Two Plus Two Publishing LLC Two Plus Two Publishing LLC
 

Go Back   Two Plus Two Poker Forums > >

News, Views, and Gossip For poker news, views and gossip.

Reply
 
Thread Tools Display Modes
Old 05-03-2015, 04:54 PM   #1051
MiRee446
old hand
 
MiRee446's Avatar
 
Join Date: Jul 2008
Posts: 1,722
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by mack's View Post
a lot of the standard sizing and plays from these days would get you crucified 5 years ago
Like what?
MiRee446 is offline   Reply With Quote
Old 05-03-2015, 04:59 PM   #1052
Kirbynator
Carpal \'Tunnel
 
Kirbynator's Avatar
 
Join Date: Aug 2006
Posts: 52,005
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

did I miss something about the streams today?
Kirbynator is offline   Reply With Quote
Old 05-03-2015, 05:27 PM   #1053
LoneUltralisk
centurion
 
Join Date: Aug 2012
Posts: 165
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by Sam Ganzfried View Post
One year a team submitted claiming to beat up to 5/10 hunl online, and we beat them.
Definition of 'beat' here is almost certainly sitting out against decent regs and waiting for recreational players to sit them. That bar is not very high.

If msoft offered $1 million in prizemoney for the poker competition, we'd see pretty close to equilibrium play in HUNL.

Pre challenge, I thought you guys were at the forefront of poker AI. After the start of the competition when the bot was doing some clearly bad things, I assumed it was because Claudico was mainly focussed on techniques that were transferable across other domains and less focussed on winning the challenge. Now, I'm not sure.
LoneUltralisk is offline   Reply With Quote
Old 05-03-2015, 05:28 PM   #1054
LoneUltralisk
centurion
 
Join Date: Aug 2012
Posts: 165
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by Kirbynator View Post
did I miss something about the streams today?
day off today
LoneUltralisk is offline   Reply With Quote
Old 05-03-2015, 05:30 PM   #1055
punter11235
Carpal \'Tunnel
 
Join Date: Mar 2005
Location: solving poker
Posts: 8,206
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
. For HU limit Texas Hold'em the current strongest agent is from academia -- the University of Alberta just "essentially solved" it to exploitability of 1 mbb/hand and would beat anyone.
Well, I am not sure if they would beat anyone but yeah, their solution is very good and won't be easy to replicate.
On the other hand they calculated it for 70 days on 4800 cores... while it's impressive they got the distributed code right it doesn't sound very impressive considering the amount of resources were spent. I mean... there is not much to compare because normal people don't have access to supercomputers and the guys who won limit competitions so far (Cepheus didn't compete yet I think) did it having way smaller hardware and they claim they solved it without approximation as well (although I have no idea if they are right about it). If they did it's safe to assume they are ahead of the curve on this one.

Quote:
There are some limitations in academia
I understand it.There is value in developing more general approach which is used to poker than focusing just on poker. I am just saying that the fact that you are the best academic team doesn't mean you have much of automatic authority here.
It seems to me you guys made some design decisions which are just very counter-productive to developing a strong playing entity.
I am a bit jaded because there was a discussion about Tartanian before and I had the impression the author doesn't understand much about poker (although I am sure he understands a lot about math/programming in general) then there are claims that Claudico is some top AI or what not while it's clear to everyone it's blundering left and right and taking 30 seconds to recalculate river on 64 cores. It should be easy to make some claims about exploitability as well (at least for flop exploitability or w/e) and those are nowhere to be seen while the author is claiming people can learn a lot from the bot. Where you claim stuff that you have this amazing thing and people can learn a lot of from it then provide the number quantifying how good it is. It's not a rocket science to calculate it or at least approximate it well.
If it's 25bb/100 from NE then we know it's decent but far away from being close to the solution. If it's 50bb/100 we know it's just not good at all (even if it's stronger than other publicly available stuff).
punter11235 is offline   Reply With Quote
Old 05-03-2015, 06:13 PM   #1056
gregorio
Carpal \'Tunnel
 
Join Date: Jan 2007
Posts: 38,285
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

punter, how do you know that limping and 6x overbets on the river are not part of a GTO strategy? Have you solved HUNL already and not told anyone?
gregorio is offline   Reply With Quote
Old 05-03-2015, 06:16 PM   #1057
Herro Prease
centurion
 
