Benefits of Permutation-Equivariance in Auction Mechanisms

---

This repo holds the code for the paper, Benefits of Permutation-Equivariance in Auction Mechanisms, which is accepted to NeurIPS 2022.

Getting Started

We mainly build our code based on the RegretNet. Thanks for their excellent work.

Before running our code, please install the following packages first:

• Python3
• Tensorflow2 (Note: It's OK to install Tensorflow v1, because we do not use any new feature for Tensorflow v2)
• Numpy
• Matplotlib
• Easydict
• Git

---

Training and Test

To train and test RegretNet-PE, please switch the git branch to master first:

git checkout master

For RegretNet-test, use:

git checkout rtest

Changing branch would modified the network structure and pipeline. Then, the experiments can be conducted with:

python run_train.py [setting_name]
python run_test.py [setting_name]

All the setting_name are listed as followed

• additive_1x2_uniform
• additive_2x1_uniform
• additive_2x2_uniform
• additive_2x5_uniform
• additive_3x1_uniform
• additive_5x3_uniform
• additive_2x1_normal
• additive_2x2_normal
• additive_3x1_normal
• additive_5x3_normal
• additive_3x1_nor51
• additive_5x1_1010

---