Halite finals begin tonight (EST)!

#1

There are only a few hours left to submit your final Halite III bots.

Halite finals begin tonight, January 22, 11:59:59PM (“midnight”) EST, and run for just under one week, to January 29, 11:59:59AM (“noon”) EST. The games played during the finals period will determine the final rankings, and thus the winner, of the Halite III competition.

Bot submission is closed during the finals period. Shortly after finals begin, all matchmaking will be paused and all bot ranks reset.

Once bot ranks have been reset, finals matchmaking for all bots in the competition will begin. Over the course of the finals period, the lowest ranking bots will be systematically removed.

The final 150 remaining bots will play until the close of the finals period, to determine high-confidence final rankings.

Good luck!

#3

Thanks exciting! a few competitors had big jumps with last minute updates in the top ranks! will be interesting to see if anyone was holding back any novel attack strategies to unleash in the finals.

#4

Will the matchmaking system try to ensure that all bots play roughly the same number of games each day, to keep the sigma value consistent?

#5

35 minutes after midnight and no games have been played.

#7

I am at 2471 which is below idle bots.

#8

I had my first three games 15 minutes ago. Seems to me it’s just a matter of time until everyone gets to play.

Still, it’s surprising that match making is not more balanced with some bots having 30+ games and others have none yet. (Just a wild idea: could it be that GPU bots get more games because less bots compete for GPU slots?)

#10

So Rachol https://halite.io/user/?user_id=134 played 37 games whiles most everybody played nothing.

#11

looks like its playing the bots in reverse order of their last commit time, so very few of the top ten been played yet.

#12

i committed ten minutes before midnight and see my first game just played, teccles also been activitated.

#13

Everyone has played games by now. The matchmaking does prioritize players with fewer games, but there are a lot of players, and only so many games we can play at once, plus it can pull in players who have already played into a match.

#14

Just a thought.

It seemed during the preliminary rounds that the 2p vs. 4p split was roughly even, but now it seems to be about 2-to-1 in favor of 4p games. I can understand if this was deliberate, as it gets more bot comparisons in with less server load, but the 2-1 split favors bots that perform better in free for all versus head to head, so would have been nice to see this information officially published in advance.

#15

Yea I noticed that, and my bot is not so good with 4p games lol

#16

The 2-1 split between 4-player and 2-player games is not to get more bot comparisons. Halite 3 plays the same number of 4-player games as 2-player games. Since 4-player games involve twice as many players, each player has twice as many 4-player games.

Before the finals the split was roughly even for newly submitted bots and a 3-1 split for bots that played a lot of games.

#17

I consider this a minor flaw in the execution of the competition.

It’s not a big deal because it hardly makes any difference (certainly not for the top ranking bots anyway), I just mention this as a possible consideration for future competitions.

When selecting bots for the next game, your algorithm seems to select a competitor with a relatively high sigma, then randomly chooses to do a 2-player or 4-player game, and then selects opponents that have a similar score than the original bot. The problem with that is that some bots tend to have higher sigmas than other bots with the same number of games, presumably due to their style of playing. These bots end up having a higher share of 2-player games, while low-sigma bots get a disproportionately high share of 4-player games, as they are never selected as first competitor. After 500+ games per bot, there are high-sigma bots that have more 2-player games than 4-player games, and there are other bots that have a significantly higher than 3-1 share of 4-player games. A possible improvement could be to take the 2p/4p split of bots into account when selecting opponents, in order to make sure that all bots have a similar split rate at the end.