Note: I walked through the following blog almost completely. Since my computer configuration is similar to this blogger, it is convenient to refer to; below I only record some of the things I encountered when I walked down step by step according to the blog process Question, the specific process is very good in this blog, so I won’t record it.
(This blog post is so detailed!!!
https://blog.csdn.net/XunCiy/article/details/89016510
First install anaconda, configure the python environment, tensorflow, keras, etc., after installing pycharm, set the operating environment to the environment we configured in anaconda called tensorflow (or other environment you configured). In fact, various environments can be configured in pycharm. This article mainly talks about the anaconda configuration environment.
(Through the blogger's explanation, my computer should install these
(CUDA 8.0 requires cuDNN 6 or cuDNN 5.1, tensorflow_gpu-1.4.0 or less, python 3.5-3.6))
1. Uninstall the previous professional version of pycharm
(Reason: The professional version has expired and I want to install the community version)
method:
Find unistall*****.exe uninstaller
It was said on the Internet that it is recommended to find a housekeeper to clean up the registry, so I used 360 housekeeper to clean up the computer.
2. Uninstall anadona3-4.1.1
(Reason: There is a blog post saying that this will cause problems, so I need to download the old version of Anaconda. But the problem I encountered later showed that the default python 3.7 mentioned here is the root environment of Anaconda, which has nothing to do with the tensorflow environment we want to configure (You need to configure python separately.) On the contrary, if the anaconda version is too old, the anaconda 3-4.2 version posted in this way, python 3.5 in the tensorflow environment will encounter the problem of not providing services, causing the configuration to fail. So it is not recommended to download the old version of anaconda.)
Find unistall*****.exe uninstaller
It was said on the Internet that it is recommended to find a housekeeper to clean up the registry, so I used 360 housekeeper to clean up the computer.
3. Uninstall the IDLE python 3.7.3 version downloaded before
(Reason: This was installed when I took the "python" course before, and it has been left. I want to clean up the computer completely, or it feels particularly messy, I'm afraid the software is also messy)
I didn’t find the unistall*****.exe uninstaller, so I uninstalled it directly with my computer’s 360 software manager
It was said on the Internet that it is recommended to find a housekeeper to clean up the registry, so I used 360 housekeeper to clean up the computer.
4. Download Anaconda3-2020.02-Windows-x86_64.exe
Installation package address: https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/
The specific installation process refers to "The first day of the first week 01 "
Don't forget to configure the environment variable +conda upgrade --all upgrade
5. Configure Tensorflow environment in Anaconda
Refer to the blog at the beginning
Note: When conda create --name tensorflow python=3.5,
I entered it on the cmd command line at first, so it kept failing.
Pay attention to the following, you should add "conda" when exiting the current environment, but it should be fine:
6. Installation and uninstallation of CUDA and cuDNN
Refer to the blog at the beginning
(Reason: TensorFlow has a CPU version and a GPU version. The GPU version requires the support of CUDA and cuDNN, and the CPU version is not required (but the difference between the two operating speeds is huge, it is recommended to install the GPU version)
(CUDA 8.0 requires cuDNN 6 or cuDNN 5.1, tensorflow_gpu-1.4.0 or less, python 3.5-3.6)
download link:
CUDA download addresses
of various versions of the official website Various cuDNN download addresses of the official website
CUDA installation:
Check there. I followed this blog: https://blog.csdn.net/jer8888/article/details/100558964 . Why? Because according to the blog at the beginning, my download progress has not moved at 0%! ! !
I specifically did this according to the blog:
Choose custom installation options. The core component CUDA must be checked, and GeForce Experience is not checked. If the current version is lower than the latest version, you can also check the remaining two options and click Next to start the installation.
(The default path is as follows:)
//自定义路径
E:\Softwares\anaconda3\cuda\Samples
E:\Softwares\anaconda3\cuda\CUDAv8.0
After a night, it was finally finished.
(Environment variables have been added automatically)
//cmd里查看版本信息
//注意V是大写，我一开始输入小写，没得到与参考博客一样的输出，还以为我安装失败
//https://ask.csdn.net/questions/656060
nvcc -V(Correct output)
Then run bandwidthTest.exe and deviceQuery.exe, all show Result=PASS
(My computer is in the E:\Softwares\anaconda3\cuda\CUDAv8.0\extras\demo_suite directory)
cuDNN installation
cuDNN cannot be called an installation. You only need to unzip the downloaded compressed package and put the files in the corresponding folder into the corresponding folder under the CUDA installation path.
Questions about copying the past:
The only person I can understand the copying process
7. Install Tensorflow in Anaconda
pip install --ignore-installed --upgrade tensorflow-gpu==1.4.0(Something went wrong, it should be that the pip version of the virtual environment of tensorflow is too low and needs to be upgraded)
python -m pip install --upgrade pipFollow the instructions on the picture, and then you can’t upgrade, don’t give up, just input it a few times.
Then enter the tensorflow installation command again: pip install --ignore-installed --upgrade tensorflow-gpu==1.4.0
(As shown in the figure below, there is still an error, red one, there is no instruction prompt, I went to search, there are various methods on the Internet, it is said that the network problem, the solution is very simple, do not use the official source to use Ali mirroring. Here first Put a few, if it doesn’t work, I’ll use these commands to download and try.)
It really doesn't work, I started to use other methods online:
(1 failed)
pip install tensorflow-gpu==1.4.0 -i https://pypi.tuna.tsinghua.edu.cn/simple
(2 success)
//博客https://blog.csdn.net/qq_43211132/article/details/94426458
//豆瓣源 - 速度嗖嗖的
pip install tensorflow-gpu==1.4.0 -i https://pypi.douban.com/simple
(3 did not try and 2 succeeded, I did not try this)
pip install tensorflow-gpu==1.4.0 -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
（---end---)
The installation test is as follows:
8. Installation of Pycharm and application of tensorflow
Refer to the blog at the beginning
smoothly
9. Install keras, ipython, jupyter in Anaconda
1. (Aliyuan) Swooshing speed
pip install keras==2.1.2 -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
2.ipython：
conda install ipythonRun it (though I don't know what this means):
ipython kernelspec install-self --user
3.jupyter：
conda install jupyter
//（多次下载失败，请多多尝试）
Open jupyter notebook in your browser: jupyter notebook (although I don’t know what this means)