Limit/HeadsUp poker (practically) solved...?

http://forumserver.twoplustwo.com/29/...ossip/heads-up-holdem-solved-1379043/

read FullyCompletely's posts.

Originally Posted by Kanu:
"Thanks for the very good posts FullyCompletely, nice to hear that things won't be solved for a while"

FullyCompletely:
"Thanks! There is the caveat, of course, that solving a game is a way harder goal than making a bot that can beat all human players. Chess programs are stronger than all human players but the game is nowhere near solved. The checkers program Chinook beat the human champions long before the game was finally solved in 2007. In heads-up limit, Polaris was holding its own against human pros in 2007 and 2008, the programs we had in 2010 were much stronger than it was, the ones we have now are again much stronger than that, and yet we're still a ways short of solving the game. The same goes for backgammon, scrabble, and a lot of other games."

cliffs:
- Bots are playing as close to nash equilibrium as it gets... they beat humans.
- No-limit heads up bots that beat humans either exist or will exist in ~2 years time.

I think the part about headsup FL poker being (relatively) close to being solved has been semi-common knowledge for a while. surprised about how close they're saying 100bb HUNL is to being solved though. kind of interesting and gross at the same time.

there's no way NLHE will be solved in 2 years. it's just too complex of a game

i play PLO mostly now anyway so who cares

On October 16 2013 04:19 TimDawg wrote: there's no way NLHE will be solved in 2 years. it's just too complex of a game

what kind of odds would you lay against the best bot beating an elite player (since there might not be a consensus on who is best) at 100bb HUNL within 2 years?

100bb NLHE HU wont probably be completely solved for quite a while, maybe more than a decade.

The size of the gamespace of the game is between 10^18 and 10^19 and last time I checked (like a year ago) the best bots were able to solve game simplifications the size of about 10^12 (info comes from university of alberta poker bots research group).

Improving it means either: way higher computation power (probably not gonna happen in 10 years) or more effective algorithms for various tasks or better methods for mapping the real game onto the simpified fully solvable one. Or the combination of all of it of course.

With that being said, I would guess top hu NLHE bots would beat most human players even now in a 100bb cap game.

On October 16 2013 05:59 Joe wrote: 100bb NLHE HU wont probably be completely solved for quite a while, maybe more than a decade. The size of the gamespace of the game is between 10^18 and 10^19 and last time I checked (like a year ago) the best bots were able to solve game simplifications the size of about 10^12 (info comes from university of alberta poker bots research group). Improving it means either: way higher computation power (probably not gonna happen in 10 years) or more effective algorithms for various tasks or better methods for mapping the real game onto the simpified fully solvable one. Or the combination of all of it of course. With that being said, I would guess top hu NLHE bots would beat most human players even now in a 100bb cap game.

You don't need to solve the game to beat all humans. It's not even close.

On October 16 2013 04:51 bigredhoss wrote: Show nested quote + On October 16 2013 04:19 TimDawg wrote: there's no way NLHE will be solved in 2 years. it's just too complex of a game what kind of odds would you lay against the best bot beating an elite player (since there might not be a consensus on who is best) at 100bb HUNL within 2 years?

idk you seem confident they will make a bot good enough to do it...you tell me what odds seem fair?

but i'd take WCGRider vs the best 100bb HUNL bot they can come up with in 2 years and be feeling pretty good about it right now lol

On October 16 2013 06:52 nlloser60 wrote: Show nested quote + On October 16 2013 05:59 Joe wrote: 100bb NLHE HU wont probably be completely solved for quite a while, maybe more than a decade. The size of the gamespace of the game is between 10^18 and 10^19 and last time I checked (like a year ago) the best bots were able to solve game simplifications the size of about 10^12 (info comes from university of alberta poker bots research group). Improving it means either: way higher computation power (probably not gonna happen in 10 years) or more effective algorithms for various tasks or better methods for mapping the real game onto the simpified fully solvable one. Or the combination of all of it of course. With that being said, I would guess top hu NLHE bots would beat most human players even now in a 100bb cap game. You don't need to solve the game to beat all humans. It's not even close.

