Combinatorial Auction Test Suite (CATS)

The Combinatorial Auction Test Suite is a generator of combinatorial auction instances for the testing of CA algorithms. It features five distributions of instances from realistic, economically motivated domains, as well as a collection of artificial distributions that have been used in the literature. Since the release of CATS 1.0 in 2000, it has become a standard benchmark for the evaluation and comparison of CA algorithms.

What's up with dummy goods? A number of people have emailed me over the years to ask how dummy goods work; this evidently needs to be cleared up. These are goods that aren't really sold in the auction, but that are added by CATS to enforce additional XOR constraints (specifically, that one bidder can't win more than one of his bids). Thus, when you ask CATS to generate an auction with 20 goods, you'll see higher-numbered goods in some of the bids; these are the dummy goods. The number of dummy goods are given in the header at the top of the file. You can learn more about them on page 99 of my thesis. Dummy goods can also be used to tell which agent each bid comes from. Every bid that includes the same dummy good comes from the same bidder. Every bid without any dummy goods is a unique bid from a different bidder. Thus, you can use CATS to model agents' valuations in a combinatorial auction, as well as for generating input for winner determination algorithms.

Current Version: 2.2. That version is only available from our new GitHub repository.

Major enhancements over version 1.0:



Old downloads: you probably shouldn't use these.

If you find any bugs or other problems with CATS, please report them on the GitHub page, or better yet, submit fixes there. At this point the code is no longer able to be supported by its authors.

Last updated May 24, 2024