Yes, I am sorry we made a fix some time ago but wasn't able to ship it yet due to a small architecture change. I won't be able to put the fix out there during the weekend but I should be able to do that (at least on 1 on 1 basis) early next week.

OK, please let us know here how to proceed to fix this issue when the solution is ready.

ty

Hi,

1. Does this mean that when an action is borderline EV-wise, Piosolver uses a mixed strategy?

For example a slighly +EV call will be sometimes a call and sometimes a fold?
And a bet which is slightly better than a check EV-wise will be sometimes a bet and sometimes a check?

2. Can you explain how to calculate the threshold between "I do action A 100% of the time" and "I do action A x% of the time and action B the rest of the time"?

I mean is that in terms of comparison of action A's EV vs action B's EV : if the best option is only Y% better than B, than we can decide to start mixing?

3. I am not sure what I mean is understandable, so here is an example :

if EV / action A > 1.1 * EV / action B, then do A always, but if not, then mix.
(it's an example for question #2 with Y = 10%)
something like that?

1. Only mixes actions that have the same EV. EV(fold) is 0, so any hands that mix call and fold are 0EV calls. +EV calls will never fold (maybe some small amount of the time depending on the solvers algorithm and how close you are to equilibrium, but if you have exact equilibrium they wont)

2. I think first part explains it. If you are asking why certain hands take actions a specific % between 1-99 and how to estimate that yourself...you cant. There are a lot of factors involved that are way too complex for humans to understand imo.

3. Also explained by part 1. If EV(A) = EV(B) mix, else strictly use action with highest EV.

First let me make a general point about EV. There isn't such a thing as "EV of an action" in vacuum. It only makes sense to talk about EV of an action against a given strategy of an opponent. Hands which are +EV against one strategy of the opponent might be -EV against different strategy of the opponent. Even if we play perfect equilibrium we may have -EV actions against non-equilibrium strategy of an opponent. We know however that if both sides play the equilibrium only the best (EV wise) actions are chosen by both sides and mixing only takes place when 2 actions have the same EV.

This is why solving for equilibrium is difficult. We may have perfect balance for our actions at some point (say mixing 50%-50% for actions which are both +10$/hand) but once the opponent adjusts it may not be a perfect balance anymore. What PioSOLVER displays is EV against current approximation of the opponent's strategy.

Now about mixing: we know the algorithm converges to the equilibrium and the balance for actions is closer to the equilibirum over time. It also means that differences in EVs for actions where there is mixing is smaller over time. It isn't exactly 0 because we don't reach perfect equilibrium but we are getting closer over time.

It also means that it's not that easy as taking EV of one action and taking EV of another action and determining the balance. The EV of actions changes over time as opponent's strategy gets close to equilibrium as well. We don't know what EV of an action in perfect equilibrium would be unless we actually get there.

With this in mind let me answer your specific questions:

No, it doesn't mean that. If Pio knows one action is better than the other it will always choose it. The problem is it's hard to determine because the EV of actions change during solving.

No. It will be 100% call assuming it's always +EV call during solving.

No. What the solving algorithm does is beyond human understanding. While it's very simple in principle it operates on millions of nodes and tries to find the balance. It's not something any human could hope to follow on anything but the simplest toy example. The algorithm tries to adjust in direction of a better action at every point but as that changes once the opponent does their adjusting it's not a simple process.


I hope it's clear from the above that it's not how that works. Mixing only takes place if actions have exactly the same EV in perfect equilibrium.

I was wondering if you could give a bit of clarification as to what exactly the 'exploitable for' percentage value actually represents. My interpretation is that it represents the change in EV(as a proportion of initial pot size) for either player that can result from either changing their current strategy at a given calculation step. Is this correct?

If this is true, then is there a way of determining exactly what player can gain from a change in strategy? e.g. How much of the exploit can be attributed to each individual players change in strategy?

Hopefully that was clear, thanks!

If you play perfect exploitive strategy (assuming you know exactly what the opponent is doing) you win X per hand vs any strategy. That X against current equilibrium approximation is "exploitable for" number. It's in chips per hand and we translate it to % of the pot.

The solver produces EVs of max exploit (in short MES) in the info. You can look at it and see how far away it is from the EV of current strategy.

Am I right in thinking that the exploitability is equal to the sum of the negative EV lines?

Exploitability is the difference between max exploit and current approximation per hand.
The equation is:

((MES_OOP - EV_OOP) + (MES_IP - EV_IP)) / 2

We divide by two to get exploitability per hand (and not per pair of hands).

hey, id like to share a EDGE license, if anyone is interested message me

This is not welcome here.
Our licenses are personal. Sharing with close friends/roommates is ok as long as you accept that only one person is owner of the license and the other has no recourse in case of trouble nor any rights to the license. Sharing with strangers is a recipe to be scammed. Here are various scams which already happened:

1)Person A buys a license for person B, accepts the money then A requires a chargeback on the credit card. They usually get the money back (or the dispute takes 2 months or something) and the license is revoked.

