tinygym

A tiny RL playground with minimal, hackable implementations of common RL algorithms.

1. algos/: custom implementations of common RL algorthms
2. envs/: custom gym environments

python tinygym.py --algo [algo] --task [task] --max_evals [default=1000] --save [True]

Test on sample tasks: python unit_test.py

RL

• reinforce (~35 lines of code)
• vpg (~50 lines)
• cma
• ppo (based on SB3)
• dqn
• sac

Converges to basic controls tasks in <1K episodes (CMA takes longer, ~10K).