Open Side Menu Go to the Top
Register
Structuring solver study (or not?) Structuring solver study (or not?)

08-23-2024 , 10:56 AM
Hi,

I recently bought GTO+ but am struggling with how to go about studying GTO, particularly in relation to the turn and river. I've watched various videos on this topic but they all seem to cover cbetting and then say at the end [oh, you can repeat this for later streets]. The problem I have is that cbetting covers only one node, but by the time one gets to the turn there are already several variations one needs to cover (different cbet sizes, xx, x/c, x/r/c, not to mention the different turn cards).

My question is, how do you look at hands when you study them in a solver? Do you look at the whole hand? One street at a time, across similar boards? One node (e.g. x/r and then turn barrel, again across similar boards)?

Thank you for any help,
C

Last edited by Chrisher; 08-23-2024 at 10:59 AM. Reason: minor edit
Structuring solver study (or not?) Quote
08-23-2024 , 12:30 PM
It's actually a very good question. Depending on the hand you'd go through the other nodes, but really it is tough to do everything at once. The short answer is it really depends on how in depth you want to go and what question you specifically want to answer and how much information you want to take into account for finding that answer.

Typically I'd take a hand or common node and then look at the general heuristics and see how they differ compared to an equilibrium output. For example, maybe you're looking at some hands from SB vs BB and you're the SB. First, you'd take a look at what your opponent/population is doing and compare it to the equilibrium output... say how often they fold to c-bets, raise your c-bet, and then how often they bet vs. check or check behind. You can easily look at some future street information to get some information on how you might want to construct your flop strategy as well: i.e. how often they fold to delayed bets on turn/river in check lines and how often they simply fold to a bet on each street. Identify the weaknesses/large deviations, use some logical reasoning to make a hypothesis on how your strategy should look, and then spend some time nodelocking and adjusting the solver's strategy and see if it matches up with your hypothesis.

This is very time consuming, but it does provide the most information. Honestly, most videos I see for reviews are actually quite superficial in nature and sort of match up with what you said. You can opt in to either approach, but ideally you'd go rather in depth and then take some notes. You can also compare/contrast your findings with your current strategy and hopefully see in which direction you want to go in.

Really there are four main steps in improving:

1. Collect data
2. Interpret data
3. Plan on a new strategy
4. Execute the new strategy

For 1) it's going to mostly be playing, but you could also use a video/coach/hh review/buy hands etc.

For 2) you'd use software (GTO+, PIO, coach, friends/review etc.). You can also choose how complex you want the data set to be... as you say looking at turn/river strategies and tendencies will certainly have an influence on flop strategy, but it is more time consuming

For 3) Taking notes, discussion with other people, and identifying specific instances you want to change are important. Compare/contrast your plan with your current strategy so you can better understand what changes you're going to try to implement. In my opinion it is usually a good idea to quantify these things at least somewhat... So if you're planning on increasing/decreasing cbet frequency then you want to see what it currently is and make sure it changes in future (which is back to 1.)

For 4) You'll want to go in to sessions with the plan of executing new strategies. It's very very easy and common to revert back to old tendencies. It requires a conscious effort to implement new strategies and information.

To circle back around to the original question on basically how to review the hand, I'd suggest you go into the review/study session with a specific question in mind... then identify what data would influence the answer, search for that data, interpret it, and then use software to help arrive at a conclusion. It's really crucial you take notes, so you have something concrete to take-away from the study session and then work on implementing whatever findings you found in the future. Again, circle through steps 1-4 repetitively is important. Improvement is an incremental process and it takes a lot of iterations to notice a solid improvement.

Let me know if that answers your question OP.

Last edited by Brokenstars; 08-23-2024 at 12:35 PM.
Structuring solver study (or not?) Quote
08-23-2024 , 02:59 PM
First of all, thank you for this reply, I really appreciate the effort put into it.

If I understand, your approach to solver study focusses on exploiting Villains' imbalances using data analysis, node-locking most simulations you run. I am quite new to poker (maybe 6 months in) and do not have a good grasp of what the equilibrium solutions look like, as such I've been studying equilibrium solutions almost exclusively, very rarely node-locking at all, albeit with the plan to do exploitative work down the line. In this sense I guess I have been following steps 3) and 4) - creating heuristics from the solver and then implementing them in game, perhaps exploiting Villains by modifying frequencies to become pure actions (and sometimes further, given the huge imbalances at lower stakes).

When you say "go into the review/study session with a specific question in mind," would you be able to please give me an example, just so I can see how broad/specific such a question should be? Should it be e.g. how do I play connected, rainbow boards, or more specific e.g. which hands should I barrel with on connected, rainbow boards after check-raising a 33% flop cbet.

