A-Jacobson / rainbow

Rainbow: Combining Improvements in Deep Reinforcement Learning

Rainbow

Rainbow: Combining Improvements in Deep Reinforcement Learning

Setup

1. Follow setup instructions at https://github.com/openai/gym to install openai gym and atari games.

pip install gym
pip install -e '.[atari]'

2. install pytorch and torchvision:

conda install pytorch torchvision -c pytorch

3. install tensorflow and tensorboardX for logging.

pip install tensorboard
pip install tensorboardX

Contents

• DQN tutorial
• Vanilla DQN agent class

Usage

To train an agent on breakout for 2000000 steps.

python train.py --env Breakout-v0 --num_steps 2000000 --agent_name breakoutdqn

To watch your agent play 20 episodes of breakout.

python play.py --env Breakout-v0 --num_episodes 20 --agent_name breakoutdqn