I am sorry for late replies. I was away for 3 days and had very limited Internet access.
Let me answer the questions from recent days in chronological order:
Quote:
Idont have the option use node locking.
Is there some other option how to force pio cbet 100%??
I am not sure why you wouldn't be able use node-locking. Anyway, it's better to just build a tree without a check. Please refer to this post for instructions:
https://forumserver.twoplustwo.com/s...postcount=5469
Quote:
heey guys
that does means the number in the ring in the picture? Average cbet EV of my rane? Or? thanks
It's EV at this point in the tree.
Quote:
How are range values calculated? Are they simply weighted average of all hands within the range?
It's not simple weighting but weighting which takes card removal into account. EV of all hands is weighted by number of possible matchups (so how often that hand happens against opponent's range).
Quote:
does exist an option how to force PIO bet on turn 100%?
Yes, put:
Code:
check, bet, check, check
in remove lines field to make IP always cbet the turn. For OOP it would be:
The goal is to remove a check at the point you want a given player to always bet.
Quote:
I use PIO exclusively and haven't worked with other solvers--I hear mention occasionally that Monker can solve multiway spots. Given my rudimentary understanding of the situation, I feel somewhat confident that there's some "catch" here? What is it?
I'm assuming it involves either:
1. very low degree of accuracy with the solutions
2. high amount/magnitude of assumptions (so as to simplify the tree)
Can anyone shed some light on this?
It's better to ask Monker's author as I am not really following that project. My understanding from the time of its release is:
-it uses bucketing to make the tree smaller (gluing various but similar hands together hoping that common strategy for all of them doesn't influence the solution negatively too much)
-it uses an algorithm which doesn't have guarantees of converging to equilibrium (nor is final solution measured in any way) but which has some good properties (like guaranteeing that hands which take pure actions (for example bet 100%) converge to that after some finite time
Those are interesting to run but have limitations. We will likely add similar funcitonality in the future. We think we can improve over existing tools but sadly some things will stay (like inherent shakiness of multiway equilibrium).
Quote:
Error: "You have selected an option to 'Process only flops from subset', but the selected file doesn't exist."
I will forward you email to Kuba (Pioviewer dev). I haven't seen it yesterday when I was answering emails.
Quote:
Does the rounding feature only work on fully solved trees ? I have a bunch of sims were I saved the trees as small (no rivers). Do I have to run these again if I want to use the rounding feature ?
Yes because recalculating EVs/exploitability after changing strategies requires the whole tree. If you just want rounded strategies without EVs/exploitablity you can do the following:
1)open arbitrary solver command window (ctrl+b)
2)type:
rebuild_forgotten_streets
[hit enter]
3)round
Please be aware that EVs won't be correct anymore when you do that.
Quote:
Would it be possible to give a hint on how to query and receive info from the solver with an external tool? I was able to figure out how to communicate with the solver using UPI but I only see the results in the command prompt. How do I get the info to an external tool?
In general you need to open the process, read its standard output and write to its standard input. Libraries for those are available in every programming language. The term to google is: "how to interact with console applications in X language".
Send us an email and we will send you sample code in Python (and maybe in C# I am not sure if we still have it though).
Quote:
dumb question which i know has been asked a thousand times so i apologise:
https://gyazo.com/c651aff135146dc10fccc719b64e98c4
IP betting is split 44%check and 56%bet 1/3rd. When comparing the EV difference between check and 1/3rd bet theres little to no difference apart from some hands preferring bet over check (although negligible).
So does this mean i can just range bet 100% and it will be exact same profitability? No need to complicate it to mix strat and simplify it to one action on this board. Is there any complications to doing that?
If there is mixing you can ignore it and choose any action which is chosen by the solver with 100% frequency without losing EV. That is only true though if your opponent doesn't adjust. General advice is: follow the solver on pure actions (100% for one action) and use your reads on mixed ones to choose the one most likely to work against particular opponent.
Quote:
Hey Punter, you mentioned on the earlier pages that there is a difference between older and newer CPUs. Are you talking about the improvements of IPC of newer generations? If so, then I guess it makes sense that newer ones are definitely faster than older ones by some X%. Like for example Ryzen 3xxx series have roughly 15% IPC improvement over Ryzen 2xxx series. Which means that when compared side by side, the same speed and the same core count, the 3rd gen is faster by 15% on avg.
Saying this, I think it will be helpful for your future customers to have like a page in your website where you have a compilation of benchmarks from people or even better have PIO gather those bench results so it will be easier for customers to find a good server for their needs.
Old AMD CPUs are just very slow. Since Zen came out they are quite fast (previous gen was around 5% slower than Intel per core with the same frequency, the newer one is probably already faster). I am not sure if it's just IPC or other architecture improvements.
The problem with benchmark is that it's difficult to build a universal one. If we set the benchmark to something which is not solving a tree then if the solver changes the way it solves the bench will not be as reliable anymore. If bench is just some iterations of solving (as it is now) then once the algorithm changes older results are not that useful anymore. That means it's not so easy to collect reliable bench results. Additional difficulty comes from various settings (overclocked or not, enough memory dies to work with quad channel etc. etc.).
Quote:
Also Pio edge will always be at least 2.5x faster than Pio pro when using the same hardware right?
Definitely not. It only faster on very big hardware. It's marginally faster if you have up to 8 physical cores.