DRL-Pytorch

Why we included this project

For someone learning deep reinforcement learning, this repository collects the core algorithms into one place with consistent PyTorch code. You get value-based methods like Q-learning, Dueling DDQN, prioritized experience replay, C51 and NoisyNet, alongside policy-gradient approaches like PPO, DDPG, TD3 and SAC, including discrete and continuous action versions where they differ. Each algorithm sits in its own folder and runs with a single main.py, so you can trace how a specific trick such as distributional Q-learning or twin critics changes the training loop. The code links back to the original papers and common learning resources, which makes it useful both as a reference and as a study path when moving from theory to runnable agents. If you are prototyping on gymnasium environments, it is a convenient self-contained base for comparing how these methods behave.