Implementations of basic RL algorithms with minimal lines of codes! (PyTorch based)
-
Each algorithm is complete within a single file.
-
Every algorithm can be trained within 30 seconds, even without GPU.
-
Envs are fixed to "CartPole-v1". You can just focus on the implementations.
- REINFORCE (66 lines)
- TD Actor-Critic (97 lines)
- DQN (113 lines, including replay memory and target network)
- PPO (116 lines, including GAE)
- DDPG (149 lines, including OU noise and soft target update)
- A3C (116 lines)
- Any suggestion..?
- PyTorch
- OpenAI GYM
# Works only with Python 3.
# e.g.
python3 REINFORCE.py
python3 actor_critic.py
python3 dqn.py
python3 ppo.py
python3 ddpg.py
python3 a3c.py