Game 2048

Python implementation of the game 2048.

Visual Mode

Game Strategies

There are 5 game strategies implemented:

• Random: Makes a random next move.
• Left Down: Try Left or Down, else Right or Up.
• Simple Max: Checks all possibilities of the next move and takes the move that maximizes the average.
• Two Max: Checks all possibilites of the next two moves and takes the move that maximizes the average value.
• Deep Q-Learning: Uses a Neural Network to approximate the most rewarding moves. Needs to be trained.

Deep Q-Learning

Replay of a Game played by the Deep Q-Lerning agent with 23k training steps, with batch size 512.

How to Run

This script requires the following Python packages:

• numpy
• matplotlib
• tqdm

For QLearning additionally:

• torch

Then run python main.py and follow the instructions and mode selection in the command prompt