2048 should be a well-known game. If you've never played it, I suggest you to play several times and learn about its rules.

This is a project making an AI to play the 2048 game. To be honest, this is my first try in making AI for a game. (And it is the first practical project about deep learning that succeeds)

In fact, it is not hard at all to make the AI for this game. (Maybe that's why I can succeed 🙂 ) So I think this project may be suitable for beginners and that's why I share this project here.

All the runnable python scripts in directory ./py. And the other scripts are placed in the subdirectory of ./py. Such as:

• ./py/game the game logic and game process
• ./py/nerve the tools for neural network
• ./py/ai the tools for game AI

The directory ./graph stores well-trained neural network graphs of the cores.

The directory ./.idea and file ./AI2048.iml are "project files" of IDE Intellij IDEA. If you are using this IDE, you can import this project.

Here I will tell you how to run the scripts in this project and that way you can have a first look of what the AIs can do.

To run the scripts, firstly you need to build the environment. This project mainly needs Python3 with TensorFlow (including Numpy). The exact versions I used are placed here for reference:

• Python 3.5.2
• TensorFlow 1.4.0
• Numpy 1.13.3

Here I will not explain how to install Python3 or TensorFlow. If you need such guides, please refer to some other sites.

All the runnable scripts are placed in directory ./py. All these scripts are supposed to run under directory ./py (i.e. working directory is ./py).

Here are the introduction of these runnable scripts:

• ./py/game_console.py - launch a console-based 2048 game
• ./py/game_random.py - the game played by the "random AI". It takes actions completely randomly. (Well, maybe it is not really "intelligence")
• ./py/game_aix-corex.py - the game played by a an AI
• ./py/game_ai2-core3_extreme.py - this is the best AI in this project. It can play until appears "2048" almost every time. But it also takes a lot of time to "think".
• ./py/game_assist2_core3_extreme.py - scripts that can let the best AI assist you in playing 2048.
• ./py/train_corex.py - a set of scripts, which can train the neural network "core".
• ./py/match.py - a script that tests all the AIs and compares how powerful they are.

Here I place the expected average max rounds of the AIs in tests. This may help you to judge if the AIs are working normally.

• random - about 100 rounds
• ai1_core1 - about 170 rounds
• ai1_core2 - about 300 rounds
• ai2_core2 - about 700 rounds
• ai2_core3 - about 1000 rounds

Things about the working principles are written in an other .md file: PRINCIPLE.md. You can refer to it, if you want to learn about the working principle of the AIs.