Quick Notes on Poker Solver
Science published a review of poker solver algorithm.
Researchers focus on two-player limit Texas Hold’em. Their grand scheme is this:
- Transform a poker game to an abstract game that has a smaller state space.
- Apply an equilibrium solver algorithm on the abstract game.
- Map the equilibrium back to the original game.
There are lossy and lossless transformations. A finer-grained lossy transformation is usually but not always better than a coarse-grained lossy transformation.
There are two popular families of equilibrium solver:
- General purpose linear programming solver.
- Counterfactual regret (CFR) minimization.
I am not sure what kind of equilibrium they are looking for. Subgame-perfect equilibrium? Sequential equilibrium? Or just simple Nash equilibrium?