Join Date: Oct 2012
Posts: 127
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by mack's View Post
a lot of the standard sizing and plays from these days would get you crucified 5 years ago
Yeeep. Which means the arguments above about "obvious" bad lines based strictly around current consensus hold little weight.
Herro Prease is offline   Reply With Quote
Old 05-03-2015, 06:21 PM   #1058
Sam Ganzfried
centurion
 
Join Date: Oct 2014
Posts: 187
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by punter11235 View Post
Well, I am not sure if they would beat anyone but yeah, their solution is very good and won't be easy to replicate.
On the other hand they calculated it for 70 days on 4800 cores... while it's impressive they got the distributed code right it doesn't sound very impressive considering the amount of resources were spent. I mean... there is not much to compare because normal people don't have access to supercomputers and the guys who won limit competitions so far (Cepheus didn't compete yet I think) did it having way smaller hardware and they claim they solved it without approximation as well (although I have no idea if they are right about it). If they did it's safe to assume they are ahead of the curve on this one.



I understand it.There is value in developing more general approach which is used to poker than focusing just on poker. I am just saying that the fact that you are the best academic team doesn't mean you have much of automatic authority here.
It seems to me you guys made some design decisions which are just very counter-productive to developing a strong playing entity.
I am a bit jaded because there was a discussion about Tartanian before and I had the impression the author doesn't understand much about poker (although I am sure he understands a lot about math/programming in general) then there are claims that Claudico is some top AI or what not while it's clear to everyone it's blundering left and right and taking 30 seconds to recalculate river on 64 cores. It should be easy to make some claims about exploitability as well (at least for flop exploitability or w/e) and those are nowhere to be seen while the author is claiming people can learn a lot from the bot. Where you claim stuff that you have this amazing thing and people can learn a lot of from it then provide the number quantifying how good it is. It's not a rocket science to calculate it or at least approximate it well.
If it's 25bb/100 from NE then we know it's decent but far away from being close to the solution. If it's 50bb/100 we know it's just not good at all (even if it's stronger than other publicly available stuff).
If you are going to attribute quotes to people, please be clear whom you are quoting. I'm not sure whom "the author" refers to.
Sam Ganzfried is offline   Reply With Quote
Old 05-03-2015, 06:42 PM   #1059
Frankie Fuzz
grinder
 
Join Date: Aug 2013
Posts: 534
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Punter,

The creators of Cepheus said previously that they initially used heuristics based on human conceptions of proper play (like never check the nuts or never fold x flush draw) but that these rules simply got in the way and made their strategy worse off so they ditched them.

Also, it's not at all clear to me that limping would be more worthless in no limit than in limit, like you claim. Can you elaborate on that? In limit we are never folding to a 3bet preflop while in NL we (probably) are.
Frankie Fuzz is offline   Reply With Quote
Old 05-03-2015, 07:10 PM   #1060
14756897412369
centurion
 
14756897412369's Avatar
 
Join Date: Jan 2015
Posts: 102
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by kingofcool View Post
I guess this is hoping for a miracle, but will the hands from this challenge be available to the public like when Hawrilenko played Polaris?
If anyone is willing to pay enough money I'd be willing to go through the vods and create hand histories for any/all of the players. It would have to be a pretty decent sum though .
14756897412369 is offline   Reply With Quote
Old 05-03-2015, 07:33 PM   #1061
tultfill
veteran
 
tultfill's Avatar
 
Join Date: Jan 2009
Location: internet
Posts: 2,928
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by punter11235 View Post
(at least for flop exploitability or w/e).
It raisefolded KdTs OTT on a T 7 4 2 board with 3 diamonds. It raisefolded to 80% stack.
tultfill is offline   Reply With Quote
Old 05-03-2015, 07:33 PM   #1062
e i pi
veteran
 
e i pi's Avatar
 
Join Date: Feb 2010
Location: ☃
Posts: 2,230
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

I'm curious what standard deviation Cepheus has when playing with itself.
e i pi is offline   Reply With Quote
Old 05-03-2015, 08:06 PM   #1063
Sam Ganzfried
centurion
 
Join Date: Oct 2014
Posts: 187
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by tultfill View Post
It raisefolded KdTs OTT on a T 7 4 2 board with 3 diamonds. It raisefolded to 80% stack.
Who is better -- someone who loses 11 BB/100 vs. the top players in the world but occasionally makes random horrible plays? Or someone who loses at 30 BB/100 but never makes any plays that stand out as being clearly terrible on their own?

