PMed you )

For the OS you said 0 was the highest then what is a lower number to use less threads? 1 or -1?

Feature request: bdfd in range explorer

0 is OS default. It's not the highest but you won't get speed increase after that.
The lowest is 1 (use only one thread).

Good idea, this is on the to-do.

Ok I think I've got it, so just to clarify then, this is my understanding of how I would model a heads up hand, a SB vs BB 6max hand, and a BTN vs BB 6max hand in Pio (assuming 5/10 blinds):

HU:
OOP money in pot: 10
IP money in pot: 5
Extra dead in pot: 0

6max (SB v BB)
OOP money in pot: 5
IP money in pot: 10
Extra dead in pot: 0

6max (BTN v BB)
OOP money in pot: 10
IP money in pot: 0
Extra dead in pot: 5

Is this correct?

Yes, this is correct. In case of BTN vs BB in 6max you want to start a tree with one possible choice of raising (to say 2.5bb or 3bb). In case of SBvsBB or HU you may choose to include a limp if you want. If there is any ante you just add it to extra dead money in the pot.

When using the preflop solver without flops, how does Pio compute the ranges?

Example: HU, in an open 2.5bb or fold scenario it will tell SB/BTN to open 60%, and BB to fold 0% (using a mix of 3b and flatting). I appreciate the power of PioSolver on computing postflop solutions. However, while I'm no HU specialist, calling the old 72o OOP vs a 2.5bb open seems loose to me.

It assumes it's check-down postflop, that's why solving those trees is instant.

It's a good play against such a small raise if the opponent can't do any betting postflop

Thanks Punter, that makes a lot of sense.

Hi punter, thanks for your answer.
How many subsets should you use for preflop solving with limited amount of ram and decent accuracy?

Would it be ok to reduce the range of the players, too?(e.g. for BB defend vs RFI instead of using a 100% range use like a 70% range and by that reduce the tree size?)

And for postflop you can use the 184 subset script and it will just solve longer than say 74 subset script?

Can anyone who upgraded for the preflop solver attest to how useful it is without buying extra cloud computing? My computer only has 16gb of ram and I'm wondering if its even worth it to upgrade if computations are going to take days to get done without shelling out a bunch for cloud.

nvm I got it

In general the more, the better. I think you want around 40 flops to have a good outline and around 60-70 to have reasonably precise solution.
As of right now we recommend using those subsets:
http://pastebin.com/AtyU84qC

out of all we ship.

Yes, the solutoin is going to be correct although you have to be careful with interpreting overall EV. As bottom 30% of hands are not in range anymore they are not included in the EV calculations. That will result in higher overall EV for BB than it normally be. Strategies and EVs for particular hands is going to be the same but overall EV will require adjustment.

Yes. It will take longer and saves will take more disk space (not very important if you choose small saves). Then the aggregation reports will take longer to generate but it shouldn't be very significant.

I very strongly recommend against upgrading to the edge version if you don't have access to hardware with a fast CPU and 64GB of RAM. You can make some simple things work with 32GB but it's going to be very limited. You can buy the solutions and read them with basic/pro version btw.

So just to be sure as this is the first time i hear about this "tweak": for preflop calcs, in heads up where we know BB will def not play certain hands we can exclude them (obs to make the tree smaller) and the strategies wont change for the remaining hands? I suppose we can we do the same for SB?

I mean instead of full BB range we can use something like this:
http://imgur.com/a/djqfN

Yes although it will only result in less space used if you reduce both ranges.

Only if you remove some of those hands from stealing range as well.
In HU it won't be much of an improvement (as BB defends a lot and SB steals a lot) but in 6max spots where stealing range is fixed to say something around 50% (or less for earlier positions) you can gain a lot by removing hands which you know are folds from BB range.

Btw, this is the reason we are stopping to respond to general solver questions sent by email and start re-diricting people here. You wouldn't see that answer if I answered the question by email and I wouldn't be able to link to it in the future. I realize it's more convenient to ask by email but it's just much less efficient. Email is going to be only for licensing/business/bug reports things and general support questions should go here.

Ok, thank you.

Yes, the solutoin is going to be correct although you have to be careful with interpreting overall EV. As bottom 30% of hands are not in range anymore they are not included in the EV calculations. That will result in higher overall EV for BB than it normally be. Strategies and EVs for particular hands is going to be the same but overall EV will require adjustment.

So it will affect only the EV of the overall strategy but the ranges should be the same right?
E.g. 3bet 10% , call 40% and BB fold 50% of the 100% range. This would translater in a 70% range something like 3bet 14% call 57% fold rest and only affect EV, because you don`t calculate -1bb/hand for the other 30% you would fold?

