New Software. Holdem-walkthrough.com -- 100 Billion+ EV Estimates

[Holdem Caulfield]

Fixed. Sorry about the delay.

[jukofyork]

[ QUOTE ]
FWIW I just emailed Stars asking PokerStarsJeff / Alex Scott to respond to your IG thread (twice lol, forgot to link it in the first email)

[/ QUOTE ]
Thanks, hopefully it might spark some kind of debate about all this.

[ QUOTE ]
SillySal thread is scary as hell [img]/images/graemlins/frown.gif[/img]

This is and old skool 2p2er, accused of vaing a recipient of chip dumping / CrazyMike says multiple accounts but probably not a bot / eventually found guilty of "botting" - whatever that may be - with zero evidence so far... damn glad I have only enough $$$ for script testing there right now.

[/ QUOTE ]
It's kinda got too big to follow properly now, but I did think your point should have got way more attention.

[ QUOTE ]
Have you not been following the spacegravy challenge thread in STTF?

[/ QUOTE ]
A little, but it still doesn't inspire me to want to move. From what I can gather he has been getting away with being a hyper-LAG pushbot and 2 weeks isn't long enough for all the regulars to adjust their calling ranges vs him yet. I have looked at the games in the past and the Stars $16s looked way way tougher than Party 33s, so I dread to think what it would be like to play the Stars 60s day in day out...

[ QUOTE ]
BTW, did you miss the quickly deleted (spam) link to the full-auto-bot in here earlier today? nasty stuff, looked far more advanced than winholdem [img]/images/graemlins/frown.gif[/img]

[/ QUOTE ]
I missed it. What gets me is if only the same people making these realtime advisers made offline analysis tools instead then all would be great - Just imagine being able to post mortem your hands with really good AI (as you can using a chess/backgammon engine). The tool presented by the OP is a classic example of this: using this offline to post-mortem hands would make it one of the most interesting tools I've seen for ages, but using it in realtime just seems like cheating to me. If only online poker followed the same play ethics as online chess/backgammon then all would be much better IMO.

Juk [img]/images/graemlins/smile.gif[/img]

[Holdem Caulfield]

[ QUOTE ]
The End Is Nigh.

The poker sites needs to seriously re-think their policy. Ok, so they might get more rake as the games get more and more infested by realtime-EV-analysis-reading-human-mouse-clicking BOTS, but in the long run this WILL kill online poker, and it'll be way way quicker than any WinHoldem "dreamers" making 50c/hour at $0.01/$0.02...


[/ QUOTE ]

Hi Juk, Thanks for the insightful comments. I don't think the situation is as bad as you think, although I definitely wouldn't want to play against someone who is using holdem-walkthrough because the expected outcome would be that we'd both tie and end up losing at rake levels, which basically already happens if you sit at a table with excessively tight players. Knowing the EV is going to get your opponents to bet.

Holdem-walkthrough is just a sophisticated odds calculator that helps you make good decisions but in no way does it play for you. Playing holdem using a rule such as maximizing EV is a very bad strategy, which is something I discuss in more detail in the manual. In a game of chance, such as video poker, where the payoffs are pre-determined, you can play by choosing to maximize EV. This optimal strategy yields a net EV of -.005 (http://wizardofodds.com/jacksorbetter). In a game of skill, such as holdem poker, however, you can expect rational opponents to fold their weak hands and bet their good hands, thus EV calculations based on a random model will be biased as the game becomes less random, that is, when people choose to put money into the pot.

Holdem-walkthrough is a good tool for telling you if you are ahead or behind, but ultimately the decision on how to play the hand is the players responsibility. This is unlike other poker advisor programs that explicitly make betting recommendations.

[jukofyork]

[ QUOTE ]
Holdem-walkthrough is just a sophisticated odds calculator that helps you make good decisions but in no way does it play for you. Playing holdem using a rule such as maximizing EV is a very bad strategy, which is something I discuss in more detail in the manual. In a game of chance, such as video poker, where the payoffs are pre-determined, you can play by choosing to maximize EV. This optimal strategy yields a net EV of -.005 (http://wizardofodds.com/jacksorbetter).

[/ QUOTE ]
But the strategy presented by your application is not "optimal" (in a game-theoretic sense)? It looks like you have used 100+ billion decision points of real play to create an ("exploitative") model which estimates the EV of each action in a given context vs the sampled population dynamic (ie: vs the average player).

