Is it somehow possible to node lock on small (no rivers) save?

What is the best way to detect optimal preflop sizings?

Here is the config for 15bb spot https://pastebin.com/SW10y0tb
Here is the solution screenshot https://gyazo.com/acbd14ac82448e10cc7045edf2b76680

The tree does not appear to converge very well due to 2500 isolate size is never used (exploitability starts to jump after ~0.4bb/100, 0.35bb/100 is best what I got after 24h solving on 4.6s bench CPU).

So, what if I want to do some more experiments in detection optimal isolate sizing? Should I solve one by one and look at EV's or stick to the previous way and resolve with 3000 and 3500 for example? Or maybe some other way? Thanks in advance.

Open notepad and type it in?
I am likely not getting your question though.

In theory PioViewer checks for those and will notify you. Sadly as this is not instant it only does so every two weeks for now. We will likely improve the mechanism in the future though.

It isn't because node-locking changes EV across the whole tree which is impossible to calculate ones the rivers are gone. It's not a problem though because you can do the following:

1)Build the tree from scratch
2)node lock on this one
3)solve

The consensus for now is that the best way is to run a lot of simulations and develop intuition for it. Sadly there is no way to do it algorithmically other than brute-forcing all the sizes.

Well, to be honest we consider exploitability below 2bb/100 for preflop trees to be very good and below 1bb/100 to be good enough for all purposes. I could make it converge to 0 but then it would use more RAM whcih in case of the preflop solver is likely the most important thing to avoid.

Yes, change size, compare overall EV, rinse repeat. Most of the time the differences will be small although you should be able to detect differences between 3x and 4x 3bet for example.

is piosolver better on intel or amd threadripper, assuming equally fast computers? what about old versions like 1.9.2.5, i'm oldschool and never upgraded

New AMD cores are about 5% slower than Intel ones for Pio. This means that 16 core ThreadRipper @ 3.4Ghz will be 5% slower than 16core Intel CPU (you need dual xeon for that) @ 3.4Ghz.

Same.

I am comparing piosolver strategy vs node lock strategy

starting pot 50
pio strategy ip 35 ev
node lock strategy ip 34 ev

the loss is 1 bb = 0.01 bb /100 or 3% of 50= 1.5bb , 0.015bb /100?

I am sorry but it looks like a statemnet with a question mark at the end. I am not sure what kind of comment you expect of me. If you want to verify your math a good start would be to post EV info, this thing:

for both trees as well as what 1bb is in the game you are analyzing.

Punter, or anyone who is knowledable about PCs and running PIO could you please help me with this problem.

Met a CPU guy at school and he is offering to sell me this PC for $950 (I am also paying an additional $150 to upgrade to 32gb RAM)

https://pcpartpicker.com/list/Vw6dXP

After talking to him he thinks that it should run PIO fine but I think he might be biased bc he wants to sell the PC at probably a decent profit.

https://pcpartpicker.com/list/4mrV7W

This is the build that I had in mind before meeting with him.


What do you guys think? Will the first one run PIO fine? How does it compare the 2nd one ( which is a little more expensive). I am mostly interested in comparing the GPUs, his processor is only 4 cores but the one from my build is 8.

I have turned off script after it finished 2X of 9X boards. Is it somehow possible to continue same script?

Buying used PC at anywhere close to a price of a new one is a very bad idea.

Your build will be much faster for Pio (by around 50%) and much more pleasant to use for other things because of the SSD. They are not close in value.

Yes, just re-run it by going to Tools -> load script. It will skip over already done trees automatically (as long as you haven't moved the saves from their original location).

Where we can find or buy convenient flop subsets for postflop studies, using for scripts etc?

Does anyone have it for share/sale?


So there are 22k different flops and around 2k different textures.
I believe the 487 flop subset PIO is offering is designed to work with wide ranges, BB defense etc.. a lot of flops are same textures.

I just renter this server: Dual E5-2660/ 192GB/ 960GB SSD. I am only using it for Pio. I am brand new to Pio and I made up a hand with 3 raise sizes, did not seem complicated but the tree took almost 3hrs to slove and the file size was 30GB. The entire time the CPU ranged between 35-50%. Not sure what I am doing wrong. Is there a standard hand I can input to run a benchmark speed test on this server? Any help is appreciated.

We ship them with the newest version. Look into preflop_subsets folder in your Pio folder.

There are 1755 strategically different flops. 22k number doesn't matter.

487 out of 1755 is 27.7% of all flops. Naturally a lot of textures are going to be repeated as there are no that many.

30GB tree is humongous. For comparison full limit tree with a cap of 4 (4 bets possible on every street) which is already way bigger than most practical NL trees is below 13GB.
As to saves, you can make small ones, see here:

https://www.youtube.com/watch?v=VhG_YsEvfwU (question number 6)

Pro version supports up to 16 threads. Notice that 50% CPU usage already uses all cores but doesn't use hyperthreading. You would get maybe 25-30% speed-up with the edge version on this computer. You can also run 2 trees at the same time if you have enough RAM to make it more efficient.

Yes, see here:
https://www.youtube.com/watch?v=5Ozjfnatwms (question number 2)

I get around 9 seconds (with edge version though) on my 3.4Ghz quad and around 3 seconds on my 16 core Xeon server.

