Installation Guide¶

Prerequisites¶

Operation system version: Linux, macOS, Windows

Python version: 3.7-3.9

Note

If there is a GPU in your setting, you can refer to NVIDIA CUDA Toolkit Installation

After CUDA being installed, you will get a correct NVIDIA CUDA version of deep learning framework (e.g., Pytorch) automaticly when installing DI-engine.

If you want to install Pytorch manually, you can refer to PyTorch Installation.

If your OS is Windows, please do confirm that SWIG is installed and available through the OS environment variable PATH, you can refer to SWIG installation.

Stable Release Version¶

You can simply install stable release DI-engine with the following command:

# Current stable release of DI-engine
pip install DI-engine

Tip

If you need to upgrade pip, you can use the following commands:

# Windows
> python -m pip install --upgrade pip
# Linux
$ pip install --upgrade pip
# MacOS
$ pip install --upgrade pip

Tip

If you encounter timeout in downloading packages, you can try to request from other site.

pip install requests -i https://mirrors.aliyun.com/pypi/simple/ DI-engine

And if you prefer to use Anaconda or Miniconda, the following command is suggested:

conda install -c opendilab di-engine

Development Version¶

If you need to install latest DI-engine in development from the Github source codes:

git clone https://github.com/opendilab/DI-engine.git
cd DI-engine
pip install .

Tip

If you hope to install DI-engine into local user directories, you can do as the following:

pip install . --user

Be careful that if you are using virtual python environment created by softwares, such as virtualenv, then the option --user may not work. Please ignore this tip.

Special Version¶

If you want to enable special version of DI-engine and install the extra packages that are required, you can use the following command:

# install atari and box-2d related packages
pip install DI-engine[common_env]
# install unittest(pytest) related packages
pip install DI-engine[test]
# enable numba acceleration
pip install DI-engine[fast]
# install multi extra packages
pip install DI-engine[common_env,test,fast]

Tip

Some certain shells require quotation marks around package names as the following:

pip install 'DI-engine[common_env,test,fast]'

Note

The whole installation procedure often lasts about 30 seconds, which depends on the the size of packages as well as download speed. If some packages installation failed, you can refer to the file “setup.py” and install the specific package manually. If meet some tough installation questions, you can ask to DI-engine’s ISSUE for help.

Run in Docker¶

DI-engine docker images are available in DockerHub. You can use the following commands to pull the image:

# Download Stable release DI-engine Docker image
docker pull opendilab/ding:nightly
# Run Docker image
docker run -it opendilab/ding:nightly /bin/bash

Installation Check¶

After installation, you can use the following python codes to check if DI-engine is available and show the version of it:

import ding
print(ding.__version__)

You can also try the command line tool of DI-engine as the folloing:

ding -v