Quote:
How come when I run scripts, half the time the script will run, it will show it solving to accuracy, but it is not placing the solved boards into the "Saves" file? I notice this happens half the time. It just sits running the script for days on end, and no boards appear in the Saves file. Then I start a new script, and the boards will appear in the saves file
I don't know, it if shows the trees are solved and there are no errors it means it saves the files somewhere. Are you sure you are looking at the correct path. You can download fast file search app (Everything from voidtools) to look at *.cfr files on your computer (sort by date of modification to find the newest ones).
If that's not the problem then we will need some screenshot or maybe pm me on Discord (punter11235#1235) and we will sort it out.
Quote:
How is it that pio2 and pio1 running the exact same tree produce 2 different solutions?
They are not that different actually.
It's just that EVs against close to 0 range are not very reliable. If you tried to play your hands according to one or the other solution you would end up with almost identical result.
Quote:
Even if there is some magic blocker effect which yields 99% of the pot to a given combo and therefor it has the EV of the pot, how can it have 97% eq vs the betting range?
It's because the betting range (for this particular sizing) is close 0 and especially as vs Ad5h (as that removes most of the hands that bet).
Quote:
While writing this post I ran the same tree with pio2 but this time with the old algorithm:
https://gyazo.com/3690952a53dd7a518e6da7ba1bb99076
So it seems that all 3( pio1, pio2 new algo and pio2 with old ago) all produce 3 completely different results.
Pio2 old is the same algorithm as pio1 with some optimizations making it much faster (on average 2x).
You seem to think those solutions are completely different but I see they are about the same. Ad5x doesn't go to that line (supposedly because betting turn has higher EV). Then if it somehow ends up there and the opponent makes a very rare action of betting 75 it has huge EV against that.
That's the case on all the screenshots you have pasted. Probably the more precise the solution the less frequent bet75 without Ad is (or it disappears altogether).
I've run it with yet another solver (Jesolver) which uses a different algorithm (more similar to PIO_CFR than to the old one but still a different one) and got this:
https://gyazo.com/50afaa6539afffe60b5d04c6b7293109
Again, this about never happens and Ad5x has huge EV in comparison to other A5o combos.
All 3 solutions want to bet Ad5x on the turn. All have bet 75 on the river by OOP with very low frequency with a range containing mainly Ad and all have huge EV for IP with Ad5x vs that range.
In general solvers that don't converge to exact 0s that well (Especially our old alg but also CFR+ based solvers that use averaging won't have this effect that pronounced because the range isn't that precise (it has a lot of noise with almost but not quite zeroed combos). New Pio converges to exact 0s and thus this effect on EV is stronger.
It seems to me that you are focusing on local EV (that is EV at this exact point assuming you got there) instead of EV multiplied by the probability of that line happening. Yes, it's a significant EV difference between the solution at this point but this point almost never happens and thus the local EV is very suspectible to small changes in the opponent range (when one combo goes from 0.001 to 0.000 may influence EV a lot). This of course doesn't matter in grand scheme of things as all the solvers/algorithm wants to do the exact same thing with the hand in question and the only difference is how exactly the opponents range is composed and how much you would make with a hand that shouldn't be in range at that point anyway.
Maybe one example that makes it more clear is this game:
-you can either pass or pay me 100$ and then I give you some money back
Solution A:
if you give me 100$, I pay you back 90$
Solution B:
if you give me 100$, I pay you back 25$
Both are GTO (you should pass and if you don't I give you some money back pocketing the difference). I can't improve by giving you back less (or more) because my EV won't change (cause you are passing 100% anyway) and EV difference if you decide to give me 100$ is huge!
Last edited by punter11235; 03-22-2021 at 04:03 AM.