I am new to the program and have a question. If I give the solver the flop , turn and specific river. It solves very fast but are the flop actions solved according to the expected turn and river ? Or is it still independent until I get to the next card?

Because when I solve it this way if I use multiple bet sizes on the turn or river and only place one bet size on flop when it solves the flop gets multiple bet sizes options when I only requested one .

Did you do any CPU comparasions?

For example i7-8700K vs my i7-4700HQ:
http://cpu.userbenchmark.com/Compare...0HQ/3937vs2728

overall doing 85% better. 61% in single core, 159% in multi core

Would in pio i7-8700K be ~85% better or maybe some other things also play a role?

Anyway to only solve for the flop?

http://cpu.userbenchmark.com/Compare...00K/3937vs3941

Or for example i7-8700k vs i5-8600k.
12 vs 6 threads!!
Effective speed difference ~7%, user 17%, multi core 44%.
What difference would probably be in pio? (I guess threads might make difference since you said pio use up to 16)

If you give it a flop, turn and river it only solves the river.
If you give it a flop an turn it solves the turn (which includes solution on all possible rivers).
If you give it a flop only then it solves the flop (which includes all possible turns and rivers).

Let's see:
1)i7-8700k has 6 cores @ 3.7Ghz which gives it 6 * 3.7 = 22.2 score
2)4700HQ has 4 cors @ 2.4Ghz which gives it 9.6 score

22.2 / 9.6 = 2.3125 whcih means 1) is going to be approximately x2.3125 faster. Approximately because those are CPUs from different generations although it's not worth that much as per core progress is very slow these days between generations.

You can solve a flop tree by giving the solver 3 cards and clicking go. It will solve all turns and rivers as well because those are necessary to have a flop solution, otherwise you couldn't possibly know EVs of actions on the flop.
I invite you to watch a quick start video, it explains fundamental functionality of the solver. It's here:
https://www.youtube.com/watch?v=JqGQoQKbCB8

Both CPUs have 6 physical cores. One supports 12 hardware threads and one only 6. This is called hyperthreading. Hyperthreading is worth around 15% speed-up in Pio case (this is not universal across all programs, it depends on the implementation). Hyperthreading is very useful if you want to run VMs (for example Windows on a Mac via Parallels so you can run Pio inside). It will also make computer run smoother if you need to run Pio and do other things at the same time and still retain decent performance.

Hi,

I am wondering.. how does PIO know who raised pre flop?

It has me checking TT OOP on an 8 high board 98% of the time.
Until now I haven't worried about who raised pre, but I am wondering is there something I can do to let PIO know?

Thanks

It doesn't.

Maybe IP range is too tight or OOP range is too loose or maybe it's just how it should be.
If there was a way to tell Pio who raised preflop it wouldn't make use of that information anyway as it doesn't influence optimal play in any way.

It seems like "Tree > Reset Strategies" in PioViewer doesn't work: "Plugin Tree:Reset strategies in the tree failed//Solver process has terminated".

Is there any way in script to reset the strategies for a loaded tree? I want to load a tree, use `force_line` and manually set the ranges for players leading up to that line, then resolve, effectively limiting things to a subtree. However calling `rebuild_tree` doesn't work since it's deprecated, and manually calling `force_line` or `add_line` on a solved tree followed by `build_tree` seems to crash the solver every time.

I could regenerate the entire tree from scratch, but that's a little annoying, complicated by the fact that `show_all_lines` shows them in UPI format, not the format you need for `add_line`. Basically all I want to do is load a file then regenerate a pristine, unsolved tree from it.

It should work as long as it's a full tree save.

As long as it's a full tree:

1)set_equal_strats
2)clear_cache*

If it's not a full tree:

1)rebuild_forgotten streets
2)set_equal_strats
3)clear_cache

In 1.10 version the saves contain the config so all the above is not needed. Just load the tree and then click "build tree" in treebuilding tab. You need commands above only if you lost the config and it's not included in the tree (pre 1.10 saves).

It won't work. Those are part of the config. You need to include them in the config. They are not commands for editing an existing tree.

Yes, that's the best way.

But you said you want more: first rebuild the tree and then edit it. In 1.10 it's not a problem because all saves have config included so you can load the tree, edit the config and build the tree. If you don't have a config anymore then it's easy to rebuild the whole tree and reset the lines but it's currently impossible to get obtain the config and edit it.

I see now--thanks for the response! I was not using a full tree save, and I didn't realize the lines were not themselves a property of the tree.

Is there a command to get the tree building config programmatically from a 1.10 save?

HI, I'm having problems with aggregation reports over multiple files. I have the newest update 1.10.16.
I'm only use one bet size on the flop. When i run all hands reports (1326 combos) it's fine, i get the percent to cbet on the flop with all hand combos. When i run the same report using only the 169 combos, then every hand is 50% and doesn't match the other file.

Yes, it's show_tree_info
In previous versions we always saved a .txt file with a config along with the maine .cfr save so you could get a config from there.

Thanks for reporting. We need to try to reproduce and investigate. I will let you know what's happening later (likely today or tomorrow morning).