2048 Reinforcement Learning

A multi-part project that implements reinforcement learning and Monte Carlo simulations to "teach" a computer to play the game 2048.

Part 1 - Game implementation

See my 1st blog post, which covers my implementation of the game using only numpy.

Relevant notebook, which allows you to play my implementation of 2048 with a rough UI:

+-- Interactive game demo.ipynb

This notebook can be loaded via binder:

Part 2 - Reinforcement learning

See my 2nd blog post, which covers my logical approach to reinforcement learning and details the parameters necessary for using the NeuralNetwork class in network.py.

Relevant notebook:

+-- Pytorch results.ipynb

Part 3 - Monte Carlo tree search

See my 3rd blog post, which covers my implementation of Monte Carlo tree search to play 2048.

Relevant notebook:

+-- Monte Carlo results.ipynb

Details

Software: This project is built in python, relying primarily on pytorch, numpy, and matplotlib.

Author: Paige McKenzie

Date completed: 4/29/2020