The repository contains a Python implementation of Differential Backward Induction for the structured hierarchical games (SHGs). Each agent is assumed to have a one-dim strategy.
tqdm, munch, networkx, pytorch, hessian
Run this command in the SHG_DBI
folder.
python -m expPGG -mode <mode_name> -cfg_name <cfg_name>
TO conduct experiments of security game, you can use expSec
instead.
Example:
- Run DBI:
python -m exp.expPGG -mode PGD -cfg_name pgg_cfg_test
2. Run BRD:
python -m exp.expPGG -mode BRD -cfg_name pgg_cfg_test
- Using BRD to evaluate DBI
python -m exp.expPGG -mode Eva_PGD -cfg_name pgg_cfg_test
- Using BRD to evaluate BRD
python -m exp.expPGG -mode Eva_BRD -cfg_name pgg_cfg_test
Configurations
- The configurations are written using a simple tool from mmcv. [doc]
- Configurations are stored in
src/exp_configs/secg
andsrc/exp_configs/npgg
. The configs used in the paper are provided. You can use your own configs following those examples.
Results
- Results are stored in
res/pgg_results
andres/sec_results
as pickles.
If you used the DBI in your work, please cite us using the following BibTeX entry:
@inproceedings{li2022solving, title={Solving structured hierarchical games using differential backward induction}, author={Li, Zun and Jia, Feiran and Mate, Aditya and Jabbari, Shahin and Chakraborty, Mithun and Tambe, Milind and Vorobeychik, Yevgeniy}, booktitle={Uncertainty in Artificial Intelligence}, pages={1107--1117}, year={2022}, organization={PMLR} }