• Ubuntu
• Python 3
• Make virtual environment

python3 -m venv venv

1. Activate virtual environment

. venv/bin/activate

2. Install baselines dependencies

sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev

3. Install Tensorflow

• Choice A: For CPU (slower)

pip install tensorflow==1.14.0

• Choice B: For Nvidia Cuda GPU (faster)

pip install tensorflow-gpu==1.14.0

4. Install baselines (Current baselines in pip is outdated)

pip install git+https://github.com/openai/baselines.git

5. Install Other dependencies

pip install -r requirements.txt

6. Import sonic roms

python -m retro.import ./Roms

7. Install retro-contest

git clone --recursive https://github.com/openai/retro-contest.git
pip install -e "retro-contest/support[docker,rest]"

python train.py

Monitoring is done through tensorboard. To run TensorBoard run the command below, make sure venv is activated

tensorboard --logdir ./tb_log

This will run tensorboard, and tensorboard will search for Events from the specified directory

Replace saved_model_filename with actual name of the saved model zip file. E.g. sonic_stable_dqn.zip

python play.py saved_model_filename

python slow_play.py saved_model_filename

Models that has successfully completed a level are stored inside saved_models