Thank you
The free version just got updated to 1.4.1. There will be bigger announcement later today but you can already get it from here:
http://piosolver.com/collections/fro...piosolver-free

It got a speed boost, more options in Range Explorer and some other features about which you can read here:

http://piosolver.com/blogs/news/3539...er-1-4-is-here

There is no video for now.
I understand it may be frustrating but we've chosen to focus on development and as the software changes quickly maintaining a tutorial would be a huge burden and the videos would be out of date quickly (as it happened with some made for early versions).

It's likely to change in near future as the solver got all the major features we wanted already and at least the main functionality will be stable for a while so check from time to time what develops.

It seems our users pick up the interface quickly though and questions are usually answered fast on our Skype group (by either devs or the users). It takes some experimenting and playing around but I believe the learning curve is only steep at the beginning.

Hey, so I bought the basic one and gotta say well worth the money and somewhat easy to use though I'm not understanding 1 thing, so when I start a tree and start with the flop and have the OOP guy 3 barrel some of the hands it suggest to bet on the river it wont have in their range since it didn't bet on the turn. For an example the program bets a K turn with 40% of hands and checks almost all of their Ax hands and the river is an Ax now it suggest that I'll shove a lot of my Ax hands though since it checked them on the turn how does the program think it still have it in their range? Am I doing something wrong or not understanding something here?

This is a common question and you are not doing anything wrong.
It is useful to understand a principle on which the solver is based: at every iteration for every possible hand it looks at EVs of possible actions thinking: "Ok, betting here has EV of X and checking here has EV of Y, if Y > X then I adjust by checking more". It often happens that the difference is quite significant and the solver starts choosing one action 100% (or close) of the time.
However, it still needs to know what EV of action X would be in case the action was chosen. For example if the opponent suddenly started doing something really silly like calling 100% on the flop and folding every time on the river the solver would quickly adjust to betting the hand. To know the answer to a question "what is my expected payoff if I choose X" the solver needs to know what it would do on the subsequent streets.

Let's consider an example: it's a 3-bet pot with heads-up ranges. The flop is Js 5d 5h, OOP player makes a continuation bet and IP player folds T7s every time as it's a nothing hand. Now we are on the turn, OOP makes 2nd barrel and T7s is no longer in IPs range but the strategy for it still exists and tells us: "what I would do if I called the flop". On an example Kc turn it's still a 100% fold, like here:



It is however useful to see what our actual range is a this point in time. We can achieve it in 3 ways:

1)By enabling "squares size proportional to weight option:

http://i.imgur.com/DtV0VZ7.gif

2)By clicking "IP Range" button, like here:

http://i.imgur.com/p3dNuP1.gif

3)By going to Tools->Range explorer->Open in current spot and looking at our range there:

http://i.imgur.com/oiG0D23.gif
(this being a dev version, it looks a bit different than a current public one)

You may notice that, similarly to a strategy, EV as well as equity still exist even for hands outside the range. This is because it's possible to calculate them still by answering the question "how much this hand would win if it was still in range". As that value is independent of the hand's weight it is possible to calculate (we can just assume for a while it exists, see how much it would win and note the number).

I hope it clears up things a bit. The questions about how strategy relate to current range are common source of confusion. We made some tools to combine the concepts in one view but it's important to realize that they answer different questions. One answers "what I would do if I were here?" and another "how often I am actually here?" one.

I admit I was confused about it as well when programming the solver for the first time and our first version wasn't able to display values of EV and equity for hands no longer in range. It was however corrected I think in 1.3 version

sometimes pio don't running with an error whatever "run as admin" or not.



what is it?

Probably this:
http://forumserver.twoplustwo.com/sh...&postcount=299

what is atm the differences between the versions? And how about the (near) future?

The summary is here:
https://piosolver.myshopify.com/admin/pages/19718372

Cliffs:

Pro over basic:
1)Pro version is able to use more cores so it's faster for i7 quads and significantly faster for hexa cores
2)Pro version gives you 2 activations
3)Pro version is scriptable (so you can run many trees during the night etc.)

Edge over pro:
1)3 activations
2)priority support
3)compiles for modern CPUs which are about 15% faster on them as of today
4)uses all available cores (which is important for systems with 8+physical cores)

If you want something specific which we can make in reasonable time but which is unlikely to top the priority list you can maybe get that in edge version as well (talk to us before).

As to near future: nothing is decided yet but you can always upgrade for approximately the difference of the price.

Hi there, I was doing some work with other users and we think the All-in threshold should probably be related to the node effective stacks instead of a % of the starting stacks since that is the way we actually think while playing, for example if the threshold is 70% and we are OTT we don't think well I have invested 70% of my stack so far so I should move in here, we think I have $100 left if the bet I'll make is $70 I'll rather just jam since I'm not folding after that, hope that makes sense if many people is very used to the way it is now maybe adding the option to choose what whay you like it might do fine.

