Why this article
安装Pytorch等环境堪称中国深度学习初学者的噩梦，本文将简单阐述安装工具辅助DL环境的安装。如果手头没有合适算力的设备推荐使用云服务器，这里个人尝试过Colab，Kaggle，矩池云，其中矩池云是付费服务器，但是综合体验值得推荐且价格低廉。因为是Linux服务器如果使用Pytorch记得更改num_workers的数值匹配服务器CPU核心数，否则会非常漫长。使用矩池云的时候个人感觉存在CPU配置过低带不动GPU的情况，推荐2080Ti级别的服务器起步。
This article is written for those DP learners located in China. Because of some well-known reasons (Firewall), access to some of the foriegn websites is not available. And, unfortunately, the libraries that are required to be installed using pip can be really hard (just nightmares!!!). Even the mirroring of '清华', '豆瓣' are not satisfactory.
In this article, I will try to state the steps of setting up training environments on Windows 10. If you would like to explore, the environment on Linux can be set up in a similar way.
Steps to set up
Install Clash for Windows
You can obtain this tool by simply searching on the Internet. It is recommended that you install from the official Github link (I am not putting the link here). To enable your access to the foreign websites, you may have to find (or buy) a route (If you find this difficult, refer to the advertisement in Clash for Windows). Paste the URL of your route in the profiles, and then click Download. Test your Prxoy list in Proxies.
Set up Clash
To proxy your device, first click System Proxy to 'ON' state. Then click the 'Launch Helper' which is included in 'UWP Loopback'.
Click 'Exempt All' and then 'Save Changes'.
Install Python
In this step you have several choices to choose. Either you can install directly using Anaconda or Python.
Personally I prefer the official Python version and you are suppose to install a Python with version lower than 3.8 for compatibility. Remember to choose 'Add Python to Path' when you are installing it.
Install libraries
After the steps above are finished, you may install the needed libraries for Python. For example, pip install numpy will install numpy for your computer. And about whether to use Pytorch or Tensorflow, it is totally dependent on yourself.
Top comments (0)