How do solvers calculate what to do?

Your approach would be an option.

Different solvers use different approaches.

Best results tend to get algorithms that "don't know" the rules just the pay-offs.

Instead they populate a tree. Each node is a choice that stores the best possible return. From this the algorithm calculates a probability to pick an option. To avoid discarding better branches by accident during training phase. After the result the updated values are stored by backpropagination.

If you check the solver results you will shown the best choices.

I don't know how helpful this was as I'm a little bit distracted right now. But if there is a specific question to one of the approaches I might try to clarify it a little.

There's 9 players and 9 different hands in one hand at poker table. Each of these hands has some % of chance of taking the pot down. Solver calculates the % of chance according to the hand value, ~AA has ~80% 1vs1 player, 72o has less than ~30% etc. More complex solvers also calculate future EV in tournaments according to payouts.

In-game the solvers are banned by pokersites, and even if you use another computer, they would note that that player played exactly according to solver, as they have all the solvers. So they would notice that your betting pattern is that exact pattern from that solver.

PokerStrategy once had Equilab which calculates both post-flop and pre-flop equities of hands, can be used for learning equities but is also banned while a poker site is open.

Originally posted by Rhodriguez
Your approach would be an option.

Different solvers use different approaches.

Best results tend to get algorithms that "don't know" the rules just the pay-offs.

Instead they populate a tree. Each node is a choice that stores the best possible return. From this the algorithm calculates a probability to pick an option. To avoid discarding better branches by accident during training phase. After the result the updated values are stored by backpropagination.

If you check the solver results you will shown the best choices.

I don't know how helpful this was as I'm a little bit distracted right now. But if there is a specific question to one of the approaches I might try to clarify it a little.

Different solvers use different approaches, but to arrive at the same solution, right? Or not?

Maybe I could have asked it a better way: not really need to know how, but rather what they calculate. Just to understand them and to interpret them.
So if I understand right: for all the options available it calculates what mathematically the most profitable distribution of % (for the options) vs villain's range is? I suppose by looking to future streets as well, and knowing villains' ranges? (and by whatever gives the best return value or something, and then it calculates backwards to correct things I understand) .

Thanks Rhodriguez!

Originally posted by taavi1337
There's 9 players and 9 different hands in one hand at poker table. Each of these hands has some % of chance of taking the pot down. Solver calculates the % of chance according to the hand value, ~AA has ~80% 1vs1 player, 72o has less than ~30% etc. More complex solvers also calculate future EV in tournaments according to payouts.

In-game the solvers are banned by pokersites, and even if you use another computer, they would note that that player played exactly according to solver, as they have all the solvers. So they would notice that your betting pattern is that exact pattern from that solver.

PokerStrategy once had Equilab which calculates both post-flop and pre-flop equities of hands, can be used for learning equities but is also banned while a poker site is open.

Thanks taavi1337, I did not know Equilab is gone. That was a great piece of software. Maybe that's why I bought something else. But it's not a complete solver.

Anyway, I saw that gto+ only has postflop stuff, and the preflop stuff in gtowiz is free. I think I'd better start with preflop stuff and that will take a long time. Maybe I'll buy 1 month of gtowiz and look in more detail at preflop ranges (assuming you can change villain's ranges to non-gto ranges and make preflop gto calculations based on that, but maybe wrongfully assumption)

Originally posted by Stackoffski
Different solvers use different approaches, but to arrive at the same solution, right? Or not?

Maybe I could have asked it a better way: not really need to know how, but rather what they calculate. Just to understand them and to interpret them.
So if I understand right: for all the options available it calculates what mathematically the most profitable distribution of % (for the options) vs villain's range is? I suppose by looking to future streets as well, and knowing villains' ranges? (and by whatever gives the best return value or something, and then it calculates backwards to correct things I understand) .

Thanks Rhodriguez!

I don't now how to answer or how pedantic I should be.

Because poker is a zero-sum game the algorithm of (AI) solvers calculate the opposite. Where you lose the least amount if villain plays perfect. The so called counterfactual regret minimization.
The solver has stored the value for each situation you are selecting the value. Which is a result of future streets. Again in hindsight after you know where you would regret the least.
It's not an equity calculation. It's the result of a simulation.
The results of different solvers should be close but depending on options, assumptions and how they cluster data the results might be a slightly different.

Originally posted by Rhodriguez
I don't now how to answer or how pedantic I should be.

Because poker is a zero-sum game the algorithm of (AI) solvers calculate the opposite. Where you lose the least amount if villain plays perfect. The so called counterfactual regret minimization.
The solver has stored the value for each situation you are selecting the value. Which is a result of future streets. Again in hindsight after you know where you would regret the least.
It's not an equity calculation. It's the result of a simulation.
The results of different solvers should be close but depending on options, assumptions and how they cluster data the results might be a slightly different.

Ah I see, that makes it more clear for me ... man I'm learning some new words and terms ... pedantic, zero-sum game, counterfactual regret minimization, thanks google ...

It doesn't even try to win anything, just tries to avoid losing by method of simulation and choosing best option, which can't be a winning one in the long run if villain is also a solver I assume. So it loses as much as it wins with every decision if it play's against itself. Some assumptions though ... hope I'm right, don't want to waste your time ... I need all that I need to know!

Thank you again

Pretty much. The solver plays against itself in two phases. Because what one player wins the other loses, game theory has proven, minimizing losses/regrets is the best strategy. This equilibrium guarantees that if one player deviate from the strategy it gets punished and loses money.

The solver will show you for each decision how much you win or lose with those cards or range.

The difficult part is when the solution is a mixed strategy. Just pruning the less likely or picking the option for a higher ev is a deviation.

I looked on the homepage, and the entering a villains range is in the elite tier - if I understand them correctly.

Have fun learning gto, recognizing where the villain makes mistakes and then exploit!

...

Originally posted by taavi1337
PokerStrategy once had Equilab which calculates both post-flop and pre-flop equities of hands, can be used for learning equities but is also banned while a poker site is open.

Equilab still exists, both as a free version, and a paid version. Both are still useful for hand analysis.

There is an Omaha version as well.

Originally posted by Rhodriguez
Pretty much. The solver plays against itself in two phases. Because what one player wins the other loses, game theory has proven, minimizing losses/regrets is the best strategy. This equilibrium guarantees that if one player deviate from the strategy it gets punished and loses money.

The solver will show you for each decision how much you win or lose with those cards or range.

The difficult part is when the solution is a mixed strategy. Just pruning the less likely or picking the option for a higher ev is a deviation.

I looked on the homepage, and the entering a villains range is in the elite tier - if I understand them correctly.

Have fun learning gto, recognizing where the villain makes mistakes and then exploit!

It's like a tug of war (rope pulling contest) with the nash equilibrium in the middle.

I also hope to see where I deviate much from gto myself of course

Originally posted by Untiltable
Nodelocking (ie. range editing) and custom AI solutions aren't available until you get the Elite $129 per month option, so you might want to go with something else if you're just looking at preflop.

Ah I see, thanks for the heads-up!
And now I know what nodelocking is ...

--
And gracias to Vorpal:

Originally posted by VorpalF2F
Equilab still exists, both as a free version, and a paid version. Both are still useful for hand analysis.

There is an Omaha version as well.