I am using a cloud right now with 120gb ram/16c which it is very good for preflop.
For preflop you should have 64gb of ram and an i7 with 4 or 6 cores, like punter said.
You can solve most spots with 64gb, for HU ranges of 100% vs 100% 64gb ram could limit you to 60-70 preflop subsets. It doesn`t really matter if you have more than 6 cores, it will just get much more expensive, especially when you want to build a pc with 2 cpus.

Atm I run one sim with ~120gb in around 50 minutes on the cloud (computing power: cloud 16coresx3ghz=48 vs Normal i7 with 4 coresx4.2ghz=16.8).
So on a desktop pc you would need like 1-2hours for 1 run and the 6-10 runs to get the exploitability down.
Still 2-3 sims per day for preflop, depending on the tree size.

It will affect overall EV, EV of hands of the other player at the beginning of the tree as well as after first (forced) bet as now they won't get any folds. Of course it only works if you force a bet with specific range as otherwise the stealing range is going to change once the player can't get any folds. This means the tricks is ideal for 6max spot where you specify stealing range anyway.

Wondering if someone can point me in the right direction. If I want to use cloud computing services to speed up the solving process, and that cloud service has, say 120gb of ram, but my computer only has 32gb of ram, are they compatible? Can you recommend where to but the Cloud services just for Piosolving?

Thanks.

Is it impossible to include a limp and 3x range from the BTN v BB if I estimate the population tendencies and use those ranges?

You connect to the dedicated server/cloud instance using remote desktop interface. The way it works is that you have the remote computer in a separate window. You can use 2TB, 64core server from 2GB laptop, it's not a problem.
The problems start when you try to move the solutions and open them on your local machine. You won't be able to open full big trees but if you make small saves that won't be a problem either.

Unfortunately not. I used to recommend one of the popular services for dedicated servers but they have run out of instances popular for Pio solving so as it is now it's better to ask on our Skype group as I have 0 experience with other options. I know people use ovh, Amazon, Azure, Google cloud though.

I am assuming you mean HU. Yes, it is possible but you need to node-lock the BTN to enforce your ranges are used and don't change during solving.

You should be able to safely run a laptop CPU at 100% with a laptop cooling mat. I've run my laptop cpu at 100% for days on end with a cooling mat.

Conversely I've overheated and broken more than a few desktop computers unless I've had office airconditioning running. The ambient temperature here regularly gets above 35 degrees Celsius. (In Queensland Australia it is close to 35 pretty much all summer). So for my high ambient temps a laptop with cooling mat is actually safer than a non-water-cooled desktop both at 100% cpu.

If we want to look at OOP vs IP and want to consider blinds just dead money can we set SB and BB at 0 and the actual blinds as dead money in the pot? I know we can let the program run like that but will it create issues in how PIOpre computes? looking to solve for call OOP vs 3bet, fold, or shove based on EV we have when it folds back to us and it now becomes HU. Example below.

https://gyazo.com/06c0ce960c2d91088e1e58794975070c

What is the recommended flop subset we should be using for PioPre to get accurate results but no overkill to save time for most trees (not referencing just this one)?

Yes, that's a good idea I agree with the rest of your post as well, mine was more of a: "if you don't know what you are doing it might be best way to stick to it" kind of advice.

Well, blinds are not dead money as they count towards subsequent bets so you need to be careful.

Yes but as it is now the same player is going to be OOP both preflop and postflop with this structure.

Other than inability to change positions preflop it will be fine.

So it looks like blinds of other players are included in dead money. In such case this is the correct way to do that. Notice that positions you consider are BTN vs SB for example then there should be 1bb as dead money (from bb) but sb shouldn't be included in dead money (as it counts towards the 3bet).

It's hard to say as there are a lot of people with a lot of different opinions about it. As it is now we recommend using one of those subsets:
http://pastebin.com/AtyU84qC

(the last column is average error in many metrics).

You can see that 60 is significantly better than 40 and then 80 is siginificantly better.
I personally feel that 74 gives nearly perfect solutions for practical purposes but less than that is needed for very good overview.
If you are just experimenting to see what sizing is better for example then it's likely a good idea to use less to speed-up the calculations.

Hi Piotrek

I got a hand simulation with an output from Pio I do not understand: IP 3bets with a very polarized range of 7.2% (some 76,65,54 and A5s-) and gets called with a tight range of 3% (some AQ, AJs, JTs, 77-99). Flop is Kh Td 2h. Equity otf is a split. Pio now wants to c-bet his entire range with a small besize. I do not get that: its like a toygame: Nuts n Air vs Bluffcatchers from oop. Why we don't split our range and choose a high betsize like 100% ?

Thanks a lot for bringing light into this one.

Stephan