The agent Walverine that participated in the 2005 TAC Classic tournament is an incremental revision of the agent by the same name that played in previous years. The original Walverine from 2002 is thoroughly described in an article appearing in Decision Support Systems. The incremental changes for Walverine 2003 and 2004 are described elsewhere on this site. The process by which we decided how to vary Walverine for 2005 is described in our 2005 TADA paper. This process resulted in just a couple of specific changes relative to Walverine 2004:
  1. Flight purchasing. The flight pricing model employed in 2004 included two small bugs: one in incorporating price observations near the boundary points, and another where values were inappropriately rounded . We fixed these bugs. The rounding bug caused the threshold parameters t1 and t2 (see the Walverine 2004 blurb) to be treated as if they were offset 0.25 on average. For Walverine-05, we adjusted parameters t1 and t2 by .25, thus approximating the 2004 settings with the rounding change. The full set of flight thresholds for 2005 was: t1 = 0.50, t2 = 1.25, t3 = 3, and t4 = 200.
  2. Entertainment trading. Walverine-05 employed a new entertainment bidding module based on Whitebear-03 (source code provided courtesy of Ioannis Vetsikas). We implemented the whitebear policy as faithfully as possible, but had to adapt the procedure in several respects: changing timing values for the 9-minute (2004) game, using as "best packages" those produced by Walverine's solution to the acquisition optimization problem, limiting entertainment bids to offering to buy and sell a single unit, and otherwise conforming to the Walverine bidding cycle.

    Walverine Team, representing the University of Michigan AI Laboratory

    Specific Walverine 2005 changes implemented by Kevin M. Lochner and L. Julian Schvartzman. Empirical game-theoretic analysis for Walverine 2005 performed by Shih-Fen Cheng, Kevin M. Lochner, Daniel M. Reeves, Rahul Suri, and Michael P. Wellman.