README.md

AMenuNet

This is the code for the paper "A Scalable Neural Network for DSIC Affine Maximizer" in NeurIPS 2023.

@article{duan2023scalable,
  title={A Scalable Neural Network for DSIC Affine Maximizer Auction Design},
  author={Duan, Zhijian and Sun, Haoran and Chen, Yurong and Deng, Xiaotie},
  journal={arXiv preprint arXiv:2305.12162},
  year={2023}
}

Generate test data (optional)

Run gen_values.py to generate all the data for the final test. And then you can find the data in './data/'

Architecture and Mechanism

The architecture of AMenuNet is in net.py. The AMA mechanism is in auction.py

Training

To reproduce all of our experimental results, run x.sh. The results will be stored in './results/x/' For example,

experiments.py includes training, validdation and testing. But if you have already got the checkpoint and only want to test it, run test.py. This may happens when conducting out-of-setting experiments. One thing to care is that to make sure that menu size and $\tau_A$ are the same between the testing time and training time.

You can also adjust hyperparameters and different configurations in experiments.py and run it.