Yea that was pretty much my point.

I would bet that bots will get destroyed by the top HUsng players in 25bb HU NLH for a long time still

On October 16 2013 04:51 bigredhoss wrote: Show nested quote + On October 16 2013 04:19 TimDawg wrote: there's no way NLHE will be solved in 2 years. it's just too complex of a game what kind of odds would you lay against the best bot beating an elite player (since there might not be a consensus on who is best) at 100bb HUNL within 2 years?

lol not even for 25bb

On October 16 2013 07:17 TimDawg wrote: Show nested quote + On October 16 2013 04:51 bigredhoss wrote:   On October 16 2013 04:19 TimDawg wrote: there's no way NLHE will be solved in 2 years. it's just too complex of a game what kind of odds would you lay against the best bot beating an elite player (since there might not be a consensus on who is best) at 100bb HUNL within 2 years? idk you seem confident they will make a bot good enough to do it...you tell me what odds seem fair? but i'd take WCGRider vs the best 100bb HUNL bot they can come up with in 2 years and be feeling pretty good about it right now lol

i'm not confident they'll make one by then at all, i really have no idea. but i don't think the chances are low enough to be insignificant. i'd take 10:1 for some amount for sure, i don't know if those are "fair odds" though, probably consider slightly worse offers.

On October 16 2013 05:59 Joe wrote: 100bb NLHE HU wont probably be completely solved for quite a while, maybe more than a decade. The size of the gamespace of the game is between 10^18 and 10^19 and last time I checked (like a year ago) the best bots were able to solve game simplifications the size of about 10^12 (info comes from university of alberta poker bots research group). Improving it means either: way higher computation power (probably not gonna happen in 10 years) or more effective algorithms for various tasks or better methods for mapping the real game onto the simpified fully solvable one. Or the combination of all of it of course. With that being said, I would guess top hu NLHE bots would beat most human players even now in a 100bb cap game.

If you simply the action you should be able to "practically solve" it. If you take NLHE and make it PLHE, then specify that you can only bet pot, raise pot, call, or fold, the solution might even be easier than FLHE. Then you can add complexity as you wish, allowing half pot bets, then 1/4 and 3/4 pot bets, etc..

Didnt Phil Galfond speak about NL holdem for 100bb or less beeing solved in a 25-50 video on Full Tilt like 4 years ago ?

("No value in this game", he said)

On October 16 2013 14:24 VanDerMeyde wrote: Didnt Phil Galfond speak about NL holdem for 100bb or less beeing solved in a 25-50 video on Full Tilt like 4 years ago ?

He's joking/exaggerating. Still its a slight peeve of mine when people claim a game "solved" when its obviously not.

On October 16 2013 08:18 player999 wrote: I would bet that bots will get destroyed by the top HUsng players in 25bb HU NLH for a long time still

I would definately bet against you. Lets hope somebody will arange something like that soon.

On October 16 2013 11:56 NMcNasty wrote: Show nested quote + On October 16 2013 05:59 Joe wrote: 100bb NLHE HU wont probably be completely solved for quite a while, maybe more than a decade. The size of the gamespace of the game is between 10^18 and 10^19 and last time I checked (like a year ago) the best bots were able to solve game simplifications the size of about 10^12 (info comes from university of alberta poker bots research group). Improving it means either: way higher computation power (probably not gonna happen in 10 years) or more effective algorithms for various tasks or better methods for mapping the real game onto the simpified fully solvable one. Or the combination of all of it of course. With that being said, I would guess top hu NLHE bots would beat most human players even now in a 100bb cap game. If you simply the action you should be able to "practically solve" it. If you take NLHE and make it PLHE, then specify that you can only bet pot, raise pot, call, or fold, the solution might even be easier than FLHE. Then you can add complexity as you wish, allowing half pot bets, then 1/4 and 3/4 pot bets, etc..

