Jaehyun-Jeong/100LinesRL
Clean RL algorithm implementations in under 100 lines each.
What's novel
Clean RL algorithm implementations in under 100 lines each.
Code Analysis
8 files read · 3 roundsMinimal PyTorch implementations of four core RL algorithms (PPO, SAC, TD3, DQN) in ~100 lines each, supporting both discrete and continuous action spaces across Gymnasium, MuJoCo, and DMC environments.
Strengths
The PPO, SAC, and DQN implementations are algorithmically correct and well-structured, with proper handling of GAE, twin Q-networks, automatic temperature, and dual discrete/continuous action spaces. The per-environment hyperparameter configs with target scores provide a useful reference for practitioners.
Weaknesses
No test suite exists, TD3 is hardcoded to Pendulum-v1 with global state and no config file, and there is significant code duplication (Buffer, make_envs) across algorithms with no shared utilities or package structure.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
56
Contributors
1
Files
11
Active weeks
14
Repository
Language
Python
Stars
1
Forks
0
License
MIT