OK I'm just gonna post this here without commentary on the actual strategy. If anyone wants to discuss that and bounce ideas off each other about why it looks the way it does that would be cool. I didn't get a chance to change V's range to be slightly narrower but w/e, this is close. Also I just noticed it's 12.5% antes, I assumed 10% but that wouldn't make a huge difference (if anything it would incentivize us to defend more I believe):
Hero's strategy vs the open
V's response to jam
There are a few things I wanted to comment on though:
1. Does anyone know how ICMIZER actually computes its supposed Nash equilibria? I ask because, if we could solve preflop spots with that program, there would be no need for PIO to have a preflop solver.
The way PIO solves preflop spots is by using flop subsets to try and approximate the full game, and you can get pretty close even with just 40-50 subsets (I used 72 for this tree which I'm told in the PIO thread in the Internet Poker subforum is an extremely good approximation. I have 256gigs of RAM at my disposal so I could've done more but I had other hands queued up that I wanna study).
I just don't see how ICMIZER can come close to this process. This tree I ran took up 80 gigs of RAM and was run since I first posted in this thread. Yet I'm supposed to believe ICMIZER can solve the same tree (or an approximation of it) in mere seconds and without the RAM constraints? I don't buy it.
FWIW I'm also gonna run a tree where hero can only defend, jam, or fold and even that tree is over 50 gigs.
I'm pretty sure PIO implements a counterfactual regret minimization algorithm, and I highly doubt ICMIZER does the same. Any comp sci masters here who can chime in?
I highly doubt ICMIZER is really giving us a true Nash Equilibrium.
I do want to note that a jam-or-fold tree (no option for hero to defend) took mere seconds to run, though the results are much different (PIO actually wants us to go even wider!!). But this is pretty useless when hero has 25bb.
Hero jam or fold
So my only guess is ICMIZER is spitting out some bastardization of a Nash equilibrium jam-or-fold tree solution
2. AZ you have way better results than I do--pretty enviable results actually--so I actually CAN'T flame you but my thoughts on this:
a.
Quote:
Will I get roasted here for admitting I don't have any NAI 3b sizings in spots like this?
and this
b.
Quote:
but ~20bb deep...hard to have a huge edge, isn't it?
are:
a.
Yeah you almost surely should have 3b NAI sizings in this spot, and my rationale would be it's part of an equilibrium strategy
b.
Obv we can't have as much an edge w/ 23.5bb eff compared to 100bb eff but I personally think we should try to eek out as much postflop play as possible if we're confident in our game and our ability to exploit holes in our V's game I think that even 23.5bb eff V's are capable of making massive postflop mistakes such that we can exploit heavily, and I think this becomes even more evident when you study NEs and realize that having an option of max{1bb, 10% pot} in your strategy profile can accumulate significant frequency at equilibrium--so it's not like we're doomed to 1-2 streets at this stack depth, far from it. In fact I personally think these shallowish spots are more interesting than deep spots because of how much closer to the edge losing one of these pots brings us.