AuctionGym: Simulating Online Advertising Auctions

AuctionGym is an innovative simulation environment designed to facilitate reproducible offline evaluation of bandit and reinforcement learning approaches in the realm of ad allocation and bidding for online advertising auctions. With its inception tied to groundbreaking research presented at the 2023 ACM SIGKDD Conference, AuctionGym aims to bridge the gap in available datasets and methodologies within this field.

Why AuctionGym?

Offline evaluation of learning-to-bid methods can be quite challenging. The issues stem from:

• Unobserved confounding in observational data paired with the high costs of obtaining experimental data.
• The phenomenon known as Goodhart’s Law: when a measure becomes a target, it ceases to be a good measure.
• The absence of publicly available datasets that researchers can leverage for testing and validation.

AuctionGym aims to resolve these hurdles by offering a unified framework for practitioners and researchers to validate and benchmark novel ad bidding methods effectively.

Getting Started with AuctionGym

To familiarize yourself with AuctionGym, follow these initial steps:

1. Run the command: jupyter notebook in the main directory.
2. Navigate to the src folder, where you can find two introductory notebooks:
• Getting Started with AuctionGym (1. Effects of Competition): This notebook simulates second-price auctions, allowing you to visualize how varying competition levels impact advertiser welfare, surplus, and auctioneer revenue.
• Getting Started with AuctionGym (2. Effects of Bid Shading): Here, you’ll explore first-price auctions with bidders engaging in truthful bidding versus shading their bids based on perceived value.

Reproducing Research Results

To reproduce the results from our research paper, follow these steps:

1. Run the following command in your terminal:

python src/main.py configSP_Oracle.json

2. This will generate a results directory containing subdirectories for each configuration file executed.
3. These subdirectories will include .csv files with raw metrics and .pdf files with visualizations.

For additional details on configuration files, refer to the configuration documentation.

Understanding AuctionGym’s Code

Think of AuctionGym like a highly sophisticated game of chess. Each piece on the board represents a different factor in the auction environment, such as bidders and their bidding strategies. The code orchestrates these pieces to simulate various scenarios within the auction. Just as a chess player must anticipate their opponent’s next moves and strategize accordingly, AuctionGym allows users to evaluate and compare the effectiveness of different bidding strategies, adapting to the complexities of real-world auctions.

Troubleshooting

Should you encounter any issues while using AuctionGym, here are some troubleshooting suggestions:

• Ensure you have all the prerequisites installed. Missing dependencies can lead to unexpected errors.
• If the command doesn’t execute correctly, double-check the syntax and paths to ensure there are no typographical errors.
• For detailed assistance, refer to the Contributing Guidelines.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Conclusion

In conclusion, AuctionGym stands as a pivotal tool in furthering the research and application of machine learning techniques in online advertising auctions. It provides a structured, reproducible, and educative platform for understanding and applying bidding strategies effectively.

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.