[ QUOTE ]
Playing holdem using a rule such as maximizing EV is a very bad strategy

[/ QUOTE ]
Please explain as that just doesn't make any sense? Other than the fact that you would play too predictably vs very good players (and hence need to use a mixed strategy against them), then pretty much every good holdem player plays to maximise EV. If playing vs "an unknown" they play to maximize EV vs the average population dynamic (as you are).

[ QUOTE ]
Holdem-walkthrough is a good tool for telling you if you are ahead or behind, but ultimately the decision on how to play the hand is the players responsibility. This is unlike other poker advisor programs that explicitly make betting recommendations.

[/ QUOTE ]
The user just looks and sees that 4.3 > 3.2 and makes the 4.3 action, and if they see 4.3 ~ 4.2 they could mix their play up a bit, but why would they ever want to rationally choose the lower EV action (assuming your models are accurate)?

Juk [img]/images/graemlins/smile.gif[/img]

[jukofyork]

I just read your manual and this bit in particular:

[ QUOTE ]
We performed our estimations using a ‘random model’ meaning that every time a player made a betting decision there was an equal likelihood that he would fold, check/call or bet/raise. Because there was no player strategy involved, the estimates describe how hands naturally fare in hold’em.

[/ QUOTE ]
The fact that you didn't create your models using real data makes it much less worrying, but don't you think it's a bit misleading to output 'EV' calculated based on opponents who take completely random actions?

Juk [img]/images/graemlins/smile.gif[/img]

[Holdem Caulfield]

[ QUOTE ]
[ QUOTE ]
Holdem-walkthrough is just a sophisticated odds calculator that helps you make good decisions but in no way does it play for you. Playing holdem using a rule such as maximizing EV is a very bad strategy, which is something I discuss in more detail in the manual. In a game of chance, such as video poker, where the payoffs are pre-determined, you can play by choosing to maximize EV. This optimal strategy yields a net EV of -.005 (http://wizardofodds.com/jacksorbetter).

[/ QUOTE ]
But the strategy presented by your application is not "optimal" (in a game-theoretic sense)? It looks like you have used 100+ billion decision points of real play to create an ("exploitative") model which estimates the EV of each action in a given context vs the sampled population dynamic (ie: vs the average player).

[ QUOTE ]
Playing holdem using a rule such as maximizing EV is a very bad strategy

[/ QUOTE ]
Please explain as that just doesn't make any sense? Other than the fact that you would play too predictably vs very good players (and hence need to use a mixed strategy against them), then pretty much every good holdem player plays to maximise EV. If playing vs "an unknown" they play to maximize EV vs the average population dynamic (as you are).

[ QUOTE ]
Holdem-walkthrough is a good tool for telling you if you are ahead or behind, but ultimately the decision on how to play the hand is the players responsibility. This is unlike other poker advisor programs that explicitly make betting recommendations.

[/ QUOTE ]
The user just looks and sees that 4.3 > 3.2 and makes the 4.3 action, and if they see 4.3 ~ 4.2 they could mix their play up a bit, but why would they ever want to rationally choose the lower EV action (assuming your models are accurate)?

Juk [img]/images/graemlins/smile.gif[/img]

[/ QUOTE ]


I calculated the EV using random players thus the EV estimates tell you the functional form of the game but do not account for strategy whatsoever. Poker, as we know, is a game of strategy and skill, it can’t be beat using EV alone.

Against random players, you’ll often find that aggressive play (betting/raising) often has the highest EV because it increases the likelihood that your opponents will fold or that you'll win a big pot. However, against real players you’ll find that the player’s with good hands are all to eager to call your bets and the players with bad hands are quick to fold. Thus, if you play blindly using the max(EV) rule you’re going to be the big sucker at the table and lose at showdown quite often.

I suggest that you treat the EV estimates in the later stages of the game as being ordinal in nature and focus more on the general pattern of what are good bets vs bad bets as opposed to whether you’ll get .1 more small bets if you raise instead of call. However, in the early stages of the game, where everything is still random, I recommend following the EV estimates pretty closely. When playing at 6-player tables, I follow a simple rule of starting with hands that have an EV of .3 or greater, and I find that my long term outcome is .3 small bets/hand. This outcome surprised me and I await to hear other players’ experiences.


There’s one more very big and very important point I want to make. Although I've stated that you have to treat the EV estimates with circumspection, I need to qualify this. I think possibly the most important information you can get from holdem-walkthrough, is that if you can’t beat a random hand, FOLD!

p.s., I'm not versed in game theory, so I can't comment along those lines.