Also this is why I think is important to have the effective stack displayed at the beginning of each street or at each node if possible so we know exactly how much stack left we have at every decition point since that piece of information is vital in actual play.

So for example here http://clip2net.com/s/3l0UE9T after we bet lets say $56 the screen could show: 0 56 102 (158) SS:458 EF:402 instead of http://c2n.me/3l0UN4u.png

And since we often times open many new windows while exploring having all that info diplayed on the new windows would be very useful as well http://c2n.me/3l0UY4B.png

Again, ty very much!!

I considered it and I think it's better the way it is. The way it is means "a player is pot committed if they already invested given % of their initial stack". This is quite natural.
What you are suggesting makes sense though so maybe it will be added as an option if there is more interest.

I agree it's useful but the information is already there.
The pot is displayed as 3 numbers: OOP invested so far, IP invested so far, dead money from preflop and there is then a total in brackets.
You can subtract OOP/IP number from effective stack to get a stack at given point.
I see how it's not convenient to subtract all the time but I am just pointing out the info is there in case you've missed it.

Yeah, this goes on to-do and is likely to happen soon.

Does anyone offer coaching for this program or anyone just good with using it and calculations? msg me will obv pay

I get to the river and I show the required strategy and it shows hands in blue (fold) that should not even be in the range anymore cause we folded them previously... Any way to only show the hands that are in our range at the street? Or am i doing something wrong?

edit: they get taken out when you check square size prop. to weight. But otherwise not. which is odd.

You're not doing anything wrong. Please read the explanation here:
http://forumserver.twoplustwo.com/sh...&postcount=328

Cliffs are that strategies for hands no longer in range still exist and not showing them for hands with weight 0 but still showing them for hands with weight say 0.000001 makes little sense.
There is "square size proportional to weight" checkbox which shows them proportionally to current weight (so it doesn't show them if they are not in range, it shows them as 0.1 of a square if they are 10% of the time in the range at this point etc.).
Anyway, please read the linked post. It goes over several options to visualize the current range when browsing the tree.

maybe a cool feature to add.. Highlite the prefered move in the big square (not sure what its called..)..

RAISE 5
CALL 12.5
FOLD 82.5

Something like that.. Just less work for the eyes and more comfort for us..

Anymore feedback on it?
I personally like it but it seems the Skype group wasn't enthusiastic. It's not very easy to do (because reasons) but if there is interest it will stay somewhere in upper echelons of the to-do

Any preflop plans?

hey punter,

first, terrific work on this software. i've always wanted something like this, had thought about working on something similar, and wish i'd heard about it earlier. This is a much better approach than any of the competitors.

i know you've made a number of speed improvements since the post quoted above -- what kind of time would i be looking at now to solve LHE situations with 4b cap to completion, starting from the flop? Assume I was renting a 8core, 32G server on rackspace or similar.

Speaking of which, do you have recommendations for running the software on cloud servers? I have a Mac laptop with only 8G, and parallels kind of blows, so I don't think it would be ideal, especially since I'd want to be working on it as the sims were running.

thanks!

EDIT: btw, add another +1 for equity graphs/info. to me, the great thing about this software is it allows us to answer questions where our intuition and heuristics break down. but ultimately, we want to use what we learn from such software and incorporate it back into our human understanding. i think connecting results to the concept of equity is important for this.

Hey yep I noticed, just wanted to avoid making the rest :-P

Thank you very much!!

Something new, when browsing the strategies the solver always tells you exactly how much you are folding or calling, but with multiple sizings for betting and/or raising you get the individual % for each sizing, so every time I have to make the calculations to see exactly how much the raising range is, so if possible I'll really love to get the solver automatically give me that number, maybe something like this: http://clip2net.com/s/3l7GHBB

btw I know you get tons of user suggestions and just can't get them all done but I really appreciate you taking the time to reply to all of them and make the customers opinion feel important! cheers!!

There are plans but I am working with a constraint to make it feasible on normal (even if quite good) hardware not on this so it will take a while


It depends what you mean "to completion". It takes somewhere around 2.5h on average flop for me on a quad if we are solving to 1mb/hand using HU Cepheus ranges. That is a very good accuracy though and like half the time is spent going from 3mb to 1mb or something like that. I can test it for you if you give me desired accuracy and ranges, feel free to send me a PM.

We are on the fence on this one and we don't officially support it. I prefer talking about it via PM.

