Quote:
Originally Posted by FullyCompletely
Hey Unguarded,
Polaris 2008's pre-flop play wasn't influenced *at all* by the programmers. It treated every one of the 169 preflop sets of cards independently, and the algorithm we used, CFR, chose the action probabilities for each preflop decision. We didn't make any adjustments to the action probabilities after CFR had determined them.
In our more recent Hyperborean programs that we enter in the Annual Computer Poker Competition, all of the flop decisions (and now, also all of the turn decisions) are handled independently, with no buckets or other abstraction techniques. We still have to simplify the game a little on the river by grouping together similar sets of cards, but even then, there isn't a programmer that goes through and decides how to group the hands together manually. Instead, we describe some features of river situations and use an algorithm that groups together river situations with similar sets of features. We then use CFR to find a strategy for the game and trust the action probabilities that CFR finds - we don't try to change them afterwards.
Awesome, tyvm! For some reason I always thought a GTO 3betting strategy would include mixed strategies with most hands... was just surprised to see Polaris using a static strategy with most.
Hyperborean sounds awesome... wish it were easier for us to see it in action or to even play against it. I loved watching Sonia vs. Polaris back in the day.
Quote:
Originally Posted by SMorris
I think the Unguarded ranges at the start of thread are slightly too loose with unsuited holdings vs EP and too tight with suited holdings vs both EP and LP. But this wouldn't be a huge problem if I'm correct.
Regarding defending vs SB open, in 07 the best high stakes players in the world used their heads and concluded that any two would be profitable. Some were loud about it and the play trickled down to the excellent middle and small stakes players. Then in 08 and 09 these middle stakes and below players checked their databases and saw the bottom 10% of hands (approximately) were for some reason losing slightly greater than .5 big bets in that situation and it became standard to fold 10-15%. Some excellent high stakes players fold non-0% these days. Rake is non-negligible at high stakes and very significant below. But again, calling with these suspect hands BB vs SB isn't a big mistake and will even very very slightly increase the profitability of the rest of the hands.
I have changed my defense strategy a lot since my earlier post... much looser now
Ahh, makes sense. I hate the idea of giving my opponent an easy time by folding some hands bvb... they get a rake-free pot with no effort, bleh. And ya, increasing the profatability of other hands has to be taken into consideration. This is prolly a lot more dramatic for players who just call a lot of strong hands bvb.