Project 4 Team 4

Project 4 for reinforcement learning

Environment information

The package gym-2048 does not work with newer versions of Python. For now I have installed python==3.7 in a new environment (I know version 3.10 does not work since it requires numpy 1.14)

Files

• main.py has an example from gym-2048 on how to run the environment
• dqn_model.py has a simple neural network to obtain the action. Since the state is a 4x4 matrix there is a flatten in the beginning and then the network will find the best action
• replay_memory.py has a replay memory buffer implementation with functions to add experience to the buffer and sample experiences with some batch size from the buffer
• agent_dqn.py has the training loop for the neural network (it is still in progress)
• new_action.py has the function to take an action from the neural network based on the observation (it is used in the main.py file)
• plots.ipynb Jupyter notebook for plotting the reward
• test.py - used just to add some quick funtion for tests
• reward.npy - File with the rewards saved during training
• play_game.ipynb - File with user input episode and runs the episode similar to main.py but from jupyter notebook
• game.npy - One user played game state transition saved

Important links

https://pypi.org/project/gym-2048/

https://github.com/voice32/2048_RL

https://github.com/SergioIommi/DQN-2048

https://github.com/navjindervirdee/2048-deep-reinforcement-learning

https://tjwei.github.io/2048-NN/ - This link also has a repository, however, it's a complex model

https://github.com/FelipeMarcelino/2048-Gym - Rendering the episodes