This repository is a project I began after feeling more familiar with reinforcement learning. I recreated the game 2048 using pygame so that I could train an RL agent to attain as high of a score as possible. Throughout the process, I trained the agent for more than 30 hours in total, and went through several itterations which improved upon the shape of the input data, calculation of the reward function, and more. I took inspiration for how to improve the model from various research papers such as this one: https://arxiv.org/abs/2212.11087. In the end, the model consistently achieves a score of around 1024 which is much better than random actions and better than the score that I am capable of achieving myself. As an extention of this project, I plan to use the model to play the actual game on the internet by recieving captured screen data.