Questions about GTO poker strategy
Posted by akissv7
Posted by
akissv7
posted in
Low Stakes
Questions about GTO poker strategy
Is there a Game Theory Optimal solution?
This would mean that there are strategies (if the game is played with a fixed set of possibilities such as fold, 1/3 pot bet, 1/2 pot bet, pot bet, 2 pot bet allin) where both player A and player B know each other strategies and they can play that strategy perfectly meaning that none of them can deploy another strategy that would make more profit against the other.
Do we know that this is solved and what the theory for the solution is or do we assume that the computer just calculates all possibilities and then finds the perfect strategy. Or do the current programs use some theory as to come to a local optimum strategy. Or do we use an AI constructed program to play against itself to come to a best strategy (meaning that we actually don't know if it is the Game Theory Optimal solution) just a damn good 'AI' solution much like the best chess bot which can beat each human player.
Loading 2 Comments...
Hey,
The only poker game that is solved is heads up limit hold'em. You can read about it and even play the bot here: http://poker.srv.ualberta.ca/.
This took many years to achieve and a very powerful computer, for a game which is infinitely simpler than NLH, especially when you start talking about multiple players being involved. For that reason NLH isn't getting solved anytime soon.
The way current solvers operate is they are given parameters of the game and they play our every scenario of every hand to see which solution gives them the highest EV for every hand individually as well as strategy as a whole. So to put that in other words, the solver will try to fold AA preflop, then it will try to raise it and play out every bet size on every possible board against an optimal defending range and come to a decision which of those decisions produced the highest EV, then it will tell you it's X amount better to raise the AA than fold.
Solvers today are clearly the most advanced we've ever seen, but they don't produce an actual GTO solution to the game, but rather an approximation of it. You would need an infinitely powerful computer to be able to solve a single situation in holdem. A program like PIO will ask you to limit it's option to a bunch of bet and raise sizes, then it will tell you how much of an error it made in it's approximation. Obviously neither of those could exist in an optimal solution.
Cheers.
Thx for the clear explanation.
So they are using AI just as google uses for lots of their algorithm and as they did for e.g. the chess & go game. They tryout a lot of plays by adjusting parameters based on the outcome of the game. E.g. when AA get cracked they will adjust parameters such that e.g. they fold AA a bit more on the flop or even preflop lol. But as most of the time AA will win they will keep playing it preflop.
If they are playing every scenario out that means that the we set limits to the moves that can be made in the form of bet sizes and sometimes even starting hands and number of players to simulate. So the end-result will depend a lot on the operators meaning rubbish in will mean rubbish out.
But if they play out every scenario there is no reason to set some standard dev as if one plays out every scenario one will find the best scenario (that is if poker is actually a game that behaves like that, also not all games have a Nash equilibrium and as poker is a game with discrete steps it could be the case that their is no Nash equilibrium for NLHE).
Be the first to add a comment