APAC-Net is an algorithm for solving high-dimensional, and optionally stochastic, Mean-Field Games.

The accompanying paper: https://arxiv.org/abs/2002.10113

If you found our paper or code helpful, please consider citing:

@article{lin2020apac,
  title={APAC-Net: Alternating the population and agent control via two neural networks to solve high-dimensional stochastic mean field games},
  author={Lin, Alex Tong and Fung, Samy Wu and Li, Wuchen and Nurbekyan, Levon and Osher, Stanley J},
  journal={arXiv preprint arXiv:2002.10113},
  year={2020}
}

In order to start training, do

python main_apac-net.py

Inside main_apac-net.py there are hyperparameter that one can choose for solving the environment. The environment can be choseb by giving the proper name to env_name. Current options are BottleneckCylinderEnv, TwoDiagCylinderEnv, and QuadcopteEnv.

Once training is finished, one can run tests of the trained model by

python start_test.py

and giving the correct path for the experiment_path argument.