensorflow2 installation quick pit avoidance summary

Preface

I thought I had found a safe TF2 The installation process （conda python/cudatoolkit/cudnn + pip tf2）, I didn't expect that after the successful installation import individual Adam All wrong -_-||. Therefore, this blog post records and summarizes the problems experienced .

Fast pit avoidance

Situation 1

The installation was completed successfully , but import Basic components （ Such as tensorflow.keras.optimizer.Adam） Times wrong

Solution

After installation , Check （conda list command ） Is there an independent Keras library ： If you have any , Make sure it's connected with tensorflow Versions, .

If it's not the same , Then uninstall and reinstall keras library ：

pip uninstall keras 

pip install kears==2.x.0  # And tensorflow Same version

Problem analysis

stay Tensorflow2 in keras It has been integrated in tensorflow in , But part of the source code is still from keras The method of directly calling the Library ：

from keras. A module  import A component

However , The fact proved that , Although independent keras The library is installed tensorflow Installed as a dependent Library in the process of , But the version does not match tensorflow Unified , It's all kinds of “ No, it shouldn't be ” Reasons for reporting errors .

More cases will be added after the treatment ...