But no chance to directly do it while (or before) solving, right?

When you include rake in preflop solver how exactly does it calculate?
On sites if hand ends preflop there is no rake. If you are all in preflop or there is postflop play there is.
Is it the same?

Not right now.

Yes. You can verify it by going to the end nodes (for example after all-in and call preflop) and looking at sum of EVs of both players. This way it's easy to verify how much rake was taken away.

I'm getting the "setting and locking strategy doesn't make sense on incomplete tree" error when I try to node lock an OOP flop action. I searched this thread and found a response to this question that suggested it's a result of a "small save" of the tree. But I'm still getting the error when I do a full save, so not sure what the issue is. Thanks!

Node locking is only possible when you have the full tree in memory, the reason for that is that recalculating new EVs as well as re-solving requires all the rivers to be there. This definitely shouldn't happen if you have a full save, are you sure the whole tree is still there (if you open task manager you will see how much the solver uses and if it's feasible that you have loaded a full save).

I'm running a Lenovo Laptop w/ a I5 processor and 8gb of ram. Will I see any noticeable speed improvments by upgrading to 16-20gb or ram, or is it more a function of processor power.

Performance when it comes to speed is proportional (more or less) to number of cores multiplied by base frequency (not turbo one). There is slight bonus (12%-15%) for hyprethreading and for newer generation.

By adding RAM you can build bigger tree but not make the solver run faster. You would need a faster CPU (it's likely that this i5 is dual core as most latop CPUs) so when building new hardware the most important thing is to get the fastest CPU possible. The choice is quite limited in laptops (you can get quad core i7 @ 2.8Ghz or something like that but not much more) in desktops it's much wider (and also you can get comparable speed way cheaper).

In the picture below what do the EV numbers mean? Eg, note that we're betting 200, and the number for the nuts (JTs) is 144. The pot is 850. What does the 144 mean?

Thanks

Looks like you've got ev/eq toggled.

right. i thought it meant show both ev and equity. ev divided by equity doesn't make sense. i'm sure I'm missing something obvious here, but I don't know what it is...

one more question, which was the deeper motivation for the above one: is there a way (via viewer or via scripting) to determine the EV of an action predicated on what your opponent does. eg:

in a single street context: the EV of raising, assuming your opponent folds immediately to the raise, versus your EV when they call. EDIT: I suppose this one is just the size of the pot (fold case) vs the EV of arriving at the node where they call, averaged across all the possible turn or river cards, ie, the average of all nodes with that action sequence and the possible turn cards? i'm guessing i'd need to do that calc myself via scripting?

in a multistreet context (this is more complicated): when you have the betting lead, your EV assuming your opponent fold at some point versus your EV when they call down. you can ignore the possibility that they raise, and assume the action trees in question are, say, "call (turn card) bet call (river card) bet call" OR one of those calls replaced with a fold.

http://imgur.com/a/ERYQA

as you can see in the image, there's some problem with the accuracy of solving a preflop tree.
I don't undestrand why it is not decreasing but it's going up and down (5.41- 10 - 6.6 - 7 etc)

It doesn't do always like this, just sometimes.
What do to? wainting until it reaches a good approximation?restart the calc from the beginning?

another question:

http://imgur.com/VOov1ls

this is a preflop tree where sb open and BB has a locked defence both vs steal and vs 4bet.
Why piosolver 4bet AJo (also ATo KJo and others) with a negative EV? could it be an error calc?

EV divided by EQ is called equity realization and we got a lot of requests to include it so we did. Personally I think it's a curiosity without much practical importance.


Yes, go to that action of the opponent (choose it in the tree browser above) and look at what EV you get in this spot. That's EV of your action assuming the opponent plays a given action. It may require rescaling (as now there is more money in the pot from your action and opponents action) but it's the number you are looking for.

This is not easy to get right now.

http://imgur.com/a/ERYQA

While this is not desirable there are certain trees which behave this way. Please email support with the config so we can re-create it and use it for future tests.
One common case is small preflop ranges with one action at the beginningof the tree which is chosen very rarely (I did some preflop adjustmetns to our usual algorithms which work great in general case but make some other cases significantly worse).

It looks like it's not solved to accuracy good enough. With node locking preflop it's in general desirable to lock first solve later as otherwise the preflop solver is going to assume it's already late iteration and might be unable to adjust to get to the equilibrium after you locked a node as its internal parameters are already set for small steps.

