Thanks for the quick answer! Yes the spot is a forced 3b spot so could be that I'll just have to stop it at some point if the explo doesn't seem to go down much further.

Hello guys.I have Question.

this is not open source so can not grasp mechanism and reliability.
ex.said GTO-solution..what mean? analytical? numerical? approximate?
so,someone or developer proved fundamental reliability of piosolver?

Hi punter, would you please recommend some single posflop sizes for us, at least as a starting point in solving 100 bb preflop trees?

Wait. Why is it bad to have "all in" checkbox checked for pf sims, and how could it affect the ranges in a bad way, making them less accurate?

It likely won't improve. Another factor which hurts the preflop solver a bit are smaller starting ranges (which is likely in your case). I already have some test cases for it and the situation improved in dev but it's still not solved entirely.

Opening the source wouldn't change much as verifying that the code works correctly is an impossible task.
We made an effort to make results exportable and browseable in many ways. We display EV of every action at every point. We also compared our results to other implementations we have as well as to those of other people (namely Oskari and his solver, Oskari's work got published in Science). Verifying things on your end would require coding your own solver (or at least a tool to verify EVs) and comparing the results but you can also try doing it manually on some toy cases.

Those are apprxoimate equilibria. We display exact distance to the equilibrium as well.

Proving software correctness is an impossible task. A lot of people did work with Pio and compared it to other solutions though. We also have several implementations which we can compare and verify the results against. At this point correctness is the least of our worries (there are thousands of people using Pio every day, many developers doing tools working with it, there was never a case of Pio producing incorrect results).

It's reasonable to use small'ish cbet (say 40% or 50%) and then 75% bets on the turn and river. 3x raise is reasonable as well. The EVs don't change that much so it's not very important to get those right.

The preflop solver operates on limited precision (to save memory). Having multiple bet sizes (and especially overbets) make it difficult for it to get the EVs right because of it. The situation is going to be improved in the future so accuracy won't be much of an issue but the part about it being not very productive (EVs don't change much so multiple bet sizes don't influence preflop ranges almost at all) is going to stand. I recommend using bigger subsets or adding more interesting preflop options instead of making postflop play more detailed when solving for preflop ranges.

Does this mean that preflop ranges from simulations with multiple sizes (or allins added) will be less accurate? Or will the simulations just take longer to converge, but be fundamentally accurate?

Hi guys im going to buy a new computer, but i was wondering if a computer with this specifications can run PIO easily?

Intel® Core™ i7-6700 processor
16 GB RAM
240 GB SSD + 1 TB hard disk
Intel® HD Graphics 530 (onboard)

thanks in advance!

It's less likely to converge well.

This is a very reasonable setup for the postflop solver and general computer usage although not enough to run things like the preflop solver.

I am not going to do those philosophical discussion.
in general,solving program has specifications for sale.

for example, I have degree of numerical analysis...then computational fluid dynamics.
http://www.ansys.com/-/media/Ansys/c...ilities180.pdf

then we can grasp mechanism "hmm..using this model..that method...great tool...want to buy " but pio has nothing like this.

needless to say,no doubt your effort,great achivements and user's feedback.lots of users seems correct.but this and that have no connection.
of course no one wants excessive information like above.but want minimum required.

so could you tell me your development policy?

"It's no use telling you LP,CFR...anyhow you know nothing.lots of users seems correct!published Science! great tool! buy it!"

as it stands I'm feeling like this.

It's different with solvers because unlike with things like fluid dynamics it's very easy to verify the solutions. All you need is a way to calculate EV of current strategies and EV of max exploit. From that you know how far away the solution is from exact equilibrium. It doesn't matter what kind of algorithm or "model" you use, it's enoogh that you verify the results. To be sure your results are correct you want:

1)The whole tree in memory (once it's not in memory you can't calculate anything about it, you can only hope)

2)Reliable way to calculate EV/MES - those are relatively straightforward and it's easy to test against other implementations or look at some spots manually.

I think you're carrying too much from your field of expertise where it's difficult to say what "true" solution is. In poker equilibrium solving it's easy. The whole tree is there for you and you can verify parts of it if you want (as many people did). Information if it's LP or CFR or something else (it's something else at the moment) has 0 significance as all that matters is that final results are close to equilibrium and verification of that happens independently of the solving method and is basedon EV and MES (EV of max exploit).

