A 2048 game api for training supervised learning (imitation learning) follow ExpectiMax agent and create your own CNN model

• game2048/: the main package.
  • game.py: the core 2048 Game class.
  • agents.py: the Agent class with instances.
  • displays.py: the Display class with instances, to show the Game state.
  • expectimax/: a powerful ExpectiMax agent by here.
• explore.ipynb: introduce how to use the Agent, Display and Game.
• static/: frontend assets (based on Vue.js) for web app.
• webapp.py: run the web app (backend) demo.
• evaluate.py: evaluate your self-defined agent. test time = 50
• online_train.py: get your own agent's weight.The structure of the model can be found
• CNN_new_141.zip: best model weight file trained by me

• code only tested on linux system (ubuntu 16.04)
• Python 3 (Anaconda 3.6.3 specifically) with numpy and flask
• Tensorflow,keras,numpy

python online_train.py

• you will get your own model taught by ExpectiMax agent

In file ./game2048/agent.py

class MyAgent(Agent):
    def __init__(self, game,display=None):
        super().__init__(game, display)
        self.model1= model_my
        # self.model2= model2
        # self.model3= model3

        # print("load_model",modelpath1)


    def step(self):
    ...
        
        return direction

cd game2048/expectimax
bash configure
make

python evaluate.py

you will get the average score of your agent (original test time is 50)

python webapp.py

The code is under Apache-2.0 License.

Please read here.