[jukofyork]

[ QUOTE ]
I calculated the EV using random players thus the EV estimates tell you the functional form of the game but do not account for strategy whatsoever. Poker, as we know, is a game of strategy and skill, it can’t be beat using EV alone

[/ QUOTE ]
See my next post - sorry I didn't read you manual before making this post and I was assuming you had used real hands to get the 'EV'.

I'll keep quite now [img]/images/graemlins/smile.gif[/img]

Juk [img]/images/graemlins/smile.gif[/img]

[Holdem Caulfield]

[ QUOTE ]
I just read your manual and this bit in particular:

[ QUOTE ]
We performed our estimations using a ‘random model’ meaning that every time a player made a betting decision there was an equal likelihood that he would fold, check/call or bet/raise. Because there was no player strategy involved, the estimates describe how hands naturally fare in hold’em.

[/ QUOTE ]
The fact that you didn't create your models using real data makes it much less worrying, but don't you think it's a bit misleading to output 'EV' calculated based on opponents who take completely random actions?

Juk [img]/images/graemlins/smile.gif[/img]

[/ QUOTE ]

If I had used real data, then the end of online poker probably would be upon us. [img]/images/graemlins/wink.gif[/img] However, poker sites don’t allow you to use information from games that you weren’t involved in so I deliberately chose to use theoretical data as opposed to empirical data.

I'm going to answer your question by quoting from the manual. I'm really enjoying our discussion. Thanks again for taking the time to dig into this stuff and feel free to ask me anything.

In defense of the random model

As we’ve just discussed, winning at hold’em requires both skill and luck. You can’t win at hold’em by robotic adherence to the max(EV) rule. In spite of the shortcomings of EV though, we believe that there are a number of situations where the knowledge gleaned from the random model can be tremendously valuable to your game. Below are just a few examples.

First and most important, if you can't beat a random hand, fold! This rule drastically reduces “tilt” behavior, ego/bravado betting, chasing your AQo down to the river, and myriad other mistakes that are so easy to make. You’ll be amazed at what a difference will come about in your game once you stop taking bets that have low expectations.

Second, random data tells you how the cards play against themselves as opposed to how humans play cards. You can use this information to exploit weaknesses in your opponents because you know the nature of the game better than they do.

Third, and as discussed above, you will find that most of the suggested courses of action presented by the EV and %WinDif estimates make sense even when you suspect that your opponent has a good, non-random hand.

Fourth, the odds approach to hold’em is also a random model. For example, although you may think you have the theoretical odds to make a call, the reality may be that your opponent has you beat regardless of what happens such as if you had a king high flush draw and he has the ace.

There are ways to correct odds calculations for such situations, but our point here is that odds based decision-making can also be nebulous when you are playing against non-random hands.

Fifth, the pre-flop betting round is very well described by the random model because of the tremendous number of ways that a game can transpire once the community cards are seen.

Choosing which hands to play is half the battle in hold’em. Knowing when to fold is the other half. Knowing when to bet or raise is the remaining 10%. Every time we use HWT we find another area of the game where we are surprised by how well the random model works. It is important to remember, however, that the EV and %WinDif data suggest courses of action that generally work well but may fail in situations such as when your opponent is eager to re-raise your raises. HWT is just a tool. You must always use your skills and knowledge of the game to determine the best way to play a hand.

[indianaV8]

Closest I agree to is jukofyork, nice posts as usual juk.

[ QUOTE ]
so they might get more rake as the games get more and more infested by realtime-EV-analysis-reading-human-mouse-clicking BOTS

[/ QUOTE ]

I don't see this will happen soon however. Human-clicking-bot or actual bot will only be an issue if the games get theoretically solved (so to say), which is not the case currently. And if this happens, site cannot do anything about it.

In the mean time, the best sites can do is to fight multi acconting. This will stop all criminal activities along with any desctructive effects, be it by bots or humans.

Actually, even if sites legalize bots but fight seriously with multiaccounting (which they can EASILY do from GT standpoint!) there will be NO ANY ISSUE for online poker.

And if the games get solved - they get solved. We all need to move on to the next game.

[indianaV8]

[ QUOTE ]

BTW, did you miss the quickly deleted (spam) link to the full-auto-bot in here earlier today? nasty stuff, looked far more advanced than winholdem [img]/images/graemlins/frown.gif[/img]

[/ QUOTE ]

:-))
These bots are usually the least problem of all (at least to all of my current knowledge which is a lot).