2)Person A promises to share the license with persons B and C, takes the money, sends both B and C some random keys and takes the money themeselves

3)Person A promises to share with person B, gives them the key, accepts the money after the software is confirmed working. After that person A requires a reset of the license (as the owner they can do it) and uses it on their computers.

In all those cases person B is left with no money nor the license. We can't do anything about it other than warn: don't share licenses with strangers, don't buy Pio licenses second hand and do not ask others to buy it for you if they don't give your email as a registered one (having a registered email means you are the owner of the license). You will be scammed if you don't follow those guidelines.

Hi.

When the option Tool---> Configuration---> Behavior---> open Range Explorer in a new tab instead of new window is selected, the two ranges involved in the analisys are 100% both (think it's a bug); maybe it's already a known problem, hope it's usefull.

Hi,

I was wondering if it's possible to run a script that solves using rounded strategies? Right now I can run a script for a subset of flops, but then I would have to go into each file one by one to apply rounded strategies. I would like to apply 108 rounded strategies (if I'm using the 108 flop subset) all at once, either as part of the script or after the script has run.

Thank you.

To be honest I am not sure. Maybe the idea was that if you want to have equity calculator/range explorer independently to the tree you can get it that way. I will add it to the list to check

Solving using rounded strategies is not possible at the moment you can however construct a script which rounds after solving and only then saves the tree.
To do that use "Commands to execute on each tree before saving" in tree generation, like here:

https://gyazo.com/33fc9c3eabec09c8b6692649d0e202d7

To see how round_up_to command works exactly, refer to the docs, here:
https://cdn.shopify.com/s/files/1/07...68544248083387

Nevermind the answer crossed my mind while posting

from a bit back, do you still think it would win at around 3-4bb? since libratus won at 15bb, though hu (and probably no rake factored)

I am not really up to date with the games these days. Edges are bigger in HU so 15bb/100 isn't very surprising as human in general are very bad at poker.
The way to win the most would definitely be to adjust just a bit from GTO in direction of what population tendencies show but even pure GTO based strategies should do quite well.

hi punter,

If i do multiple analysis,
is there a way either through command line or something else to ask the question :
what is the frequency of a bet size on a certain street ?

for example if I input 50% of pot and 75% of pot on the turn ,
there are many different lines/node that reaches the turn and I would like to know accross all these possible lines/nodes, is 50% used more often than 75% ?

thank you in advance

If you mean a single tree then those are displayed in the right upper corner when you browse the tree, here:

https://gyazo.com/9dd003080e1da7e1c4d99b2ed69074df

If you mean across many trees then you need to run multiple files aggregation report and then look for overall line in there.

yes but let's say if I input 3 bet sizing on the flop (35%, 70%, 100%) and one raise size , and 3 bet sizes on the turn (35% , 70%, 100%)
then we can reach the turn through
r0:c:c
r0:bet35%:c
r0:bet70%:c
r0:c:bet35%:c
and so on and so on
may be there are a dozen way of reaching the turn

If I want to know on average if the bet size 70% is frequently used on the turn ( to remove it for example)

I would need to run aggregation report on all possible line that reach the turn and then compare the number with the frequency at which the specific line is reached.

That is very very time consuming.

Is there any way to get this number more easely ?

thank you for your answer btw

There is show_all_freqs command. You can read about it in docs here:
https://cdn.shopify.com/s/files/1/07...68544248083387

I go through one example in the video, here:
https://www.youtube.com/watch?v=zLGpcZavxeQ (14:20, see the video description).

The output of that should be easier to work with as it aggregates action frequency for all lines in the tree and prints them one by one.

Thank you ,

That is what i am looking for.
In the video though you load a preflop tree calculated on 100flops and then run the command.

Is it possible to do the same with postflop tree ( aka loading all the flop from a certain folder and then run the command)

Or do I have to execute the command on each tree through a script and then process the data ?

What does "create subtree configuration" do?

It's not possible to run it on all trees in the folder right now. You can either write a script to do it (you would need to insert all file names though) or program it in a tool which interacts with the solver. It's probably the easiest to generate a script and then parse the output.

It lets you take part of the tree (for example a branch after bet-call on the flop starting on the turn) and copy the configuration from there so the new tree starts from that point. This is useful if you would like to for example take turn ranges from a flop tree and build more detailed tree from there.

I go through one example on the video, here:
https://www.youtube.com/watch?v=1PEmsGaGLvk (3:00, see the video description).

When choosing one specific flop or turn for a preflop solution, the strategy shows up pretty slow. How could I speed this up?