While errors in calculations are possible the likelyhood of them occuring in simple cases like those are close to 0 as the solver is tested against various tools on millions of hands. It's possible that the solver is bad at solving certain kind of trees though.
Again, the best way is to copy the whole config and send it to us. We can then use it as tests for future releases.

got it, thanks.



let me ask your opinion on the underlying motivating question. i am trying to find criteria to classify aggressive actions as bluffs vs value bets. except on the river, all "bluffs" are really "semi-bluffs" to one degree or another, so that complicates matters.

you could just go with a simple rule like "under 50% equity means it's a bluff". the problem with that rule is:

1. a bet/raise with 49% equity is probably not fundamentally different from one with 51% equity.
2. it doesn't capture situations in which you are raising/betting as a current equity dog, knowing that the the turn or river card will usually put you in a situation that gives you a strategic advantage because your range is polarized. thus factoring in the advantage you expect to have on later streets, the money may actually be going in good. i *think* this is possible, though i'm not sure.

i'd love to hear your thoughts on how to approach this problem.

Yes, this is what makes it complicated. Recognizing bluffs/value bets is relatively simple on the river (although even then there are plays which are neither in some spots) and is on the to-do already. For flop/turn it makes sense to introduce more advanced method for classification. We will think about it at some point but there is some poker judgement involved so it will never be black and white issue.

One method I was trying in the past is to compare immediate equity to equity vs say top 20% of opponents range. Hands which have good ratio are semi-bluffs, hands which have both terrible are pure bluffs etc. It's not a simple problem, it would be nice to have some idea what the desired output is (what kind of categories for example, only bluff/semi-bluff/value or more?).

Sorry not really following you on that due to my lack of computer knowledge, is it something in task manager having to do with CPU and/or memory being used by PioSolver that would indicate to you I don't have a full save?

Yes, you can also look at the size of the file itself. If it's a few GBs then it's likely a full save, if it's a few/tens mbs then it's a small save.

Why would your range being polarized give you a strategic advantage?

It's 130 MBs.

So what's the issue then? I've not saved it correctly or my computer doesn't enough disk space (I checked this and it seems ok, it has 40 GBs/220 of space available)?.

To clarify, the file size is 129 MBs and the tree size is about 2500 MBs.

Running a script on a relatively small (11GB) tree and the solver is taking forever to converge close to .3% accuracy on a few flops. Almost flops are taking 30min-1.5hrs but one or two have taken over 10hrs.
1) is there a likely explanation for the solver struggling to make improvements after so many iterations and
2) is there some command I can type to either finish and save the current flop as it is and move on to the next one or change the accuracy settings of the script without losing what has been done

Thanks and congrats on building such a great product. Glad you guys are having the success you deserve.

on the river, would you prefer to:
1. 100% of the time have a hand which you knew had a 50% chance of winning vs your opponent's range.
2. 50% of the time have the nuts, and 50% of the time have hair. from your opponent's perspective, in this case, you simply hold a hand that beats him 50% of the time.

I wonder if this difficulty means that we haven't found the right criteria, or simply that the concepts "bluff" and "semi-bluff" are ill-defined in multi-street games and shouldn't be pursued. maybe they're just a relic of old-school heuristics and aren't useful in thinking about GTO strategies?

11GB is gigantic tree taking into account memory optimizations for recent releases.
A tree with 2 sizes everywhere is less than 8GB for example (on a rainbow unpaired board).

A lot depends what you mean by "forver" what kind of hardware do you have and what exactly the tree is. Some things in a tree like a lot of overbets makes it way more difficult for the solver to get to good accuracy fast.

30 minutes isn't bad for a tree that big (but again depends what CPU you have available). Feel free to send us the one which takes so long. Again, maybe you are using a lot of overbets - those make things way more difficult for the solver and are rarely useful for the final solution (with the exception of the river). I would recommend using combination of accuracy and timeout in scripts to avoid unpleasant situations like that.

-overbets make it difficult, try not using "add-allin" checkbox too much
-the solver is optimized for memory use mainly as that is what most of our users want (build bigger tree within reasonable RAM limit), that makese convergence not as good as it would be if we were allowed to use say 1.5x or 2x memory

Not when the script is running but you can use combination of timeout and accuracy when generating the script, the tree is stopped and saved when one of the condition is fulfilled so something like .3% or 40 minutes would be a good idea.

Thanks
It's true though that the solver could be faster and it will be in the future

I don't know, I don't think much about human concept lately as I am not an active player anymore and it's the task better suited for coaches/experience players etc.
Bluff/value bet thing works quite well on the river (not all cases but still) but yeah, on the turn/flop those don't make as much sense, especially taking into account that optimal play contains a lot of betting for protection or at least it looks like betting for protection