I have been trying to take notes on hands I look at in the solver, do you have any guidance on how to structure this? The note-taking process reflects the struggle I am having with structuring solver study in general. Initially I was looking at rainbow, triple Broadway flops, BUvBB. I took notes on cbet strategy - which hand classes did what, and it worked quite well - I developed several heuristics to help me e.g. Two-tone boards are bet smaller compared to their rainbow counterparts. When it comes to the turn though, I began to struggle. I tried to take notes across similar boards (rainbow, Triple Broadway) on strategy if the turn was a gutshot completing card, for example. However, some of the flops had different flop cbet sizes and so come the turn the SPR and ranges were very different, making it hard to compare between boards. Or instead of b/c, the flop went b/r/c, leading to another entirely different range/SPR combination.

I've just realised I'm rambling, writing has never been my strong point so I apologise for that. Is there any better approach that you know of when it comes to note taking across several, similar boards, particularly turns and rivers? As mentioned, I tried taking notes on 'turn barrelling after b/c' across these boards but there's just so many different branches of the game tree to make this unfeasible.

C
Structuring solver study (or not?) Quote
08-23-2024 , 06:04 PM
Sure, I'll give you an example of a hand I recently played.

    Poker Stars, $1/$2 No Limit Hold'em Cash, 4 Players
    Poker Tools Powered By Holdem Manager - The Ultimate Poker Software Suite.

    BB: $228.07 (114 bb)
    CO: $276.78 (138.4 bb)
    BTN: $760.90 (380.5 bb)
    Hero (SB): $257.65 (128.8 bb)

    Preflop: Hero is SB with A 2
    2 folds, Hero raises to $6, BB calls $4

    Flop: ($12) T 2 A (2 players)
    Hero checks, BB bets $8.55, Hero calls $8.55

    Turn: ($29.10) 8 (2 players)
    Hero checks, BB bets $20.74, Hero raises to $72.59, BB raises to $213.52 and is all-in, Hero calls $140.93

    River: ($456.14) A (2 players, 1 is all-in)

    Spoiler:
    Results: $456.14 pot ($3.00 rake)
    Final Board: T 2 A 8 A
    BB showed A 4 and lost (-$228.07 net)
    CO mucked K 4 and lostBTN mucked 2 5 and lostHero showed A 2 and won $453.14 ($225.07 net)


    So, some questions:

    1. With made hands is it better to bet or check (i.e. what gets more money in more often)

    2. With made hands should I x/r the flop here?

    3. With made hands should I x/r the turn here?

    ---------------------

    So, some notes for my population that I knew before going into this hand:

    a) I believed my population overfolds vs. small c-bet otf SB vs BB
    b) I believed my population would bet flop vs. check here higher than equilibrium
    c) I believed my population overraises vs. small c-bet otf SB vs BB
    d) I believed solver would rarely x/r flop vs. large sizing here, but wasn't sure
    e) I believed solver would rarely x/r turn vs. large sizing, but wasn't sure
    f) I believed my population would probably give up on a lot of rivers if I x/c turn
    g) I believed my population would not fold very often vs. my turn x/r (like never fold AX)

    So, those were my thoughts during the hand and now to verify/see if those were the case. I also wasn't really sure if I should just bet small and face a higher frequency of x/r or check to them and have them bet at what was probably a higher than equilibrium value... I also wasn't 100% what those frequencies would be on a semi-dynamic two-toned A-high texture.

    Quote:
    a) I believed my population overfolds vs. small c-bet otf SB vs BB
    My populations folding frequency in BB vs SB SRP vs. a c-bet sizing of 25-35% are the following for these board textures:

    all boards: 27.8% ---> PIO = ~24.9% ---> slight overfold
    two toned: 28.2% ---> PIO = ~21.7% ---> large overfold
    A high : 32.2% ---> PIO = ~32.2% ---> medium overfold
    A + tt : 30.8% ---> PIO = ~26.2% ---> medium overfold

    ^ looks like my thoughts were correct, population overfolds a bit vs the small sizing

    Quote:
    b) I believed my population would bet flop vs. check here higher than equilibrium
    My population's bet vs skipped cbet otf in BB vs SB SRP are the following for these board textures:

    all boards: 43.7% (any size) ---> PIO = ~45.4% (only 30% sizing allowed)
    tt boards: 44.7% (any size) ---> PIO = ~46.4% (only 30% sizing allowed)
    A boards: 34.5% (any size) ---> PIO = ~36.7% (only 30% sizing allowed)
    A + tt : 35.7% (any size) ---> PIO = ~38.1% (only 30% sizing allowed)

    Frequencies were comparable, but for the agg report I was looking at the solve used only allowed a small bet sizing, so comparably population is likely betting a bit more often than they should be.

