Resquest: Let the solver play against the solver to get GTO frequencies for mass data analysis.

[gui999]

From the other issue, it seems that you are working on saving strategies, aggregate reports for flops and play vs GTO solution on "random" flops. Considering those, it would be great if there was an option to let a GTO opponent play against another GTO opponent to either create stats report or better, directly create "gto" hand histories.

Something that is extremely hard to come by are "global GTO frequencies". It's something that is kept in closed circles and/or are extremely costly $$. Those are used to compare against different pools of players in data analysis to find exploits.

For example, considering a single raised pot BB vs BU. flop goes X/X, on the turn the GTO frequency for BB to probe/bet is ~30%, while a pool of NL50 players are ~40%.

The question is, how do I know the GTO probe frequency is 30% to start with, or any other turn/river frequency? After asking around and does some research, it seems that people manage to create hand histories based on the strategy found by solvers.

Basically, if you've got a subset of solved flops you can take those strategy to make 2 gto players play against each other and save the result as a standard hand history from any site, and once you've got enough hands you simply import them in any hud software like PT4 or Hand2Note (for the range research). Obviously, 2 gto players would give gto hand histories.

In the case of your project, considering that you're already planning a "play against GTO" feature, I don't think it would be that difficult to implement. Play vs GTO would already use the strategy to simulate a gto villain, while showing the gto options to hero and wait for him to click a button.

I might even tinker with this myself if/when the play vs gto feature comes out. Can't wait!

Hopefully that made sense... thanks again for your work on this!

[Thomas-MMJ]

The question is, how do I know the GTO probe frequency is 30% to start with, or any other turn/river frequency? After asking around and does some research, it seems that people manage to create hand histories based on the strategy found by solvers.

The preflop ranges are solved (such as with MonkerSolver), using a (representative) subset of flops, then using those preflop ranges, you input the sizing options for Flop, Turn and River and a specific flop you want to solve. Then from that the frequencies of the various actions are solved for that specific flop. The 30% probe frequency is based on the assumed preflop ranges and the assumed allowed actions. That probe frequency is likely not the same probe frequency you'd get for a slightly different set of assumptions. These solutions are 'approximately GTO' not 'True GTO'.

The Solver already effectively 'plays against the solver' (that is a crude way to describe counter factual regret but it is massively faster than a naive 'play against the solver') .