Installation

First, you need build your conda env

conda create --name atari python==3.8
conda activate atari

Then, the necessary dependencies must be installed. You can see all installations in condalist.txt files

Pytorch installation. It can be different for your device according to CUDA version. Please check Pytorch Installation Page: https://pytorch.org/get-started/previous-versions/

pip install torch==1.11.0+cu102 torchvision==0.12.0+cu102 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu102

Gym installation

pip install gym[atari]
pip install gym[classic-control]
pip install autorom[accept-rom-license] // Bugfix for old gym enviroments

Using msgpack for saving weights

pip install msgpack==1.0.2

Install tensorboard to see graphs.

pip install tensorboard==1.15.0

Train and Run

For training run following code

python dqn.py // Start training process
python observe.py // Open gym env and algorithm plays the game