Yeah 8GB is a big limitation when it comes to limit trees. You can fit a lot of them under 16GB (not all though, full tree with flop donkbet allowed with HU ranges and on rainbow unpaired flop is above that). On the other hand cloud instances are in general very expensive comparing to buying your own hardware so it depends how much you want to run it and how long you are happy o wait for your solutions.

Already done in dev and first version will be available in 1.4.2

Yeah, on the to-do but unfortunately features like that won't be a priority for a long time as they don't add functionality and there are many urgent things which do. It has chances one day though

Hi there, I was thinking about the rounding feature and I understand why is so popular as people like simplifying things in a way where strategies can be memorized to a certain level, so I was thinking on maybe another type of rounding so instead of getting raw frequencies for each combo on every runout I could get a more condensed strategy for some runouts.

So when we are studying often times we are thinking on what to do on certain type of turns/runouts, for example what to do when the Flush hits, the board pairs, turned over cards, etc so maybe it makes sense to be able to group the turn cards and get an overall strategy to play the turns in the fixed grouping. I'll make an example with one flop from Tipton's book:

Flop: 8s7s3h I'll be interested on making some turns grouping like Ts: when the TS comes, F: when the flush comes ST8: for when a T,9,6 comes, A for any A, Over: for any over cards J-K, P: for the board pairing and then Any for any other card...etc

This way I can have some "rounded" strategies for the most important scenarios, maybe to be able to change from one sub group to the other using the vertical search, and users could make any number of sub groups from 49 (one group for each turn card-the default) to 2 (splitting into two different group of turns)

Maybe use this grid: http://clip2net.com/s/3ldOmw2 could be used and so you name the first goup, then select what is going on it, the cards then are grayed out from the grid and so untill all hands are selected into a group you can run the rounding feature.

I know something like this may be overly ambitious, I don't even know if is possible to implement such a thing or how to deal with Rivers (maybe having the same option), or if you guys would be interested but I just had this idea and thought it might be worth sharing

Cheers!

PD. for anyone reading this and thinking about getting the software, go for it!! Pio is revolutionizing the way to study the game!

Exploitable for: 0.012 = nash distance i assume

set_accuracy = sets target nash distance or % of the pot?

You are mentioning one of the two related features:

1)a way to show strategies by hand group (so what top pair is doing, what flush draw is doing etc.)
2)a way to show strategies by grouped board cards

I have to admit I am very used to seeing 1) but 2) is rare.

I am not sure what you want to round here though. Rounding makes sense in feature number 1 mentioned above (although I think the penalty for rounding all the hands in a group to the same strategy would be very steep). Can you elaborate what you mean here?

Making an interface for this is a bit of a pain and not likely to happen in the near future. I can offer you a work around though. When you make an aggregation report for given line, like this:



You cas choose rows corresponding to interesting turn cards (say the ones feeling a heart flush draw in my example) and then calculate % of RAISE/CALL/FOLD yourself.
To do that you need to take the percentages in columns I/J/K and calculate weighted average where the weight is number of match-ups (or Global %).
While this is painful to do by hand it should be possible to make a spreadsheet form to handle the calculations and then you would just need to choose and copy-paste some rows there.

It's not exactly the most convenient solution but it's one way to achieve what you want


Yes, the solver reports exploitability which is sometimes called Nash distance. The solver reports it in chips (so absolute values). in your example:

0.012 / 17 = 0.0007 of the pot = 0.07% of the pot.
The number is displayed as % of the pot at the bottom in tree-building tab as well so you can verify it by hand.

The solver accepts absolute values here as well. While this is inconsistent with some places in GUI (as well as recalc_accuracy which only make sense as % of the pot) it is sadly probably too late to change as there are a lot of scripts around which would be broken by the change.

In "generate script" window the GUI accepts % (as you can see from description) and translates it to absolute units for the solver.
If I were to code it today I would make it fraction of the pot everywhere but I don't think I can break it at this point.

Btw, more about what we mean by exploitability/nash distance here:
https://piosolver.myshopify.com/page...solution_trees

(Solution trees, bullet point number 3)

For Mac or Linux users...

As previously reported, PioSOLVER works in Windows running in Bootcamp or in a Parallels VM. Now it also works with a Windows guest running in a VirtualBox 5.0 VM; 5.0 was released earlier this month. VirtualBox is a free fully-featured VM.

[Thanks punter11235 for help in confirming that the problem under previous versions of VirtualBox was lack of support for Intel AVX/AVX2 instructions, which are now supported by VirtualBox 5.0.]

punter,

I finally got to play around with software today. I think it's terrific. It does so many things the way I'd have done them myself, which is a rare thing in software.

A couple questions, and some feature requests.