Yeah i always imagine the ultimate GTO bot varying his raising sizes preflop all the time, minbetting or overbetting all over the place, etc. I'd be very surprised if a true GTO bot would 1/2 pot 3/4 pot or pot everything, that's only cuz as you said they just use fixed betting patterns, but doesn't mean those are truly the optimal ones.

^ I would agree a GTO bot varies sizes in a real game, but I still think its useful, interesting, and computationally possible to solve (practically) for a toy game where you're only allowed a few different bet sizes. If limit bots can crush any human now I see no reasons why a bot couldn't crush any human in such a toy game right now.

On October 16 2013 11:56 NMcNasty wrote: Show nested quote + On October 16 2013 05:59 Joe wrote: 100bb NLHE HU wont probably be completely solved for quite a while, maybe more than a decade. The size of the gamespace of the game is between 10^18 and 10^19 and last time I checked (like a year ago) the best bots were able to solve game simplifications the size of about 10^12 (info comes from university of alberta poker bots research group). Improving it means either: way higher computation power (probably not gonna happen in 10 years) or more effective algorithms for various tasks or better methods for mapping the real game onto the simpified fully solvable one. Or the combination of all of it of course. With that being said, I would guess top hu NLHE bots would beat most human players even now in a 100bb cap game. If you simply the action you should be able to "practically solve" it. If you take NLHE and make it PLHE, then specify that you can only bet pot, raise pot, call, or fold, the solution might even be easier than FLHE. Then you can add complexity as you wish, allowing half pot bets, then 1/4 and 3/4 pot bets, etc..

Sure they use betting restrictions in the simplified games to get a solution. But its far from enough. Even FLHE is far from being completely solved. They use many other simplifications, like card grouping pf, on flop, etc.
And you cant just add complexity as you wish. Most of the time you will be restricted by max size of the memory you can use at once.

On October 16 2013 20:36 locoo wrote: Show nested quote + On October 16 2013 11:56 NMcNasty wrote:   On October 16 2013 05:59 Joe wrote: 100bb NLHE HU wont probably be completely solved for quite a while, maybe more than a decade. The size of the gamespace of the game is between 10^18 and 10^19 and last time I checked (like a year ago) the best bots were able to solve game simplifications the size of about 10^12 (info comes from university of alberta poker bots research group). Improving it means either: way higher computation power (probably not gonna happen in 10 years) or more effective algorithms for various tasks or better methods for mapping the real game onto the simpified fully solvable one. Or the combination of all of it of course. With that being said, I would guess top hu NLHE bots would beat most human players even now in a 100bb cap game. If you simply the action you should be able to "practically solve" it. If you take NLHE and make it PLHE, then specify that you can only bet pot, raise pot, call, or fold, the solution might even be easier than FLHE. Then you can add complexity as you wish, allowing half pot bets, then 1/4 and 3/4 pot bets, etc.. Yeah i always imagine the ultimate GTO bot varying his raising sizes preflop all the time, minbetting or overbetting all over the place, etc. I'd be very surprised if a true GTO bot would 1/2 pot 3/4 pot or pot everything, that's only cuz as you said they just use fixed betting patterns, but doesn't mean those are truly the optimal ones.

Well I am not sure about minbetting or overbetting all the time, I would say most of the time the bet sizes will actually be pretty similar to what top players use, but sure, there will be a lot of varying.

If I remember correctly, there were some partial GTO solutions for specific low stack situations in the Elky&co book Raisers Edge. It was quite interesting how some holecards only had like one pf raise size 100% of the time and some had like 4 or 5 different raise sizes with different percentages. Like KK in one 3b scenario (I dont remember exactly), was something like 3% shove, 70% raise very small, 15% raise normal (2.5x or so), 12% raise big (i am making up the numbers, but it was along those lines).

If these capable bots exist, why aren't they filling up poker rooms?