2048-api

Code structure

• game2048/: the main package.

  • game.py: the core 2048 Game class.
  • agents.py: the Agent class with instances, is a strong agent based on rule.
  • displays.py: the Display class with instances, to show the Game state.
  • expectimax/: a powerful ExpectiMax agent by here

  My Design

  • 'model1/': training code for model1 (online)
    • 'dataloader.py': dataloader for my model1
    • 'models.py: my Neural Network Model based on Vgg
    • 'vgg.py: vgg model
    • 'utils.py': a tool for batchsize
    • 'train.py: my code of online imitation training
  • 'model2/': training code for model2
    • 'model.py': my Neural Network based on CNN
    • 'train.py': my code of CNN training
  • 'model3/': training code for model3 & model4
    • 'CNN_4096.py': include Network Model and training code for model3 & model4
  • 'model5/': training code for model5
    • 'CNN_deep.py': include Network Model and training code for model5

  agents

  • 'agent_voting.py': my design agent, load five models and voting to get direction

  params

  • 'model1_pretrain_params.pkl': pre-train model for model1
  • 'model2_pretrain_params.pkl': pre-train model for model2
  • 'model3_pretrain_params.pkl': pre-train model for model3
  • 'model3_pretrain_params_2.pkl': pre-train model for model4
  • 'model5_pretrain_params.pkl': pre-train model for model5
  • For privacy and github limitation, model5 params haven't been uploaded.

• 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.

• evaluate_voting.py: evaluate my voting agent.

Requirements

• code only tested on linux system (ubuntu 16.04)
• Python 3 (Anaconda 3.6.3 specifically) with numpy and flask
• Pytorch 0.4.0
• numpy 1.14.5

To compile the pre-defined ExpectiMax agent

cd game2048/expectimax
bash configure
make

To run the web app

python webapp.py

LICENSE

The code is under Apache-2.0 License.

For EE369 students from SJTU only

Please read here.