AI player for 2048 game using Machine Learning concepts
pip install -r requirements.txt
After training the agent for 100K episodes, the agent was able to choose the most optimal move everytime. I didn't try the same for 4x4 settings as it is going to take days to get the same output.
The Test still gets into infinite loop in some games. Meaning the agent needs still more training.
- Double DQN code reference: https://github.com/rlcode/reinforcement-learning
- Reinforcement Learning Basics: https://www.youtube.com/watch?v=2pWv7GOvuf0
- Original 2048 Game: https://github.com/gabrielecirulli/2048
- Valuable discussion: https://github.com/matthiasplappert/keras-rl/issues/38