- Created Tuesday, June 17th 2014 @ 20:39:20
I noticed that the rankings of the bots fluctuate pretty wildly. At the current time there are quite a few aggressive bots out there and the hand can be very decisive in that case.
I also noticed this short game: http://theaigames.com/competitions/heads-up-omaha/games/53a07b804b5ab20b4221f527
My bot puts the opponent all in while having a 97% chance of winning, but loses at the river. Now my bot is not very good, but I can't disagree with it here.
So that got me thinking of paired matches. After each match, you simply play the same match, but with the hands reversed. Only winning both matches would count as a win. This would greatly reduce the effect of the strength of the hands from the equation and might lead to a more stable ranking. Of course it would not be allowed for a bot to try and recognize a reversed match.
- Updated Wednesday, June 18th 2014 @ 23:31:09
This will make sense for matches later on, but currently it wouldn't make more sense than just running an ELO-ladder where we play semi-randomly (remember your suggestion would take 2x the amount of matches to get 1 result, which in this sense is equivalent to just running 2x the amount of matches as it is now, be it against different opponents).
Note that I also mentioned that in the game of Omaha we will in this competition we can only have a roughly 65% chance to win against a simple "AllRaise". So yea, the variance is killer, and indeed your suggestion would make sense for the matches later on.