Btw, information what kind of solving algorithm is used is about useless as it's very easy to implement it incorrectly. It's very easy to make mistake in CFR so it doesn't converge. It's very easy to implement LP so it solves for something else not real equilibirum. It doesn't really matter because to verify things you look at the final solution and perform relatively straightforward calculations on it.

Hey guys please help with this problem..when i try to build a tree on piosolver basic and i click on build tree it tells me to connect to piosolver first which im not really sure how to do that...i went to my firewall and turned it off for a moment and still didnt work...

also this is another noob question but how can i input different cbet sizes on pio?

It should happen automatically if you follow the installation process described in the registratin email. Make sure all the files are in C:\PioSOLVER (or other folder if you have changed the default). It's important that you don't move anything from there.

It's explained in the quick start video. If you want more than 1 size you can just separate them using a a space or a comma. Like here:

https://gyazo.com/bf9fdbb193b156dbabaf8ff0cd31c290

Hello,
I got you pm, ty.
I replied today.
Let me know, ty !

Looking for an experienced piosolver expert to get me up to speed quickly. Will pay $150/hr. Send me a PM, references preferred.

Sorry not sure if this is the place to ask but, PIO seems to be running quite slow on my server in comparison to my laptop, could it be anything to do with the .avx engines? I get an access denied message when I select them on the server, but they work fine on the laptop.

i7 6700hq @ 2.6ghz 4 cores and 8 logical processors

vs

2* Xeon x5650 @ 2.67ghz 6 cores and 12 logical

Ive tested it by running two identical trees and the only difference I can see is in memory usage, on my laptop is says its using 5bg as opposed to 26 on the server.
There is a diff on accuracy setting .1 to .o5


Is this more likely a problem with windows, memory leakage or something along the lines of that?

Usual reasons for it are:

1)Energy saving settings (you need to turn them off, "balanced" is not good)
2)Some external antivirus which eats the resources

Try checking those and see if that helps.

The speed is more or less proportional to number of cores multipled by base frequency (and there is more or less 15% bonus for hyperthreading) so you get 2.6 * 4 = 10.4 here and:

6 * 2.67 * 2 ~= 32 here. This means that the server should be about 3x faster.

There is no way that Pio uses 26GB vs 5GB for the same tree. There must be some misunderstanding, try using task manager and look how much Pio itself is using. If the values are different make sure you are really running the same tree and if you are using the same version of Pio.

It's very likely you are doing something different on both computers so I would investigate that first. Pio uses the same amount of RAM for the same tree (mabye +/- several megabytes if you look at task manager).

The suspense is thrilling me ... most bets are on a PLO solver it seems. I still hope for some Snowie style Hold'em learning tool (bot), though. Whichever it will be, surely gonna be fantastic!!

What does the following error message mean?:

ERROR: go missing/incorrect tree

Any ideas what I'm doing wrong, or would you need script?

Thanks!

Yo it's past the first week of August, where is PLOsolver at dog

Quick question, I'm trying the scripting interface and the way set_strategy works is very confusing to me.
I'm trying to programmatically reduce raising ranges to match the more passive population. I want the freq of some combos to flow from raise -> call (not folds!).

Just to make sure there is no confusion, example:
It seems that piosolver by default uses 'fixed' option for set_strategy, except for fold. However I could not see a mention of this anywhere, not in the documentation and the dev-console doesn't show anything special either, it seems to be happening under the hood.

The problem even shows up when I LOCK THE FOLDING RANGE. I've tried locking the folding range before & after locking the raising ranges. It seems no matter what I do, it wants to put raises into folds.


Any solution/workaround would be very much appreciated

You have to internally in your program distribute strategy for all actions and then call set_strategy for all actions starting with the first one.

Example pseudo code:

for i in range(3):
solver.send("set_strategy r:0 %d %s" % (i, range[i]))

I'm looking at getting the edge version. How long does it take to solve a pre flop spot? It obviously depends on the tree, so lets say it's 100BB deep, IP (with user inputted opening range) vs BB, and a single bet size on each street. Is it less than a day? Or can it take multiple days? I'd be using cloud computing with 8-16 CPU's.

It depends on the tree and the CPU. For trees in 40-60GB range (normal preflop spots without many bet sizes preflop and sizeable preflop subset) it should be around 6-8hours on 16core CPU.

And is the time taken proportional to the number of CPU's? For example, if a tree takes 8 hours on 16core CPU, will it take roughly 16 hours on a 8core CPU and 4 hours on a 32core?

Read post 2816