There has been some concerns that the number of matches played in the Arena was too small. Some players feared that they could not reach their real rank after pushing, especially for high-ranked AIs (< rank 300). The real ranking would occur over time from the matches generated by other players when they themselves push into the Arena but not until several hours have passed.
We heard you all right and we have acted accordingly.
Every 15 minutes since yesterday every player in the top 300 plays an additional match. This keeps the Arena live with the purpose of having the AIs more properly and fairly ranked.
At the end of the contest (Sunday, October 4 at 18:00 UTC, 2pm New York time, 20:00 Paris time), we will wait for all the current matches to play. Then we will relaunch 100 matches for the top 300 players. Once all the matches from this second wave are completed, the leaderboard will be final.
So do not hesitate to push close to the end of the contest!
Thank you for letting us know the process and thank you for running more games. However, I still have some concerns. I have been watching some players in the top 100 and their rank can vary from 40 to 80 within the same day (without them resubmitting). This tells me that rankings have a huge variance (even for bots with many games) and running an extra 100 matches may not be enough.
Suppose the startingposition of a match is asymetric and favors one player. Does that matchup get repeated, so each player gets the favorable position once?
Some 1v1v1v1 (4 players) have some permutations, but I didn’t see any permutations for 1v1v1 (3 players) while the positions can’t be symmetric (unless samecell=true) (or the replay is not available?). It’s not fair
Being equidistant from each opponents does not mean having the same available space. The map is not a circle.
It can explained why the rank is random.
It would be nice to have every permutations for each last extra game (<300)
As far as I remember, in “The Great Escape” game 4-player games have been cancelled completely as it takes too long to play all permutations.
Do you provide the guarantee that all permutations for 2-, 3- and 4-player games will be held?
Thanks.
For a 2 player match: either we launch a symmetric match (and only one match is played) or an non-symmetric match (and 2 permutation matches are played).
For a 3 player match: 6 permutation matches are played (unless all players start from the same cell)
For a 4 player match: either we launch a symmetric match (and only one match is played is played) or a non-symmetric match (and 24 permutation matches are played).
Number of players is random, so on average it should be 33 / 100 of each type.
Why is the ranking algorithm matching AIs against opponents which have just been pushed too (with low progress status)? These would be the least reliable indicators of the level of the ranked AI, no?
Or is the progress of each AI taken into account when computing the score?
My code is the same as yesterday. (I’ve uploaded an old code, not a new). But I think the problem occurred only in the first match, because of the stress in the servers.
What happens when somebody from the top 300 fall below position 300 after the final relaunch? Does that mean that the player which was originally “below the line” will be ranked in top 300 without rematch?
Yep, I had a few random timeouts as well. Replaying the game, everything was fine.
I was also surprised by my final ranking, as after the computations were done I was in the top 20. Then the final 100 games were launched and I instantly dropped below 40. I didn’t expect such a change as hundreds of games has already been launched when I was in the top 20…
I wonder if it would not have been easier to have a deterministic start position, like “everybody in the center” for all games. The ranking would have been much more deterministic.
Anyway, it was a lot of fun, thanks for it.
