Now that I got that to work

I am see some very small difference in the numbers shown in aggregation reports and what I see in PIO viewer? ie report might say fold 55.5% and viewer show fold 55.0%

I am pretty sure I am comparing apples to apples, and the trend or pattern is just that all the numbers in the report are just a little bit off from viewer. Is this normal? Or what am I missing?

thanks

Those differences are because of card removal. PioViewer shows simple weighted average (weighted by weights of hands in our range) without adjustment for card removal. Starting from 1.9 version you can change that, I talk about it here:

https://www.youtube.com/watch?v=zLGpcZavxeQ (16:10 - see the video description).

The reports show "how often it really happens" frequency. Very short example is this:

-our range: AK, QQ
-villain's range: AA
-board: A3322

Villain bets and we call with AK, fold QQ. Now:

1)if we only look at what we do with our hands, we get:
12 combos of AK call
6 combos of QQ fold

Overall we call 12 out of 18 times which is 66%

2)if we look at what really happens:
There are only 12 matchups of AA vs AK (one A is on the board, we have one of the three remaining and villain has 2 remaining. There are 6 * 3 = 18 combos of AA vs QQ (we can have any of 6 combos and he has 3 possible sets). Overall we call

12 out of 30 matchups which is 40%

Notice that both numbers are correct they just describe a different thing. You can play around with that setting on some toy examples to see the difference.

cool - thanks

I'm thinking of building a custom desktop computer mainly for doing Pio calcs. Unfortunately, I'm not the most savvy when it comes to hardware. Could you tell me what things will have a significant impact on calculation and what won't? How important is the CPU? I know memory is important (does calculation speed scale linearly with RAM?). There must be other things I'm not thinking of...

Search this thread (set 2p2 to display 100 posts per page and then ctrl+f) for the following terms: cpu, server, performance, hardware. It's impossible to suggest anything without knowing your budget and expectations (if you plan to run just postflop trees or preflop as well).

As most of us are limited to 32GB RAM (which give about 45 flops for solving BB defense spots) what do you think about this concept:

I start with 45 flops and BB 100% range to get some solution, then I cut hands from BB range which I am sure solver will fold about 100% of the time (f.e if in 45 flops solution solver fold 80% of time K6o and 100% of time K5o I cut from BB range K4o, K3o and K2o)
then I repeat this process with more flops and do it again and again until it stops help. Imagine I will end with 180 flops (4 x 45).

Will this end solution be something what I can get on 128GB machine (4x32GB) running on 180 flops?

(Of course I know average frequencies will be another - f.e if BB has only 25% range in last point and solver CCs 15% range and 3bets 5% range Solver will show 60% call and 20% 3bet but I mean about frequencies for particualar hands not average)

RAM usage is proportional to the size of the range containing hands which are at least in one range s oif your BTN stealing range is 50% you will never do better than 2x.
Overall it makes sense though but you are going to burn a lot of electricity. It might be cheaper to rent a 64GB server for like 50$/month and just letting it run.

Yes, I know it. I am talking of course about small opening ranges (UTG on 9-max, etc). Just want to know if end solution will be correct or I overlook something important.

P.S. If I do one open raise sizing for 15% open range then I know defend vs 12% will be tighter, so then is only one step in this case.

Are there any general guidelines as to what exploitability you should run your sims to? Is there any threshold level where solving further doesn't really make much difference?

It will be correct as long as you don't cut any hands which would like to start calling at some point. This is unlikely to happen if you leave some leeway in BB range (keep some hands which are good candidate for calling even though they still fold 100%).

This is matter of personal opinion and experience. My view was that 0.35% is good for most practical purposes and 0.25% is really good but after talking to a lot of people I now refrain from giving guidelines on this topic. Some people are really happy with way higher exploitabilities, some are more pedantic and want it as low as possible.

I've done some similar explorations for BB defense, first using 60 flops to identify the always-fold hands and then checking that against a reference calculation on 100 flops with tight convergence. The folding hands came out pretty much identical. So it seems a loose screening process to narrow down the ranges before you do the final demanding calculation is a good way to go.

If you want OOP to auto-check on the flop just leave OOP bet sizing blank, like here:
http://imgur.com/A5c9pkn

Hi there,

I am thinking about buying Piosolver, can you help me with these questions please?

Is there a possibility to see the optimal betsize on a flop (or any street), or pio just shows the GTO for all the betsizes typed in? How do I know if 1/3, 2/3, 1/2, pot, overbet has the highest EV?

Are there optimal ranges modelled for all situations (in raised pots and 3bet pots), according to positions and betsizes? Or I have to enter ranges manually?

Thanks

You need to try them out. Sadly it's not easy to find an algorithm to determine the best sizing, brute force or basically brute force is the best I can offer for now. Luckily EV differences between sizes are very small in most cases.

