Very interesting thanks.

Sure. But you do it for a river-subgame with ranges calculated from the tabled preflop-to-turn strategy. However, I assume (not sure if it is actually the case, though) that the bot is already solving this very subgame to determine his river strategy - and probably also with more bet-sizes than the other streets. Any imperfections aside from the remaining bet-size abstractions would then have to come from the estimated river-rages and your method would not reveal errors in those as they are already part of the sub-game definition - for the bot as well as its river-nemesis.

So basically you are pitting a GTO-strategy against its nemesis and this - by definition - should give an exploitability of zero.

ignatius

You are doing it for the whole tree obviously. From leaves to root.

No. You can calculate it for the whole tree going to one river at a time (depth first). The imperfections would come from not perfect preflop/flop/turn strategies and from not getting the rivers exactly (just replicate the way the bot is playing). It's easy to get rivers almost perfectly although I guess they are not doing it because it's hard to imagine some of the plays would occur no matter what the ranges were.
I mean, it really is simple I am amazed you can do a poker AI project and not measure this.
Tbh the conversation should go like that:
-"I have this Nash equilibrium estimation algorithm"
-"great, how good is it?". "I think it's pretty good! I am convinced it's great!"
-"What's the number though?"
-"Well we didn't measure but we are doing it for 4 years now!"
-F-, come back next year

Srsly.
Again, it makes 0 sense to optimize for something you can't measure. It's like trying to play high stakes poker not knowing if straight beats 3 of a kind.
This is bad but claiming anything about "pretty good estimation" while not measuring it is well... not really something you should say.

In a more than 2-handed game you can be colluded against, even if villains aren't cheating by revealing cards.

People can decide to play slightly too LAG when IP to hero, but slightly too tight when OOP. They might be hurting their own winrate, but at the expense of both themselves and hero. The ones who gain from this are the other players, and there's nothing hero can do about it, and there might be no strategy available to hero that doesn't lose.

Multiplayer nash equilibrium isn't unique. Simple example- 3 way rock paper scissors. Player 1 throws R,P, player 2 throws P,S, player 3 throws S,R all with 50% prob. This is a nash equilibrium. But player 1 can decide to f over player 2 or player 3 with his throw next round and so on. NE isn't that important when it doesn't even guarantee a breakeven strategy.

Also this thread is so cringe with the results oriented posts. Doug was a beast, and Jason was a failure of humanity a few days ago.

Doug was never playing 100% seriously even from the start and ppl were trashing the bot as terrible. Now they complain that Doug isn't taking this seriously, and Bjorn is the beast.

It's a high variance game, and Claudico will only get stronger.

how does this get no love?

To be fair though, if you tuned into the stream yesterday, you could see Doug visibly fatigued and uninterested in playing vs. the bot. He might not have been taking it 100% seriously from the beginning, but he was definitely a lot more serious and focused at the start of the match. I think this would happen to any high stakes player in this situation who is being significantly under-compensated for his time. Being in a casino for 2 weeks straight, playing for 10 hours a day + additional hours of study, living out of a hotel room and most of the compensation coming from just showing up.....I'm sure it was fun for the first 5-6 days, but I'll wager that they're all hating the thought of getting out of bed in the morning now and that the end can't come soon enough.

I would say it's mostly due to him being on a downswing vs the bot and it being no longer likely that he wins the most.

Also first couple of days he was really spewy and ran insanely hot whilst justifying it by saying he was testing the bot (I distinctly remember him 2x pot cbet with 33 on like 742hh and calling a check raise, turn 3h river 3 vs bot's 44).

These are some pretty wild claims. How are you able to define what optimal play in chess is in order for you to say that "programs are nowhere near optimal play"?? Are you a chess grand master?

It is a bold claim indeed, but from his post history he seems to know what he's talking about .

And it's not an absurd statement to me although I have played chess to a high level before so have more intuition about this.

Also idk how the end result optimal play in chess will be defined seeing as it's almost certainly a draw at gto. You say that computers have another 300 elo points to gain- but this is a pretty arbitrary way to define "distance to gto".

Yes

Current scores?

The link has only been posted a dozen times itt.

http://www.twitch.tv/claudico_extra

-737k at the moment. Bjorn Li up 459k alone.

Sick sweat on whether cheet will end up positive. Rooting for him.

Remove any challenge or mystery? So once it's solved any monkey will play an unexploitable game due to the fact an unexploitable strategy exists? So long as you don't play against bots, you'll be fine.

pretty sure most people have learned to just ignore kirby, but yea, i liked it

Smart, funny and vocal. Yeah, most people dislike those traits combined.

Of course not, every game can be reduced and solved with enough computing power.

can someone clarify the perfect/imperfect recall distinction? is imperfect recall the academic way of saying claudico doesn't have a full strategy stored and calculates its strategy on the fly?

It would kill the online version of any game it happened to, botting would be too hard to police effectively considering the effort which currently goes into designing strategies for a bot could go into anti-detection mechanisms instead.

Well, it's already been doing that as it is. There were fixed-limit bots playing as high as 100/200 ten years ago (not crushing, tho). I just meant when it becomes an epidemic, y'all will have to play live.

wait what?

I think it needs to be pointed out that every hand starts at exactly 200 bb deep for this challenge. In a real heads up match stack depth will often be long ways from XXX bb. It is an entire dimensional reduction.