Polk recommends following frequencies to BB Open vs 3x: call 40% and 3bet 21%.

I can look at these opening frequencies with Snowie even more later on, perhaps with how it tends to defend against 2.5x and minraise.

One interesting thing to note is that solvers tend to use 80% cbet frequency but Libratus and Snowie use more like 50%. Solver has to do more checking on the turn to balance that high flop frequency (downside to that strategy, imo). Although the good thing in that strategy is, that it denies opponents ability to see free turns so often.

Would be interesting to put these two playing styles to play against each other to see how it would go.

Those simulations have to see showdown or fold to mean anything. If you can solve preflop poker you can solve the whole game so preflop is definitely the hardest, not the easiest street.

River is the easiest street to solve via simulation or otherwise. Any unknown information adds a lot more work for the sim and preflop is where the least amount of information is known.

That is correct. Preflop is not solved yet.

Does not work like that with AI.

If you use AI you have to know what kind of hand ranges it is going to arrive to the river with. But you don't know that if you do not have the "solution" (or should I say approximation) to how it is going to play earlier streets.

When Libratus was learning poker, it started the learning process from preflop play. If it would have started it from the river, then it would have not lead to any kind of solution.

You are right in that if we plug in river ranges, we can solve the game (actual solution within those ranges) without trying to solve preflop first given that our assumptions about those ranges that players arrive to the river with holds true. However, if players play earlier streets differently and arrive to the river with different ranges than we are going to assume, then we do absolutely nothing with that solution.

I am certainly no expert but I thought I read sometime that they actually start at the river and work backwards to preflop.

I.e. they build a tree and start at the terminal nodes and work backwards. But again I am no expert. I thought it was a paper on cfr. I could be wrong though I could have been mistaken about the trees/nodes represented in the paper.


Right but Idk what all of this has to do with your argument about preflop being easier to solve? You literally have to play to the end of every hand for every possible scenario on every board.

But if you have a board on the turn you could quite quickly and easily solve the river just because the amount of possible outcomes is greatly reduced, even if you considered all possible available 2 card hands.

Edit: if I could use a chess analogy to me it would be like saying it would be easier to solve a chess game from the first move than it would be 4 moves from checkmate.

I just checked and it folds 43% vs a 3x open and it 3bets ~22%. There is nothing exploitable about that. Of course IRL if you're playing with weaker opponents you can get away with calling more. The 10% of hands bellow the folding region are so close to 0ev that it can easily be made up postflop against weak opposition.

I meant it is easier to "solve" for the AI. Just because it starts the play from preflop and not all hands end up to the river. So after millions of trials it has more simulations regarding preflop play. It is this trial and error - approach that I was talking about. I thought that I mentioned it pretty clearly in my message.

It is quite different story if you just take river ranges on top of your head, plug them into solver and solve one street of poker than if you have to simulate whole street of poker to arrive to the river with some kind of "solution".

Yea I think I just took your quote out of context is all. Thanks for the conversation.

Anybody else have problem accessing the "Preflop advice" window in snowie? I get error code 361, and "Range advice" still works fine.

[QUOTES=Jerman]



You are right in that if we plug in river ranges, we can solve the game (actual solution within those ranges) without trying to solve preflop first given that our assumptions about those ranges that players arrive to the river with holds true. However, if players play earlier streets differently and arrive to the river with different ranges than we are going to assume, then we do absolutely nothing with that solution.
-------
It is quite different story if you just take river ranges on top of your head, plug them into solver and solve one street of poker than if you have to simulate whole street of poker to arrive to the river with some kind of "solution".


[/QUOTES]

This reconciles what several posters were seemingly disagreeing with. I knew you knew but this part was not explained as clearly as it is in your later posts.

Thanks for all the work you put into your posts.

Yeah, I figured afterwards that I could have been even more clearer. English is not my native language, so that makes part of the difficulty, but I am more than happy to clarify my thoughts.

I don't have those following frequencies for some reason.

I have NL50 hand played at Stars and both players have 100bb. Opponent raises to 3x, and when I am looking Snowie's range advice, it is calling 15.14%, raising pot 26.02% and folding 58.83%.

The rake at HU 50NL must be brutal. :/

I have played little over 11k hands and it shows that I have paid 931.52$ in rake. Rake is approximately 0.08 per hand so isn't that 16bb/100 off from winrate immediately?

I guess it is like you said Arty, that rake just affects so much to those frequencies.

Just checked them on level NL500 and on that level Snowie is calling 28.24%, raising pot 23.19% and folding 48.57%, so defending more than 50%.

Seems like it is better to do the donating on higher stakes.

Checking "optimal" ranges at 50NL HU is useless because it is impossible to beat that rake against any sort of competent opponent. You end up at this strange equilibrium where both players overfold their BB because while it is a -ev play it's less -ev than calling and paying rake.

The only circumstance in which you should ever play this kind of game is against very weak opponents in which case "optimal" ranges go out the window.

+1 to the above. I remember a 2+2er that did some HU 50NL on Stars and he clearly had an edge on the field, but was paying 15bb/100 in rake, which made the game basically unbeatable. It certainly wasn't worthwhile to battle against other regs.

May I ask that did he quit poker, or did he just switch to playing other format?

Would be nice to know how the NL50 pool in general is doing. For like, what is the biggest winrate at the moment on that player pool.

He went back to 6-max, saying something like "I have more fun playing the wide ranges at HU, but the rake is ridiculous". I don't follow many PG&C threads any more, and I think everyone I follow is playing 6-max cash (especially Zoom). I was under the perhaps mistaken impression that heads up cash was basically dead. There might be some HU players that have PG&C threads though.

You know a lot and we are lucky you are here.

Is pokersnowie's support here. How long it take to have a mail respond from support.
I registered a trial account long time ago. And today i buy a 1 year sub in a different account. But when i login it pop up a warning " Unfortunately our system encountered ............. "
Anyone has the same problem, please help me !

I'm having the same problem. Emailed support.

With deep stacks in 2nd position, about how often does Snowie three bet rather than call when UTG opens for a typical raise? Six and nine handed.

I just downloaded the free PokerSnowie trial (so I'm a complete beginner with it). At first glance it appears to use it's own preflop ranges. Can you not implement your own preflop ranges vs villian? I'll continue to look through the program and google/youtube search.

It's playing against itself.You can try to play your own game/ranges against the bot and it will tell u what thinks about that.