The following packages need to be installed in advance anaconda
pytorch2onnx actual combat

pytorch install

pytorch brief introduction ： Pytorch It's an open source Python Machine learning library , from Facebook AI Institute maintenance , It's one of the mainstream deep learning frameworks . Basic learning can be referred to Pytorch Chinese document & course .
pytorch install ： It is suggested that pytorch Official website The installation order of （ firsthand experience ）. Check it according to your own configuration , It is recommended to use pip install , And then in anaconda prompt Enter the command line given on the official website ：

tensorflow install

tensorflow brief introduction ： tensorflow Is a data flow graph , Open source software library for numerical calculation , It is also one of the mainstream deep learning frameworks , from google Team maintenance .tensor Indicates that the transmitted data is tensor ( Multidimensional arrays ),flow Said to Calculation chart Carry out operations , That is, the calculation task is represented by a graph , also The diagram must be executed in the session .

tensorflow install ： It is suggested that anaconda The new one in is called “tensorflow” Of A virtual environment , The environment comes with the ability to run python The basic package of , The reason for this is to operate in the newly created environment Will not affect the external environment , What do you want to do , Of course, install directly in the external environment tensorflow It's also OK Of .
New environment ：

conda create --name tensorflow python=3.6

designated python Version is in the new environment python edition , Nothing to do with the external environment . If you want to install 3.7 edition , yes , we have conda Version may not support , You can check the installable python edition ：

conda search  --full --name python

Activate the environment ：

activate tensorflow

Appears before the command line “(tensorflow)” It means that the current is tensorflow In the environment .
If the work is completed, you can exit the environment ：

deactivate

The above is New environment The process , We just named this environment tensorflow, It's not installed tensorflow, Next, we show how to install in the new environment tensorflow：

pip install tensorflow

perhaps ：

pip install --upgrade --ignore-installed tensorflow

Waiting for the installation , After installation, you can enter pip list see tensorflow edition . And then verify tensorflow Is the installation successful , Enter python, Will jump out python Version information , And enter python edit , Enter the following command ：

import tensorflow as tf
hello = tf.constant('hello!')
sess = tf.Session()
print(sess.run(hello))

Output b'hello! That means the installation is successful ！
Reference blog ： Hand in hand teaches you how to install Tensorflow(Windows and Linux Two versions )

onnx and onnx_tf install

onnx brief introduction ： onnx(Open Neural Network Exchange) It is an open neural network exchange format , It is a general deep learning framework , It is through onnx take pytorch Model to tensorflow Model .
onnx install ：

pip install onnx

onnx_tf install ：

pip install onnx_tf

Or go https://github.com/onnx/onnx-tensorflow download , Enter after completion onnx-tensorflow Enter the command line under the directory :pip install -e

Possible problems and solutions

SSL Module not available

Problem description ：“pip is configured with location that require TLS/SSL, however the ssl module in Python is not available”

Go to the website ：https://slproweb.com/products/Win32OpenSSL.html

Download the first one （3M）.exe Just install the file .

Cannot open …pip-script.py

Problem description ：“Cannot open D:\Program Files\Anaconda3\envs\tensorflow\Scripts\pip-script.py”
This is pip The problem of , Need to be reinstalled pip, Enter the following command line ：

python -m ensurepip
easy_install pip

ReadTimeoutEorror

Problem description ： Problems may occur during installation "ReadTimeoutEorror" Or something like that , Because pip The default resource pool is foreign , Without scientific Internet access, the download speed is very slow , If the default time is exceeded, an error will be reported , There are two solutions to this kind of problem ：
If the installed package is small ： You can set the default timeout , The default is 15, We can set it to 100, For example, we update pip To the latest version ( Only 1.4M)：

python -m pip install --default-timeout=100 --upgrade pip

If the installed package is large ： You can replace the resource library with a domestic image source ：

pip install  Package name  -i  Mirror source 

Domestic image sources mainly include ：
https://pypi.tuna.tsinghua.edu.cn/simple( tsinghua )
https://pypi.douban.com/simple( douban )
http://mirrors.aliyun.com/pypi/simple/( Alibaba cloud )
For example, we install tensorflow( A few hundred M)：

pip install tensorflow -i https://pypi.tuna.tsinghua.edu.cn/simple

If you don't want to add one every time you install -i Mirror source , You can click on the Pycharm Medium File—settings—project interpreter—“+” Number —manage repositories, Change the default source to Tsinghua source or other sources .
Here are more detailed usage methods of domestic image source ：https://www.cnblogs.com/microman/p/6107879.html

import tensorflow Report errors "ImportError"

Problem description ：import tensorflow It's a mistake ："ImportError: DLL load failed: The specified module was not found ." Generally installed tensorflow The version is too high or too low , Reinstallation of version succeeded import.
Install the specified version ：

pip install tensorflow==1.15.0 -i https://pypi.tuna.tsinghua.edu.cn/simple

conda Commonly used instructions

Finally, record a few conda Commonly used instructions ：
Install the specified version of the package ：conda install package=version
Specify the source of the package ：conda install --channel Mirror source
Restore default source ：conda config --remove-key channels
Check which environment variables are currently installed ：conda info --envs
Check what is currently available python edition ：conda search --full --name python
Install new environment ：conda create --name Environment name

If you have any questions, please leave a message in the comments section , At the same time, my level is limited , I hope you can correct some mistakes .