TAC-01 Competition

Of the 19 teams that entered the qualifying round, ATTac-2001 was one of eight agents to make it to the finals on the afternoon of October 14th, 2001. The finals consisted of 24 games among the same eight agents. Right from the beginning, it became clear that livingagents [Fritschi DorerFritschi Dorer2002] was the team to beat in the finals. They jumped to an early lead in the first two games, and by eight games into the round, they were more than 135 points per game ahead of their closest competitor []<SouthamptonTAC,>Jennings02. 16 games into the round, they were more than 250 points ahead of their two closest competitors (ATTac-2001 and whitebear).

From that point, ATTac-2001, which was continually retraining its price predictors based on recent games, began making a comeback. By the time the last game was to be played, it was only an average of 22 points per game behind livingagents. It thus needed to beat livingagents by 514 points in the final game to overtake it, well within the margins observed in individual game instances. As the game completed, ATTac-2001's score of 3979 was one of the first scores to be posted by the server. The other agents' scores were reported one by one, until only the livingagents score was left. After agonizing seconds (at least for us), the TAC server posted a final game score of 4626, resulting in a win for livingagents.

After the competition, the TAC team at the University of Michigan conducted a regression analysis of the effects of the client profiles on agent scores. Using data from the seeding rounds, it was determined that agents did better when their clients had:

1. fewer total preferred travel days;
2. higher total entertainment values; and
3. a higher ratio of outer days (1 and 4) to inner (2 and 3) in preferred trip intervals.

Based on these significant measures, the games in the finals could be handicapped based on each agents' aggregate client profiles. Doing so indicated that livingagents' clients were much easier to satisfy than those of ATTac-2001, giving ATTac-2001 the highest handicapped score. The final scores, as well as the handicapped scores, are shown in Table 10. Complete results and affiliations are available from http://tac.eecs.umich.edu.

Table 10: Scores during the finals. Each agent played 24 games. Southampton's score was adversely affected by a game in which their agent crashed after buying many flights but no hotels, leading to a loss of over 3000 points. Discarding that game results in an average score of 3531.

Agent	Mean	Handicapped score
ATTac-2001	3622	4154
livingagents	3670	4094
whitebear	3513	3931
Urlaub01	3421	3909
Retsina	3352	3812
CaiserSose	3074	3766
SouthamptonTAC	3253	3679
TacsMan	2859	3338