Open Side Menu Go to the Top
Register
PioSOLVER - postflop equilibrium solver for Holdem PioSOLVER - postflop equilibrium solver for Holdem

03-28-2015 , 09:04 PM
ok - I am sure this has been asked before (but was hoping you had a better idea now) - First it is really cool that your implementing a more custom tree soon (next update?) - but my question is about a Min/Max exploit function and an E.T.A. on when that will be available? IMO, that is the next most important feature besides the bet sizes (custom trees) and possibly the most useful tool for dedicated users (players).
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-29-2015 , 05:54 AM
Quote:
but my question is about a Min/Max exploit function and an E.T.A. on when that will be available?
I am not sure what Min/Max exploit function is. I was talking to a lot of people about what they want here and most of the use cases come down to forcing one player's strategy on the flop and solving with that assumption. That's going to be there in the solver (but not in GUI yet) in the next update.
Maybe you could describe the functionality you would like to see a bit deeper? Preferably give me an example or two about what you you would like to be able to do.
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-29-2015 , 03:50 PM
I purchased the pro version and have been playing around with various flop textures to get my head around what it considers to be optimal strategies.

For the below example I looked at BB calling a Button open using typical ranges. I used a dry flop (K73 rainbow), had the BB check and then looked to see what the Button betting range should be. See the below image where highlighted the Button strategy and EV for A4s. In this case the button is betting 2/3 pot.

What I can't understand is why PIO recommends bluffing more with Ad4d which has less EV than the other 3 combos. The other 3 combos have backdoor flush possibilities and in real life those are the combos I would more likely bluff. This pattern seems to repeat across all the suited combos.

Would appreciate someone providing a likely explanation - it doesn't make much sense to me


PioSOLVER - postflop equilibrium solver for Holdem Quote
03-29-2015 , 04:25 PM
Quote:
Originally Posted by Ramblinman15
What I can't understand is why PIO recommends bluffing more with Ad4d which has less EV than the other 3 combos. The other 3 combos have backdoor flush possibilities and in real life those are the combos I would more likely bluff. This pattern seems to repeat across all the suited combos.

Would appreciate someone providing a likely explanation - it doesn't make much sense to me
If you think about clairvoyance game and how equities should be distributed that makes a perfect sense. Betting range always should be more polarized than checking range. Ideal betting range would constitute with proper ration of pure nuts hands (i.e. hands that never loses) and pure bluffs that never improves. Since that's impossible in texas holdem solution gravitates towards that anyway, by leaving more pure bluffs in betting range and less hands that have more equity (could win but not that often) which belongs to checking range more than to betting range.
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-29-2015 , 04:34 PM
Interestingly, all of those A4 hands have a higher EV for checking than for betting, but the hand that is bet most frequently is the one hand that has the highest difference EV(check) - EV(bet) = 23 in this case, and 14, 17, 17 in the other cases ... how does that come about ?
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-29-2015 , 05:02 PM
So this is difficult question. There are three aspects to it:

1)the solver isn't perfect
2)sometimes things are different than reality than common sense dictates
3)it's close anyway

Let me elaborate on those in relation to this case.

1)the solver isn't perfect

The solver works in such a way that it tries to move into direction of the most profitable action at every step. Unfortunately what is the most profitable changes depending on what the opponent is actually doing. This is the reason why finding equilibrium solutions is very hard - it's impossible to predict what the strategy for given hands is going to be at the end.

The solver doesn't care if EV of A4s with a backdoor is higher or lower than EV for A4s without a backdoor. What it does care about is EV difference. The bigger EV difference the more aggressively it will move the strategies in direction of more +EV action. It needs to be careful not to overstep though, as then opponents counter could bring severe damage to overall strategy.

In the situation you described, all the A4s's will go into checking direction next step but it's not clear how long that will continue.

What probably happened in this solution is that Ad4d was more +EV for betting than for checking and it didn't go back to the checking more yet (again, it's unclear if it should, it's impossible to predict how opponent's counter are going to influence EVs at next step.

2)
This is something which we just need to accept: sometimes it will be impossible to understand every specific detail of GTO strategy. The trees are huge and even slight river ranges implications can influence what the flop strategy is. The solver just looks at numbers and take it step by step. It knows what the EVs are now (that's the ones displayed in your screen) and it's clear it did pretty good job so far because the EVs of checking and betting are very close for this mixed action (in a perfect solution the differences would be 0).

It's worth noting that the mechanism for calculating EVs and exploitability is independent from the solver. It's a way to measure its performance and keep it in check. The solver tries to strike the balance and arrive and the best strategy possible - the evs and MESes are the way to assess how well it did.