We don't ship optimal ranges. A lot of people do sell preflop solutons for various situations though, one of them is our friend at PioCloud, here:
http://piocloud.******.com/

Notice that the concept of "optimal range" is very shaky for everything outside heads-up poker.

Hello guys, I like to deactivate my license since a windows reinstall. When I type "deactivate" to the Solver Console, it says: " ERROR code 4:"

If anyone knows what is the problem, I appreciate to share it here.

GL

It means there is a problem with connection. The reason for it is that the licensing service we use is down right now. Please try again later today (the service is http://wyday.com/ so if it's up it will work). I am sorry for the inconvenience.

Can Nash distances be translated into a maximum % error bound for the strategy of any given combo? If you know of no rigorous way of doing this, do you have any way of estimating this that, in your experience, generally seems good?

I don't think it's possible in general. Differences in strategies on specific combos are not a good way to think about how different the whole strategy is. For example, when we consider a toy game like this:

1)board 32222, pot 100$, stacks 100$, only all-in bets
2)IP range: AA, 66,55,44
3)OOP range: KK

OOP autochecks and now IP bets. They may choose any of the 66-55-44 to bluff with and check the rest or they may choose to mix on every single one of those combos. You may get to perfect solution and you will still have huge differences on resulting strategy.
I think similar situations may happen across the whole tree. Even if not in that pure form, it may well be that playing a combo 100% or 0% results in miniscule EV difference (and therefore no significant difference in exploitability).

It would be more feasible to try to estimate how far away EV for a given combo is from what it would be in the equilibrium.

Quick question from a GTO solver newbie.

I've been playing around with the free QJ2 two-tone flop and find the software very interesting. However, just how does one put the recommended frequencies into practice? It's interesting that for instance AA should be checked/bet as the PFR in a 1:1 ratio, but in practice it's just not feasible to play every hand you can hold in a GTO ratio.

What I do like, is the general feel the solver gives about particular spots. For instance, on this flop out of position with a 12% RFI against a reasonable CO calling range, pioSolver only c-bets 40% of the time and vastly prefers a 1/3 PSB. This is very useful information, since I always cbet a larger amount when the flop is wet and I have a strong, uncapped range.

I understand how GTO solvers like pio help one to develop a better understanding of general GTO play - but how do you use it to help you decide how to play the hole cards you happen to hold?

It's difficult and not practical. It's probably better to focus on other things such as:

For example:

You can also analyze what kind of range is betting and what is the composition of it using Range Explorer. This way you can get a feel for how to construct ranges. Later you can try node-locking a bit different play from your opponent and see how the solution changes and how range composition changes as well. Focusing on a bigger picture as well as patterns (which blockers are the best to bluff with etc.) is in my view more productive than focusing on frequencies for specific combos.

After you played the hand you can look into the solution to see how it plays with your exact hand. To prepare though it makes more sense to focus on things mentioned above. There are many tools to pick up patterns: range explorer, hotness, aggregation reports. We add them to help with understanding how the optimal play looks like. We don't think people should try to memorize the exact strategy

Thanks for the quick reply.

One funny thing I noticed. When OOP is allowed to check or put in a 33% PSB, and IP is allowed a 67% PSB, PioSolver says IP should bet 42% of its range after a check.

But when I change it so OOP is only allowed to check, PioSolver wants IP to bet 53% of its range.

Shouldn't GTO play for IP be the same in both cases? A check is a check, no?

It's not that easy
Notice that OOP being forced to check means they will check a different range than they would having all the options (well, at least for spots where it makes sense to bet OOP like for 3bet pot or SB vs BB in 6max). This means that there might be adjustments to auto-checking which yield higher EV.
This is what you are seeing, the solver "exploits" forced check by betting less than it would normally need to ensure max possible EV.

Notice that a different strategy against a forced bet would seem very natural. I understand it's a bit more difficult to visualize against a forced check but that's the same principle at work.

It would maybe be possible to say more about the case if you shared the config. You can do that by clicking "copy to clipboard" button once the config is filled and then pasting it to a service like pastebin and linking the page.

Aah... so basically, IP "knows" that OOP is not allowed to lead out, and it adjusts its strategy with that in mind?

Yes, because one of the conditions for a Nash equilibrium is that all players know each others' strategies.

Equilibrium is a state where neither player can improve by changing their strategy. In 2 player game it means that equilibrium strategy guarantees the highest EV against all possible opponent's strategies (no matter what they chooses we get at least X).
By "all possible opponent' strategies" we mean strategies constructed within the limitation of the tree. The opponent may choose any actions in a tree but only those actions so basically the solver "knows" that only check is allowed or that either check or bet are allowed.

I prefer a simple definition: "equilibrium is a state in which neither player can improve their EV by changing their strategy" or in other words: "both strategies are already the best against each other".