minimalRL

Why we included this project

minimalRL is a compact way to see how the core reinforcement learning algorithms work in actual PyTorch code. REINFORCE, DQN, PPO, DDPG, SAC, and the rest each live in a single file of about 100 to 150 lines, so you can follow a full training loop without hunting through a large codebase. Since every environment is pinned to CartPole-v1 and runs finish in around thirty seconds on a CPU, it is easy to experiment and to compare how value-based, policy-gradient, and actor-critic approaches differ. That makes it a good fit for students, educators, and engineers refreshing their RL knowledge who want to get from code to the underlying math quickly. Just remember it is built for clarity, not production scale, so treat it as a learning companion rather than a drop-in library.