3)
In theory if the action is mixed the evs for mixed actions have to be the same. That would be the case in the perfect world but the solver isn't perfect. As you can see though the evs are really very very close. Maybe if you bet every backdoor it would be too easy (well, from maximum exploitive strategy point of view) to use that information and maybe the solver tries to exclude the best backdoor combos (A4s is a pretty strong hand, it's not purely a bluff, something like Js9s is more of a bluff with good potential).
There is EV to be won by checking now... it's not about how good betting is but about how big the difference is between actions - here it's very close to 0.

Unfortunately I can't give you more satisfactory answer as I don't understand every nuance of GTO either. The solver is very new and probably over time people are going to pick up more patterns and the understanding is going to improve - if there ever is "human" explanation for that situation though I don't really know.

One way to discuss those things with other users of PioSOLVER is to join our Skype group and post questions there (PioSOLVER on Skype or drop me an email and I will add you to my personal one). Maybe someone will have some insight. More likely though this is something between points 1) and 2) but fortunately also qualifying for point 3) - the difference isn't huge and the solution isn't perfect yet.
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-29-2015 , 11:18 PM
before i buy the basic version i want you to tell me if i will have to buy your soft again after formatting

reason is i tend to format my system every 6 months so i just want to know if i will have to be buying your software everytime im formating

how will you handle this ?

table ninja system we can activate and deactivate the computers anytime so maybe youre folowing this system ?
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-29-2015 , 11:20 PM
and another queestion

if i buy the basic version i will be able to upgrade to the pro version anytime in the future for a discount or something?
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-30-2015 , 02:15 AM
Quote:
before i buy the basic version i want you to tell me if i will have to buy your soft again after formatting
No, you wouldn't need to. Just contact us before doing the re-format so we can deactivate the software. You can then activate it again on either the same or different computer.

Quote:
table ninja system we can activate and deactivate the computers anytime so maybe youre folowing this system
We have something similar but we restrict deactivations to reasonable amount (if you don't ask every 2 weeks it will be ok).

Quote:
if i buy the basic version i will be able to upgrade to the pro version anytime in the future for a discount or something?
The update to pro version costs 249$ which is 23$ difference between buying it straight away. The difference is there to cover our payment processing and licensing fees.
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-30-2015 , 05:48 AM
Thanks Punter, I sent an email to your support group to join the Skype community.

As for the A4s, it now makes a bit more sense when I look at the total combos being bet - see below image. Of the 4 available combos the solver says bet 1.08 combos with backdoor flush (As4s, Ah4h, Ac4c) and 0.64 combos without (Ad4d). So betting 43% of the time in total - 27% with backdoor flush and 16% with no backdoor.

I guess the solver is striving for balance across all combos....trying to reach a point of indifference between checking and betting.

I'm trying to map this against the toy games described in MoP - trying to see if the solver is bluffing at near optimal frequencies given the bet sizing (2/3 pot). The toy games in MoP are mostly half-street and in this particular case the solver is describing a strategy for the flop (with two streets to follow) - but I'm trying this mapping exercise to further my appreciation of poker.

Anyone trying to do the same with some helpful insights ?


PioSOLVER - postflop equilibrium solver for Holdem Quote
03-30-2015 , 08:19 AM
Quote:
As for the A4s, it now makes a bit more sense when I look at the total combos being bet - see below image. Of the 4 available combos the solver says bet 1.08 combos with backdoor flush (As4s, Ah4h, Ac4c) and 0.64 combos without (Ad4d). So betting 43% of the time in total - 27% with backdoor flush and 16% with no backdoor.
Just in case, you can go to:

tools->configuration->show total combos in range boxes

to switch the view from weights to combos. The default is weight (it's the same for first decision for hands if they were played 100% of the time preflop, because then starting combo is 1 but it is different in general case)

Quote:
I guess the solver is striving for balance across all combos....trying to reach a point of indifference between checking and betting.
The solver tries to improve moving the strategies in the direction of most +EV action at every step. The process often results in many indifference points but those aren't inherent part of optimal play - they are just very common symptom optimal solutions have.
There are situations where there aren't any indifference points though (like with very strong range against a weak one, every hand from the strong one would bet 100%).

