2048-ql

Implementation of a 2048 AI using Q-Learning and following an algorithm very similar to the one described in Playing Atari with Deep Reinforcement Learning. Currently, the AI plays better than random moves but can't make it to 1024 :(.

Getting Started

Set up virtualenv

(pip install virtualenv if you don't have it)

cd 2048-ql/
virtualenv venv
source venv/bin/activate

Installing python dependencies

(venv)$ pip install -r requirements.txt

Installing Chromedriver

You can download Chromedriver here: https://sites.google.com/a/chromium.org/chromedriver/ After unzipping the folder, there will be an executable chromedriver (OS X). Place chromedriver in venv/bin/. Chromedriver should be located in venv/bin/chromedriver

If you aren't using OS X, place the equivalent executable in venv/bin/ and change the CHROMEDRIVER_DIR variable in main.py to the appropriate directory.

Train the Model

Train the model using Selenium and Chromedriver:

(venv)$ python main.py

Train the model locally, using my Python implementation of 2048:

(venv)$ python main.py -t local

Credits

Credits to Mikhail Sannikov's 2048 bot for without-animation.js and inspiration to use Selenium.