Last edited by Sam Ganzfried; 05-03-2015 at 08:20 PM.
Sam Ganzfried is offline   Reply With Quote
Old 05-03-2015, 08:16 PM   #1064
ambientsilence
journeyman
 
Join Date: Oct 2014
Posts: 391
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by punter11235 View Post
Well, I am not sure if they would beat anyone but yeah, their solution is very good and won't be easy to replicate.
On the other hand they calculated it for 70 days on 4800 cores...
You're making it sound like it's a huge number but you don't have to buy a supercomputer for that. You can rent computer power through amazon or google. I don't know exactly how much it would cost but I'm guessing something on the order of $250.000. A good chunk of money, but it might be worth it if high stakes limit holdem was still alive.
ambientsilence is offline   Reply With Quote
Old 05-03-2015, 08:25 PM   #1065
g-p
Carpal \'Tunnel
 
g-p's Avatar
 
Join Date: Apr 2005
Location: 11
Posts: 6,705
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

sam i'd love a few sentences on why you think a NE might exist for nlhe. formal explanations are fine. thanks
g-p is offline   Reply With Quote
Old 05-03-2015, 08:35 PM   #1066
Sam Ganzfried
centurion
 
Join Date: Oct 2014
Posts: 187
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by g-p View Post
sam i'd love a few sentences on why you think a NE might exist for nlhe. formal explanations are fine. thanks
Well, in a nutshell, every game has a Nash equilibrium, and NLHE is a game. I imagine this explanation is not very helpful for you though

http://en.wikipedia.org/wiki/Nash_eq...stence_Theorem

"Nash proved that if we allow mixed strategies, then every game with a finite number of players in which each player can choose from finitely many pure strategies has at least one Nash equilibrium."
Sam Ganzfried is offline   Reply With Quote
Old 05-03-2015, 08:37 PM   #1067
feedmykids2
Pooh-Bah
 
Join Date: May 2012
Posts: 4,527
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by g-p View Post
sam i'd love a few sentences on why you think a NE might exist for nlhe. formal explanations are fine. thanks
It exists by definition bro, unless you think Nash was wrong lol
feedmykids2 is offline   Reply With Quote
Old 05-03-2015, 08:40 PM   #1068
g-p
Carpal \'Tunnel
 
g-p's Avatar
 
Join Date: Apr 2005
Location: 11
Posts: 6,705
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

all games do not have a nash equilibrium


nash can be right and a NE might not exist for nlhe
g-p is offline   Reply With Quote
Old 05-03-2015, 08:44 PM   #1069
feedmykids2
Pooh-Bah
 
Join Date: May 2012
Posts: 4,527
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Still waiting for you and durrrr to make billions and win a nobel prize with this theory you have.
feedmykids2 is offline   Reply With Quote
Old 05-03-2015, 08:50 PM   #1070
punter11235
Carpal \'Tunnel
 
Join Date: Mar 2005
Location: solving poker
Posts: 8,206
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
If you are going to attribute quotes to people, please be clear whom you are quoting. I'm not sure whom "the author" refers to.
For example this article:

http://www.cardplayer.com/poker-news...apon-for-poker

Quote:
When asked if he thought, in the future, if learning from what his program has discovered would be mandatory for all those wanting to compete at the top of the game, Sandholm didn’t mince words.

“I think so. It’s a bit of a nuclear weapon for poker. You don’t want to be bringing a knife to a gun fight.”
Is kinda a big claim when you have something exploitable for probably 40bb+/100 and playing in a way that its untypical things it does is major source of profit for the opponent.

Which also brings us to:

Quote:
Having found what he considers to at least be an approximation of the Nash equilibrium for heads-up no-limit hold’em, Sandholm entered the same bot into both categories.
There is no considering here. Just calculate the number. It's a relatively straightforward programming exercise to do so (at least for well defined betting abstraction). The number on some chosen flops would be enough to give at least the ballpark.