Quote:
I'm trying to map this against the toy games described in MoP - trying to see if the solver is bluffing at near optimal frequencies given the bet sizing (2/3 pot).
This is a good exercise to try on river games. The optimal bluffing frequencies are different in multi street games. Even if we assume our standard game from MoP (0-1 and vbets/bluffs vs bluff catchers) the optimal bluffing frequencies raise with every additional street to play.
It's even more difficult in Holdem because not only it's multistreet but also hands change value as new cards come. It's really hard to produce anything similar to pure 0-1 games outside river games with carefully chosen ranges.
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-30-2015 , 10:27 AM
Quote:
Originally Posted by punter11235
I am not sure what Min/Max exploit function is. I was talking to a lot of people about what they want here and most of the use cases come down to forcing one player's strategy on the flop and solving with that assumption. That's going to be there in the solver (but not in GUI yet) in the next update.
Maybe you could describe the functionality you would like to see a bit deeper? Preferably give me an example or two about what you you would like to be able to do.
Sounds like you are already doing what I most want - I guess the best way I know of to explain the differences between min/max exploit - is w/ a pre-flop example. OK, so your playing HU w/ a guy who folds to 3b's 75% - max exploit would be to 3b him all the time - but he will eventually notice and you will lose the exploit - whereas a min exploit would still have you 3b much more often than you would vs a guy that defends more, but you will not be 3b so much that he can easily spot it and start adjusting. (tbh, it sounds like the min exploit function would be extremely hard to program and I wouldn't have any idea how to do it....I was talking w/ a friend and he explained it to me and said he had heard of some work being done on it).

Could you explain this a little more and what it means for someone w/ no programming experience at all (to me it sounds like "it will be possible but you would need to write a script to use it right away") -
Quote:
"forcing one player's strategy on the flop and solving with that assumption. That's going to be there in the solver (but not in GUI yet)"
I think this is the max exploit function but I am not 100% sure (just figured I would be greedy and ask for both and have all bases covered).
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-30-2015 , 10:44 AM
Quote:
tbh, it sounds like the min exploit function would be extremely hard to program and I wouldn't have any idea how to do it....I was talking w/ a friend and he explained it to me and said he had heard of some work being done on it
Maybe it's hard, maybe it's not. The problem is it's not very well defined concept.
You can probably calculate something kinda/sorta between equilibrium and max exploit vbased some assumptions about your opponent's strategy. I doubt we will be doing it anytime soon though as other things are more important now.

Quote:
I think this is the max exploit function but I am not 100% sure (just figured I would be greedy and ask for both and have all bases covered).
Max exploit only makes sense if you define a strategy in the whole tree. Changing just one node and assuming the rest stays unchanged is very easy to program but unfortunately almost completely useless as analytic tool.
There are two ways which are useful:

1)change and lock strategies in some decision nodes, recalculate the equilibrium
2)define strategy for the whole tree and calculate max exploit vs that

1) is doable and there are chances it's going to be available soon.
2) is very hard (the trees are huge) and would require you walking the whole tree or defining general conditions for behavior. This is actually a question of writing max-exploit AI which profiles the opponents. The project isn't close to being at this stage.
PioSOLVER - postflop equilibrium solver for Holdem Quote
03-30-2015 , 02:23 PM
OK yeah I was confused on the terminology - option 1 is actually what I have heard referred to as min exploit and option 2 as max exploit. I was confused because I thought they were essential the same thing only less extreme for min exploit. But yeah sounds like you are working on a useful tool, so thanks
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-02-2015 , 06:53 AM
Hey Punter,

I was trying to use the scripts, I am a little confuse at the results. I opt to run the script using piosolver-pro (right-click+open with piosolver-pro). The command prompt comes up and runs the program. But what happens after the program completes the task? I was using the example script 3. I cant find the solution anywhere and the script doesnt have an autosave command in it.
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-02-2015 , 08:53 AM
Quote:
I opt to run the script using piosolver-pro (right-click+open with piosolver-pro).
You can do that but easier is to start the PioSOLVER (double click on PioSOLVER-pro.exe, the console with welcome string will pop-up) and type: "load_script yourscript.txt" (without the quotes)

Quote:
I was using the example script 3. I cant find the solution anywhere and the script doesnt have an autosave command in it.
It's possible that the bug with spaces in the file paths is still there in your version (will be fixed for good this Sunday for everyone). If your PioSOLVER is in a location which has space in its name, like:

C:\documents and settings\user\something

Then that will be truncated at the location of the first space.
The files should be saved though and you can look for them using some decent windows file search, like this one: http://www.voidtools.com/ (it's 100x better and faster than built in Windows search)

The best for now is to move the whole folder with the solver to something like: C:\PioSOLVER or wait till Sunday for it to be fixed.

Quote:
and the script doesnt have an autosave command in it.
There is dump_tree in that script:

Quote:
#using rebuild_tree saves time for allocating/deallocating the trees
set_board Ks 3h 2h
rebuild_tree
go 300 seconds
wait_for_solver
dump_tree 2bKs3h2h.cfr
It's in the last line in this fragment (which is repeated for every new tree)

Last edited by punter11235; 04-02-2015 at 09:05 AM.
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-02-2015 , 09:20 AM
did not realise the dump_tree was the save command. That make sense now, the command prompt says

ERROR: dump_tree couldnt open specified file.

