Installation 🛠
Quickstart
Open the project in Gitpod and start coding immediately.
Docker
# 0. Prerequisites: Docker
# 1. Run DQN algorithm
docker run --rm sdpkjc/abcdrl python abcdrl/dqn_torch.py
# 0. Prerequisites: Docker & Nvidia Drive & NVIDIA Container Toolkit
# 1. Run DQN algorithm
docker run --rm --gpus all sdpkjc/abcdrl python abcdrl/dqn_torch.py
Note
Docker Container parameters and the detailed installation process of the NVIDIA Container Toolkit can be found here: Nvidia Docker.
Warning
Using our docker image and train on GPU. Nvidia Driver needs to support CUDA11.7.
Using nvidia-smi command, look at the CUDA Version: xx.x in the top right corner. It need to be 11.7 or greater.
If your device only supports CUDA11.3-11.6, you can use sdpkjc/abcdrl:cu113 image. For older CUDA versions, we don't officially support them, please refer to our Dockerfile to adjust and build your image.
Pip
# 0. Prerequisites: Python3.8+ & Pip
# 1. Pull git repository from github
git clone https://github.com/sdpkjc/abcdrl.git && cd abcdrl
# 2. Install dependencies
pip install -r requirements/requirements.txt
# 3. Run DQN algorithm
python abcdrl/dqn_torch.py
# 0. Prerequisites: Conda & Nvidia Driver
# 1. Pull git repository from github
git clone https://github.com/sdpkjc/abcdrl.git && cd abcdrl
# 2. Create and activate virtual environment
conda create -n abcdrl python=3.9 pip && conda activate abcdrl
# 3. Install cudatoolkit and the corresponding version of Pytorch
conda install pytorch torchvision torchaudio cudatoolkit=11.6 -c pytorch -c conda-forge
# 4. Install dependencies
pip install -r requirements/requirements-torch.txt
# 5. Run DQN algorithm
python abcdrl/dqn_torch.py
Note
There are many ways to install pytorch, refer to Mu Li's video tutorials for details.
Version selection of cudatoolkit is related to Nvidia Driver version, refer to Mu Li's video tutorial and Pytorch installation page.
# 0. Prerequisites: Conda & Nvidia Driver
# 1. Pull git repository from github
git clone https://github.com/sdpkjc/abcdrl.git && cd abcdrl
# 2. Create and activate virtual environment
conda create -n abcdrl python=3.9 pip && conda activate abcdrl
# 3. Install cudatoolkit
conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_PREFIX/lib/
# 4. Install dependencies
pip install -r requirements/requirements-tf.txt
# 5. Run DQN algorithm
python abcdrl/dqn_tf.py