A score is just a means to an end. An oracle which ranks candidate solutions is sufficient. But as I said before, bankrolls after a sample serve as an oracle. I wouldn't call it a function though.

They don't calculate an entire game tree.

A few notes.

1 - It is possible to compute the maximum exploitability of a given computer strategy for a game like hold'em, but it is not easy. In the case of hu limit hold'em it has been done (see Johansen et. al - they have code for this, but it is resource intensive). In fact, in my experience, it is more resource intensive than the code to actually generate the AI in the first place (although I haven't tried too hard here to be fair). The variable bet sizes with nlhe makes it much harder.

2 - Even if you can't easily compute the max exploitability of a given huhu nlhe agent, it is often possible to generate some lower bounds (that is to say, this AI is exploitable for _at least_ XX bb/100).

3 - Even if you have the maximum exploitability numbers, it isn't at all clear how useful they are other than as a very rough comparison tool. It has been shown in the literature that you can have an agent X with lower max exploitability than agent Y that nonetheless loses to agent Y.

So its possible to know what snowie has in every spot. So if you are a opponent and you play a hand against snowie, its very easy to optimize because you know snowies range 100% and what snowie will do. So now you just look for the best counter strategy.

This strategy will lead to perfect exploitation and consequently snowies response model must be quite well balanced. If you look at how snowie plays this is very apparent.


The QQ hand from above is very good example of how so many things snowie does seems unnatural and totally against what we have learned, however when you go thru it it starts to all make sense.

In our games QQ might not be a fold, however if an opponent were to try to exploit and good it make sense how qq is a fold.

Against snowie qq is definitely a fold.

Yeah, if Snowie only cbets bottom set on this board, as you mentioned before, then QQ is obv a fold. However I doubt that cbetting only bottom set is "close to GTO".

Blablablub. Can you prove that?

i think i may be able to prove that it is correct that he is doubting that cBetting only bottom set+ is GTO. not sure if anyone is interested in this though ;-)

Go ahead! Isn't it possible that he could be spreading FUD?

What's FUD? ****-You-Display?

That would be interesting to see. It makes perfect sense to me after thinking about what snowie is doing. But of course i can't prove it. So seing how someone is going to prove this would be nice. Till then I will trust the machine over the rambling 2+2er on here.

Also snowie never would have bottom set in this spot (as preflop raiser) and has many situations where it checks 100% of its range on the flop.

It's an acronym for "fear, uncertainty and doubt". It's usually applied when you think someone is stirring **** purely for the purpose of making people unsure of something.

Thanks.

Yeah, but assuming it would have raised 22 preflop, then it would have cbet only this, right?
Lol, matchball!

You're making the claim that since it's possible for snowie to be gto because it does something that gto does, it therefore is more likely to be gto. That's like the worst circular argument possible. Or maybe not so much circular but just dumb.

knircky misunderstanding mme.

zachvac misunderstanding knircky.

wat

hello, joshua.

we (@moreev) have been testing the beta version of 'challenge pokersnowie' and the least we can say is that the HU snowie is very solid, balances it's range very well and can make some ballsy moves (from hero call to overbet on river)

we came out really impressed with it

I think opposite, challenge proves that pokersnowie is weak. Just bet, overbet to contest the pot.

THIS, it really bluffs too much...took him a few buy in very easily

I played a hand where the BTN opened 2.5 SB 3 bet 8, I 4 bet to 17.5, the BTN folded the SB shoved 100bb effective, and I called.

Pokersnowie liked my cold 4 bet but thought I should have folded to the 5 bet. Says I lost 20.03bb in EV.

You need to make a min raise on the btn for it to give this advice. Snowie would not min raise, it likes to pot the btn. A 9bb 3 bet vs a btn min raise is a 1.56 pot 3 bet.

Lol I just finished a 2700 hands session of 6-max vs Pokersnowie Challenge. The snowie AI is very exploitable and FAR away from a GTO solution to poker. I won 1100 bb, with a 1550 bb positive redline. The chances of this being just variance is approximately 1% according to the Pokerdope variance calculator. The first couple of hundred hands Snowie did ok, but after making some adjustments I completely ran over the AI players.

I'm having trouble importing hands after playing against pokersnowie, could it be my firewall?

I've tried reinstalling it.

If your talking about importing hands into Pokersnowie it works better if they come direct from the poker site than from PT4 or HEM2.

Really ???
I just downloaded the free trial today and played 600 aggromonkey hands HU at something like 300bb/100.
Sample size is ridiculous but so is the winrate (and I'm not gonna play 100k play money hands vs a nitfish anyway). I almost never used the fold button (not preflop anyway) as there was no point in folding when you could bluff later streets with any overcards/flush/straight cards.

I guess if you calculate the variance there's still something like 20% chances that I should lose on the long run but I don't see how that could happen with that much fold equity.

100bb deep I even think going AIPF every hand is profitable.