ICFP Programming Contest 2002

$Revision: 1.19 $, $Date: 2002/10/25 22:49:37 $

The scoring process for the normal division

How did we pick a winner among the ~168 submitted entries, you wonder? How did we decide which robot was the best? Below, we describe our judging procedure.

A robot is considered good if exhibits some degree of "intelligent" behavior, i.e. it performs well in a wide variety of scenarios. The robots were thus tested in a number of single-submission scenarios (a submitted player on its own, or possibly meeting one of our instrumented players) and a number multi-player scenarios, and the best robot was the robot with the highest final score, obtained by taking a weighted sum the scores from the different scenarios.

There are at least two sources of randomness that introduce "noise" in the scores, and we have been careful to limit the effect of this noise:

We started by running all robots in a variety of single-submission scenarios (round 0). We then ran all robots in random groups of 8 in the symmetry scenario (round 1).

Time would not allow extensive testing of all robots, so after round 1, we picked the 20 best robots so far, confident that the best two would be among these. We made a number of runs in two additional multi-player scenarios, were all 20 robots participated in each run, and summed up the scores so far (round 2).

After round 2, we limit further testing to the 8 best robots so far, again under the assumption that the two best would be found among these. The 8 robots all participated in a number of runs of two multi-player scenarios (round 3).

The top two robots after round 3 were then run in a number of two-player games (round 4). The total score after this determined the winners.

Tables of results for the normal division

Tables of results for the lightning division

The lightning division was scored in much the same way as the normal division. Here are the results.

Submissions and log files

Finally, here are links to the submitted programs and the raw log files from all the games we ran to obtain the above results. Also, the tables contain links to relevant log files and scenario descriptions.

The log file format is described in the README file for the Game Simulator.


2002-10-25 15:45 PDT
We reran submission 243, which got around 200 points.
2002-10-10 12:00 PDT
It was discovered that the log file for submission 130 in the sokoban scenario was incorrect. The test was rerun and the log file was replaced. As a result, submission 130 lost 31 points. (Thanks to Thomas G. Rokicki for pointing this out.)
2002-10-15 16:15 PDT
For no good reason, scores for submission 257 were missing for the bridge, maze and sokoban scenarios. These were added, and submission 257 gained 33 points.

ICFP Programming Contest 2002