    Quote:
    c) I believed my population overraises vs. small c-bet otf SB vs BB
    My population's raise vs. small c-bet otf in BB vs SB SRP are the following for these board textures:

    all boards: 16.0% ---> PIO = ~ 9.6%
    tt boards: 16.6% ---> PIO = ~10.2%
    A boards: 11.7% ---> PIO = ~ 5.5%
    A + tt : 12.1% ---> PIO = ~ 5.9%

    So, vs. small sizing my population raises quite a bit more than PIO. It also looks like they do it a lot especially on two toned boards.

    Quote:
    d) I believed solver would rarely x/r flop vs. large sizing here, but wasn't sure
    equilibrium output: SB check F, BB bet 75%, SB response:

    ^ looks like low x/r is correct, centered around TT and AT

    Quote:
    e) I believed solver would rarely x/r turn vs. large sizing, but wasn't sure
    equilibrium output: x-bet-call flop, check turn - B67

    Surprised by low folding freq, likely function of inefficient turn sizing by IP... Also would be good to do further analysis on population's turn sizing strategy here has b67 I'd perceive as relatively weak/capped likely more towards middling AX and possibly some combo draws like 2Xdd, 8xDD, or TXcc.

    Quote:
    f) I believed my population would probably give up on a lot of rivers if I x/c turn
    Pop's bet frequency by street (all boards): f/t/r = 43.8, 47.7, 39.1
    Pop's bet frequency on turn if they bet flop (all boards): t = 54.9, r = 40.7
    Pop's bet frequency on river if they bet flop + turn (all boards): r = 45.4

    It does seem like there is a little drop off on river aggression. For this specific hand pio follows through on rivers ~50% of the time.

    Quote:
    g) I believed my population would not fold very often vs. my turn x/r (like never fold AX)
    In aggregate it was 54.8%
    When flop bet = true it was 58.9%
    Flop bet = true + A high flop = 46.2
    Additional filters I don't really have sample size to evaluate.

    Nodelocking:

    Here is equilibrium output for the sim I created for BB vs SB b30:



    Here is equilibrium output for the sim I created for BB vs SB where SB x:



    ^ BB only betting 22% here which I don't think will ever happen

    Nodelocks:

    BB vs SB b30: increasing raise freq from 9.5 to 12.1, I also adjusted more of the 2x to less raising and more of the flush draws, made hands, and straight draws to more raising

    BB vs SB b30: marginally increasing fold freq with some of the ~0 ev hands



    BB vs SB; SB check: increasing bet v check freq from 22% to 36%

    I specifically reduced 2x betting, increased betting from all AX significantly, all FD significantly, and added some more air and middling hands



    Conclusion:

    With the above nodelocks, the sim heavily prefers checking. C-bet frequency goes from over 50% to around 11%. Strong made hands (AT+ are pure checks), middling hands go into a mix with a neutral EV difference here, but in reality would also likely prefer checking since population will overfold to a delayed c-bet line here in my opinion.



    The semi-surpise was how often oop is x/r. And I'd have to go into more depth here to determine whether or not I would want to implement that.



    Back to beginning and answering the original questions:

    Quote:
    So, some questions:

    1. With made hands is it better to bet or check (i.e. what gets more money in more often)

    2. With made hands should I x/r the flop here?

    3. With made hands should I x/r the turn here?
    1. Looks like I should check a lot

    2. Looks like x/r aggressively is very reasonable, esp with a mergey range (would want more analysis on turn/river player in bet-call node)

    3. Population's bet-fold turn value was 40.9 over all boards which seems pretty reasonable. I didn't have much data for further filtering, so this one is a bit of a toss up in terms of conclusion, though the hand itself provides a data point it sort of aligns with my original thoughts and don't want to put too much weight into it because of that.

    So you'd likely go through something like this, evaluate your current strategy and probably also come up with some new questions and continue down that line. In this example you can draw a lot of other conclusions aside from just answering the original questions, and in addition to that come up with a lot of other questions. You can also analyze population values based on villain archetype as well as see if your conclusions for the specific hand apply to other board textures or similar situations. Again, in my opinion it is best to be as thorough as possible and draw as many conclusions as you can. You should make sure you take thorough notes.

