4 Tips to better use NLH poker solvers in 2021
A brief overview of basic poker solver mistakes and best practices
By Brandon Wilson
Solve to a sufficient Nash distance
A common mistake is to terminate solves prematurely or assign an insufficient Nash distance from the outset. DTO recommends solving down to sub-0.5% of the pot, at least. Simulations that don’t sufficiently converge will produce unreliable and more exploitable results. One telltale sign that a simulation has not sufficiently converged is a holding mixing its action frequency despite showing different EV numbers between actions. In true GTO play, actions only mix when they have the exact same EV.
Construct parameters around what you expect to implement
If a solver’s parameters include three flop bet sizes for a range, but in the hand in question, the player would have employed, say, two sizes, then whatever question is posed to the solver will receive a much less precise answer (more on asking better questions ahead). Of course, players don’t know exactly what strategy their opponent plays in a given hand, but input parameters can range from reasonable (evidence-based, inputting only the strategy one encountered in-game) to highly impractical and not useful for study (allowing four check-raise sizes or no all-ins on any street).
Furthermore, game conditions must be accounted for in the simulation’s parameters. For example, solving a final table hand without including ICM considerations like stack distribution and payouts in the inputs is a mistake (insofar as the user if attempting to elicit useful ICM concepts). Similarly, reviewing multiway spots as if they were heads-up presents a similar issue. While some concepts from heads-up pots may carry over, a specialized multiway solver would more precisely answer a user’s questions and avoid erroneous takeaways.
For a more detailed explanation of the ramifications of solver parameters, see DTO’s article on the pitfalls of using pre-built solutions to study.
Ask good study questions
If solver use is contained to determining whether the computer merely takes one’s chosen line at any frequency (“Was this a thing?”), learning will be limited and important concepts overlooked. This binary form of study misses a crucial component of solver study: discovering underlying equilibrium mechanics, or “why.”
To that end, develop clear questions. “Should I have a big bet range on this texture? How much EV is lost if I don’t?” is a good solver question. As is “What are the shared or unshared properties of non-made hands that prefer to bet here?”
Most importantly, questions should be range-focused as opposed to hand-focused. Because range frequencies are the foundation of NLH, myopically examining your specific holding’s action can mislead if it encourages ignorance of the range’s overall frequencies. Confirming or disproving the validity of a line you may have taken with a specific holding is much less important to progressing than understanding what the entire range is supposed to be doing in any given spot.
Run separate sims to determine optimal bet sizes
A reasonable question to which an answer is often unsoundly sought is, “Which bet size is best here?” A user may input four sizes, run the sim, and conclude that the one most often used by the solver is the “best.” While in cases where there is a size used overwhelmingly more than any other, the answer may withstand scrutiny, in cases where different bet size frequencies are close, deciding that an option used, say, 2% more than any other is the best could be wrong.
A better way to determine bet size EVs (a method utilized in the creation of DTO Cash) is to create separate sims with consistent independent variables and solving for the “best” bet size, the dependent variable. For example, instead solving of a game tree consisting of b25, b75 and b120 for a given node, solve three separate game trees with one of the three featured in each, then compare the EVs of the player in question’s overall strategy across all three. In doing so, you may find that a simplified, single-size strategy is best (and easier to implement), or you might uncover a mechanic that proves it is important to employ multiple bet sizes insofar as the EV of not doing so pales in comparison.