p-christ/deep-reinforcement-learning-algorithms-with-pytorch

This repository is a collection of Python implementations of deep reinforcement learning algorithms. Reinforcement learning is a branch of AI research where software agents learn to make decisions by trying actions and receiving rewards or penalties based on the results, similar to how a person learns a game by playing it repeatedly. The code here uses PyTorch, which is a popular framework for building and training AI models. The repository covers 18 different algorithms, ranging from foundational approaches like Deep Q-Learning to more advanced methods like Soft Actor-Critic and Proximal Policy Optimisation. Each algorithm is a different strategy for how an agent figures out the best action to take in a given situation. Some work well when actions are discrete (like choosing left or right), others when actions are continuous (like adjusting a value on a sliding scale). Alongside the algorithms, the repository includes several custom game environments used for testing. These include simple grid-based puzzles and simulated robotic tasks. The README also shows graphs comparing how well different algorithms perform on these environments, so you can see which approaches learn faster or reach higher scores. To use the code, you clone the repository, set up a Python environment, install the listed dependencies, and run one of the results scripts. The setup process is documented with step-by-step terminal commands. If you want to test an algorithm on a different game, you can point it to any compatible environment from the OpenAI Gym library, or build your own by following the provided examples. This project is primarily aimed at researchers and students learning about reinforcement learning, but anyone who wants to see working code for these algorithms alongside experimental results can use it as a reference or starting point.