Must the script file be located in a specific place on my computer?

My piosolver folder is on the desktop, there isnt any space in its name.
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-02-2015 , 09:39 AM
Quote:
Must the script file be located in a specific place on my computer?
Well... it depends on OS behavior (if you do right click thing) because PioSOLVER saves in "current directory".
I am not sure how Windows behaves here to be honest. The best would be to put the script in the same folder as PioSOLVER (for now), double click PioSOLVER.exe and then using load_script command. If that works then we can investigate further.
I suspect that if you do it the way you described (right click->open with) then the solver tries to put a save where the script is located. I will look into that later, for now maybe try the way I described above.

Quote:
My piosolver folder is on the desktop, there isnt any space in its name.
Desktop is usually located in:
C:\Users\username\Desktop

If your username contains a space it will not work (for now, it will be fixed this Sunday).
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-02-2015 , 10:07 AM
I retry it putting the script in the same folder as PIOSolver and you where right, it saves the file in the same folder as the location of the script. I previously had it somewhere with space in the name, thats probably the reason it couldnt save earlier.

thx
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-02-2015 , 12:10 PM
Hi it would be nice if piosolver detected PokerStars is running and auto close itself, becouse its porhibided soft and one could get ban if he do not remember to close it during session couple of times.
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-02-2015 , 12:24 PM
Quote:
Hi it would be nice if piosolver detected PokerStars is running and auto close itself, becouse its porhibided soft and one could get ban if he do not remember to close it during session couple of times.
I've already contacted them. They understand it's a new software and their policy is very new as well (they put it in the same list as ICMIZER/CREV/Coffe Calc etc. about 4 days ago).
The next version of PioSOLVER will detect poker sites running. It won't auto close though but it will make noisy in you face warning asking you to close one of the two (the solver or the site) the moment you open a poker client (so it will get in your face before you click log in).
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-05-2015 , 08:51 AM
Hello all,

Today we are updating the solver and the viewer. You can read about details on our blog.

The biggest change is that it's now possible to build trees with multiple bet sizes. Amount of bet sizes is for now limited to 10 per decision node. Games from turn/river should be very manageable with as many bet sizes as you wish. Here are some examples of flop spots which are solvable:

0)up to 4GB of RAM:
-almost every sensible turn/river game with 2-8 bet sizes, those are small and fast to solve;

1)up to 8GB of RAM:
-100bb single raised pot without flop donkbet and 2nd river sizing added (on all rivers)
-100bb 3 bet pot with 2 sizes everywhere (say 40% and 67%)
-25bb tree with 2 cbet sizes and 2 river sizes everywhere
-CAP tree with all-in added everywhere

2)up to 16GB of RAM:
-100bb single raised pot with 2 river sizings (every river)
-100bb single raised pot with one flop sizing but 2 turn and river sizings (no flop donk bet)
-as above but with donkbet but only one sizing if there already was a raise
-25bb tree with 2 bet sizes everywhere
-CAP tree with 3 bet sizes everywhere but without flop lead OOP
-limit holdem with cap 4 (standard) without OOP flop leas

3)up to 32GB RAM:
-full 100bb tree with 2 bet sizes everywhere (for example 40% and 70%)
-full CAP tree with 3 bet sizes everywhere
-full limit Holdem spot with CAP 4

How it looks like:

1)setup (for CAP with 2 bet sizes everywhere)


2)browsing:



3)Toy river game:



The schedule for updates:
-edge customers have their download and notifying email already delivered
-pro users can expect it later today
-basic users can expect the update tomorrow
-free version will be updated later this week; multiple bet sizes (maybe a bit limited) are expected to make it to there as well

Schedule for near future:
-add some missing basic functionality of the solver/viewer (including setting/locking strategies, range analysis, expanding chosen nodes etc.)
-optimize speed/memory so all above mentioned settings go down at least one category down in memory requirements (and hopefully one generation of CPUs as well but that's a bit more problematic)

Stay tuned!

Last edited by punter11235; 04-05-2015 at 09:16 AM.
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-06-2015 , 08:18 AM
Everything is great, but you could leave old bet size vizard option. For qiuck simulations.
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-06-2015 , 01:37 PM
Quote:
Everything is great, but you could leave old bet size vizard option. For qiuck simulations.
Yeah, you are not the first to ask for it. I didn't expect to be as popular.
We will add it in the next release, maybe we will make a quick bug-fix one and sneak it in earlier.
It's still available from scripts as build_tree_old just in case you use those.
PioSOLVER - postflop equilibrium solver for Holdem Quote
04-06-2015 , 02:35 PM
booked^^
PioSOLVER - postflop equilibrium solver for Holdem Quote

      
m