Testproblems from the report "Combinatorial Auctions: A Survey" (INFORMS J. Computing, (2003) Vol. 15, No 30) and the comments regarding the 700 MHz auction
by FCC titled "On the Computational Feasibility of Combinatorial Auctions on all Subsets" by Sven de Vries and Rakesh Vohra

Here (use ``Save Link As'') are the files for all the comparisons we did in "Combinatorial Auctions: A Survey". (To uncompress them on a unix box, use either ``gtar xvfz problems.tgz'' or ``mv problems.tgz problems.tar.gz; gzip -d problems.tar.gz; tar xvf problems.tar''.) The archive contains 2560 files that we used for benchmarks; together they take about 280MB (the tgz-file is just 47MB light:). Each file has the following structure:

• first line: number of items
• second line: number of sets
• each of the following lines contains:
  • bidamount for a set
  • number of items in this set
  • a (space separated) list of indices of items contained in this bidset

• prob.1.[Number of Items].[Number of Bids].[Number of this Instance]: SandholmRandom

• prob.2.[Number of Items].[Number of Bids].[Number of this Instance]: SandholmWeightedRandom

• prob.3.[Number of Items].[Number of Bids].[Size of Uniform Sets].[Number of this Instance]: SandholmUniform

• prob.4.[Number of Items].[Number of Bids].[Alpha in Percent].[Number of this Instance]: SandholmDecay

• prob.5.[Number of Items].[Number of Bids].[Size of Uniform Sets].[Number of Bidders].[Number of this Instance]: ourQuadraticModel

Here (use ``Save Link As'') are the files for all the comparisons we did in "On the Computational Feasibility of Combinatorial Auctions on all Subsets". (To uncompress them on a unix box, use either ``gtar xvfz almostadd12.tgz'' or ``mv problems.tgz almostadd12.tar.gz; gzip -d almostadd12.tar.gz; tar xvf almostadd12.tar''.) The archive contains 2550 files that we used for benchmarks; together they take about 245MB (the tgz-file is just 71MB light:).

The files are named by the following convention:

• prob.8.[Number of Items].[Number of Bids].[Lower Bound on Average Value].[Upper Bound on Average Value].[epsilon*1000].[Number of this Instance]: all provided instances have a lower bound of 90, an upper bound of 110, and epsilon*1000 in steps of 2 from 0 to 100. For an explanation of the problems see "On the Computational Feasibility of Combinatorial Auctions on all Subsets".

This page used to reside at: http://www-m9.ma.tum.de/~devries/comb_auction_supplement/ until about 2007.