    Last edited by Brokenstars; 08-23-2024 at 06:12 PM.
    Structuring solver study (or not?) Quote
    08-24-2024 , 05:18 AM
    As far as I understand, your question boils down to "poker is really complex" and yes it is.
    You can structure your study in different ways but Ill give you some beginner tips.
    1. Dont try to memorise postflop solver stuff. It doesnt help, you can't. You will naturally remember more stuff with time spent in the solver.
    2. Its not about GTO. Its about understanding how stuff works, also about developing the skills to process poker information in real time.
    3. Dumb it down. Depending on your skill level, you will have to dumb it down more or less. You cant pretend that youre more skilled than you are since it has to do with neural connections you strengthen over time.

    Trying to learn GTO before exploiting is not a good approach. Its procrastination and its an attempt at ensuring you never give your opponents a chance to beat you.
    Your opponents do have a chance to beat you and you cant change that.
    Its also not possible since GTO is not based on some mafic rules that differ from exploits, they go hand in hand.

    Last edited by aner0; 08-24-2024 at 05:25 AM.
    Structuring solver study (or not?) Quote
    08-24-2024 , 08:58 AM
    For the most part I also agree with everything Anero said. I've overlooked the fact you said you only recently started. You should really have a simplified approach to begin with and just look for very obvious spots to exploit and keep it simple stupid (KISS). Strive to achieve a reasonable understanding of theory and start with memorizing/implementing basic preflop ranges (RFI/3b/BB defense).

    Getting into solvers after only 6 months is very early. You do seem intelligent in your posts, but such an approach could lead you to misunderstanding a lot of things/implementing things wrong/straight up burnout.
    Structuring solver study (or not?) Quote
    08-24-2024 , 01:05 PM
    You might be better off focusing on 3b pots where the ranges are narrower and the lessons more obvious/tangible. Look for patterns. Build up a modest library of solves from the subsets that you can dip in and out of to test theories.

    You can also drill spots, eg SB 3b BU, in the trainer. If you haven't yet you can break the game down into simpler common scenarios (e.g. blinds 3b BU/CO, or BU/CO/HJ 3b UTG) where range variation is slight enough there won't be that much difference in how GTO goes about playing its range.

    Play hand...make questionable play...check hand after sesh in solver. If play was 'correct' feel proud, if play was bad say 'gto is useless' out loud.
    Structuring solver study (or not?) Quote
    08-24-2024 , 03:38 PM
    This is incredible Brokenstars, thank you! I think I will find this very useful going forward, as guidance.
    Structuring solver study (or not?) Quote
    08-24-2024 , 03:42 PM
    Thank you aner0, I think you are spot on. I was initially drawn to GTO because I liked the idea of 'unbeatable' poker but I see what you mean when you say that it's inseparable.
    Structuring solver study (or not?) Quote
    08-24-2024 , 03:46 PM
    I agree with your recommendation to start with the 3bet pots (or even 4bet) I think Ceres, I was musing on that earlier. Bigger EV differences and importantly the simulations don't take as long to run, as I might be quite inefficient at running them to begin with. Thank you for your response!
    Structuring solver study (or not?) Quote
    08-24-2024 , 04:01 PM
    I bought the solver, Brokenstars, after reading 'Modern Poker Theory,' all the solver print-outs made me want to start doing them myself. I'm playing NL25, trying to work my way up, and as you say, I was applying solver principles inappropriately against the vastly unbalanced (mostly calling station) population. I'll try to keep it simple going forward. Thank you once again. C

    Last edited by Chrisher; 08-24-2024 at 04:04 PM. Reason: minor
    Structuring solver study (or not?) Quote
    11-06-2024 , 11:23 PM
    Commenting for later reference. Amazing stuff.
    Structuring solver study (or not?) Quote
    11-10-2024 , 01:39 AM
    Some good stuff in this thread so far. I'm not the best at utilizing solvers, but I can be pretty lazy so I try to maximize the value of the time I do spend studying.

    A few things I do:

    1. On the flop look for patterns. Which hands are pure bets or pure checks and why? Look at hand classes vs individual hands.

    What percentage of flush draws bet vs. check? Why? How about top pair hands? Middle pair, bottom pair, gut shots, etc.? If certain hand classes mix a lot, what characteristics determine which hands are bet more often?

    2. On the turn it is very helpful to look at the heat map that compares each player's equity on various turn cards. Again look for patterns. When the turn card gives you a significant range advantage how does your strategy differ from when the card is bad for your range?

    Basically you want to find generalizations. Stuff like if your opponent's range is very strong then you tend to want to do a lot of checking. That type of stuff.

    Also thinking about how people's actual play differs from what the solver does is helpful, like brokenstars was talking about. These types of exploits are a little more advanced though. Anyway good luck at the tables.
    Structuring solver study (or not?) Quote

          
    m