When the simulation is running, it periodically shows "exploitable for X% of the pot" at the bottom, and "explotable for 1.77" (or similar) in the right panel. Is the number in the right panel a dollar value number? Is there a way to see the exploitability in BB/100?

One of things I like a lot is the ability to click on "Raise" "Call" or "Fold", and see a filtered, normalized view in yellow of your range when taking those actions. Is there a way to also see the equity view of that filtered view? That is, I want to see not only the range I should be calling with, but how my equity for just that range looks. I'd also like to see to see the average equity when calling, raising, folding.

What I'd truly like to see is multiaction and even multistreet strategy. Currently, if I want to know the hands I'm intending to c/r, I have to look at my check range, remember it or open it in another windows, then click check and bet in the navigator, then see what my raise range looks like. Most poker players think something like: "I'll c/f with this portion of my range, c/c this, c/r this."

To reflect this, you could have the option to view not only "raise" "call" "fold" buttons, but also buttons for these multiaction strategies:

cf/ cc / cr
bf / bc / br

Taking this a step further into multistreet territory, I'd like to see things like my "check calldown" range:

For each hand, for each combo of turn / river hands, how often does that hand reach showdown with the history c/c, c/c, c/c? Then, considering only these hands and this history, what does my normalized range look like?

Similarly, I'd be interested in my range when I peel once or twice and then fold.

When looking at my "cc / cf" range, I'd like to see that side by side with a heat map of turn cards, the heat map showing the relative likelihood of taking that line -- hence showing which turn cards are bad.

Before I found PioSolver, I'd been playing around with some of these interface ideas so I could better explore Cepheus's strategy. Which brings me to my last question, I know these ideas are a lot of work, and really they are just different "views" of the full strategy PioSolver calculates. So I could just create my own frontend that "displays" the 7G .cfr files. Is it possible to get those files much smaller? For example, by losing some data and storing the frequencies at a smaller precision?

Thanks again for the awesome software!

What is shown on the right (EVs/MESes/exploitability) is all in the same units (chips) and what is shown on the bottom is that number translated to % of the pot.

No, although using 5/10 or 50/100 blinds makes mental calculations required not very taxing

Yes. Like this.

This is something I had in mind with my first solver prototypes back in 2013. This is a very nice feature

I think that once you go beyond one street those become problematic. There are some choices to be made about what exactly to display. For example knowing that QsJs is a cc/cf % of the time on Ts 6s 2c 2d isn't very interesting.

We already did some experiments with a heatmap thing and it is a nice idea although at this point it becomes difficult to execute without opening everything in a new window as the GUI is already quite packed. I like this idea but it's not something likely to happen soon.

Let me answer this point-by-point:

1)Parsing .cfr files is not a good idea. While the format is just a binary dump of the tree in depth first order it wouldn't do you much good even if I described how it's constructed; the problem is that the save contains only information about betting structure and strategies. There are no EVs/EQs or anything else the solver shows - all of those are calculated on the fly when browsing and not stored in either RAM nor the save; not only you would need to code the logic to calculate those yourself but you would also need to write an algorithm to turn our strategy format into something you can work with and as there are several thousands different sub-formats in every tree and isomorphic parts of the tree are not even there (the solver construct them on-the-fly as well to make the tree look like it's a full one from user perspective) - it's not as simple as reading the numbers and converting them to a Python dictionary or something.
You would need to re-create big part of the solver logic to be able to make use of the saves in the binary format.

2)The ideas like using less bits to store strategies and optimizing to store only what's needed can bring memory usage to about 25%-30% of the current one. This is great but it's not good enough for storage, there is much better idea out there:

3)You can make a small save (flop+turns or just flop). The small trees are still possible to browse and from user perspective there is very little difference - the solver calculates missing information on-the-fly in the background. It usually takes about 200ms-300ms to recalculate a river so even vertical browsing (changing river cards and refreshing strategy/ev/eq) goes smoothly.
Flop+turn saves are about 50x-150x smaller than the full one and it's 100x-150x in case of limit trees from what I remember. Saving flops only produces files up to several megabytes but they are not very practical as recalculating turns takes several seconds.

4)The point and the good news is that the solver is designed to be easy to work with for programmers. The solver exposes standard input/output text interface which you can easily use from any programming language. It's the same interface PioViewer uses (you can even read the whole log with communication to see exactly what is going on by going to Tools->Configuration->enable logging). Doing it this way you can work with small saves as well: just load them up and make inquires - you don't need to care about recalculating rivers - everything is taken care of in the background without any user/programmer intervention.

To start with the solver text interface go here: http://piosolver.com/pages/text-interface-scripting

There is also a documentation in pdf with all the solver commands described:
https://cdn.shopify.com/s/files/1/07...27913292418743

ability to load trees by drag&dropping pls