Quote:
You're making it sound like it's a huge number but you don't have to buy a supercomputer for that. You can rent computer power through amazon or google. I don't know exactly how much it would cost but I'm guessing something on the order of $250.000. A good chunk of money, but it might be worth it if high stakes limit holdem was still alive.
Yeah I think 150-200k. At this point it's probably better to harvest Cepheus's preflop (or even flop) ranges and recalc from there.
punter11235 is offline   Reply With Quote
Old 05-03-2015, 09:39 PM   #1071
g-p
Carpal \'Tunnel
 
g-p's Avatar
 
Join Date: Apr 2005
Location: 11
Posts: 6,705
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by feedmykids2 View Post
Still waiting for you and durrrr to make billions and win a nobel prize with this theory you have.
its not nobel worthy, it doesnt contradict nash

http://en.wikipedia.org/wiki/Bayesian_game
g-p is offline   Reply With Quote
Old 05-03-2015, 09:40 PM   #1072
Sam Ganzfried
centurion
 
Join Date: Oct 2014
Posts: 187
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by punter11235 View Post

There is no considering here. Just calculate the number. It's a relatively straightforward programming exercise to do so (at least for well defined betting abstraction). The number on some chosen flops would be enough to give at least the ballpark.

PM me your algo for computing best response/exploitabillity in imperfect-recall games/abstractions and maybe we can work something out

Last edited by Sam Ganzfried; 05-03-2015 at 09:55 PM.
Sam Ganzfried is offline   Reply With Quote
Old 05-03-2015, 10:07 PM   #1073
Keruli
adept
 
Keruli's Avatar
 
Join Date: Jul 2009
Posts: 703
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

what's a pure strategy in poker?
Keruli is offline   Reply With Quote
Old 05-03-2015, 10:13 PM   #1074
dodgybob
old hand
 
dodgybob's Avatar
 
Join Date: Jan 2006
Posts: 1,899
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by Keruli View Post
what's a pure strategy in poker?
It just means 'not a mixed strategy'.
dodgybob is offline   Reply With Quote
Old 05-03-2015, 10:17 PM   #1075
Sam Ganzfried
centurion
 
Join Date: Oct 2014
Posts: 187
Re: WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot

Quote:
Originally Posted by punter11235 View Post
Well, I am not sure if they would beat anyone but yeah, their solution is very good and won't be easy to replicate.
On the other hand they calculated it for 70 days on 4800 cores... while it's impressive they got the distributed code right it doesn't sound very impressive considering the amount of resources were spent. I mean... there is not much to compare because normal people don't have access to supercomputers and the guys who won limit competitions so far (Cepheus didn't compete yet I think) did it having way smaller hardware and they claim they solved it without approximation as well (although I have no idea if they are right about it). If they did it's safe to assume they are ahead of the curve on this one.



I understand it.There is value in developing more general approach which is used to poker than focusing just on poker. I am just saying that the fact that you are the best academic team doesn't mean you have much of automatic authority here.
It seems to me you guys made some design decisions which are just very counter-productive to developing a strong playing entity.
I am a bit jaded because there was a discussion about Tartanian before and I had the impression the author doesn't understand much about poker (although I am sure he understands a lot about math/programming in general) then there are claims that Claudico is some top AI or what not while it's clear to everyone it's blundering left and right and taking 30 seconds to recalculate river on 64 cores. It should be easy to make some claims about exploitability as well (at least for flop exploitability or w/e) and those are nowhere to be seen while the author is claiming people can learn a lot from the bot. Where you claim stuff that you have this amazing thing and people can learn a lot of from it then provide the number quantifying how good it is. It's not a rocket science to calculate it or at least approximate it well.
If it's 25bb/100 from NE then we know it's decent but far away from being close to the solution. If it's 50bb/100 we know it's just not good at all (even if it's stronger than other publicly available stuff).
I object to this post. You are implying that I stated all the things you are saying. E.g., "you claim stuff that you have this amazing thing and people can learn a lot of from it then provide the number quantifying how good it is." But in post https://forumserver.twoplustwo.com/sh...postcount=1070 you only give links to quotes by Professor Sandholm, and you never clarified whom your different "the author"'s refer to. Please stop attributing quotes to me that I did not say.
Sam Ganzfried is offline   Reply With Quote

Reply
      

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off


Forum Jump


All times are GMT -4. The time now is 03:43 PM.


Powered by vBulletin®
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.
Copyright © 2008-2020, Two Plus Two Interactive