Trying to get more accurate solutions from PIOsolver in HU
I have been trying to learn HU recently, and I have been trying to create a GTO library with scripts and whatnot. For context I am running typical heads up preflop ranges for srp with 200bb stacks with 4 sizes for flop, 3 sizes for turn, and 3 sizes for river, and the exact flop that is getting stuck is 333 (has currently been running for 10+ hours with no end in sight). I have the accuracy set to 0.2% of the pot, but it seems that some boards are just not able to get to that level of accuracy, and looking at the log of the cmd that is running, it seems like its not even getting close to 0.5% of the pot. Could I have a setting incorrect, or could it just be a hardware cap? Any suggestions on how to build a gto library for HU as well would be appreciated.
2 Replies
I have been trying to learn HU recently, and I have been trying to create a GTO library with scripts and whatnot. For context I am running typical heads up preflop ranges for srp with 200bb stacks with 4 sizes for flop, 3 sizes for turn, and 3 sizes for river, and the exact flop that is getting stuck is 333 (has currently been running for 10+ hours with no end in sight). I ha
Running BvBB 100bb with 3 sizes for flop, turn, and river with a custom built PC with 64gb RAM and a 12core/24thread 4.8GHz I was probably averaging ~20min per flop for an agg report.
Not sure what kind of machine you’re using, but if it’s not very very powerful, the settings you’re using are likely beyond its practical capcity
Paired boards at 200bb with that many sizes are brutal on solve time. I ran into the same wall building a 6-max library last year and the issue wasn't accuracy settings, it was that the tree size was just too enormous for the hardware to converge efficiently. 333 specifically is rough because the range interaction is weird and the solver keeps iterating without finding stable equilibrium quickly.What helped me was dropping to 2 flop sizes and 2 turn sizes for the library build, then only expanding to full sizes on boards I actually wanted to study deeply. You can always re-solve a specific board at higher resolution once you know it matters for your pool.
