2007PBWC NLH FinalTable - Bot mistake? or misfortune?

[DevinLake]

Ummm....no...

This is how I started each response in this thread.

[ QUOTE ]
RIIT, you already

[/ QUOTE ]

[ QUOTE ]
Well, what makes this a bad push

[/ QUOTE ]

[ QUOTE ]
what is ROR? I know it means risk of ruin

[/ QUOTE ]

[ QUOTE ]
Ok, because this dumbass thread

[/ QUOTE ]

[ QUOTE ]
I'm not sure exactly what you are saying here

[/ QUOTE ]

[ QUOTE ]
I'm usually pretty good at understanding things,

[/ QUOTE ]

[ QUOTE ]
Ok, I get what u were saying before

[/ QUOTE ]

[ QUOTE ]
I think if this bot was programed correctly

[/ QUOTE ]

[ QUOTE ]
This is why we convert the HH to be readable

[/ QUOTE ]

[ QUOTE ]
Your calling with 1.4%

[/ QUOTE ]


[ QUOTE ]
Ummm....no...


[/ QUOTE ]

[DevinLake]

[ QUOTE ]
The short answer is I don't know. What I'm most interested in is the win,lose,tie river chances for the hero. The tool I use deals 2 random hands and a random river ending and measures the hero's hand against those 2 hands for a win a loss or a tie and sums the results over many trials to get a statistical sampling of the current situation.

[/ QUOTE ]

I find this interesting coming from an engineering background. If you don't know what hand range your equivalent win/lose/tie equities are associated with, how do you verify that your model is valid? IOW, how do you know that the equities calculated would represent a realistic hand range for the villain here?

[Guaran]

[ QUOTE ]
Your calling with 1.4% (AA-QQ). For it to work 97% you bot would have to be raising 46.7% of hands.

[/ QUOTE ]

The percentage that I'm raising would have nothing to do with the percentage of the play working. For it to be working 97% of the time, is just that, it works 97% of the time. It doesn't matter what percent of the time I'm raising or what I'm calling with, because you don't know my range of what I'm reraising with.

[DevinLake]

[ QUOTE ]
[ QUOTE ]
Your calling with 1.4% (AA-QQ). For it to work 97% you bot would have to be raising 46.7% of hands.

[/ QUOTE ]

The percentage that I'm raising would have nothing to do with the percentage of the play working. For it to be working 97% of the time, is just that, it works 97% of the time. It doesn't matter what percent of the time I'm raising or what I'm calling with, because you don't know my range of what I'm reraising with.

[/ QUOTE ]

Is that the way it works? It has nothing to do with the frequency you are raising and the frequency you are calling? Just simply that you've decided that the play will work 97% of the time? Even though unless you have a redicously loose min 3-betting range, you have AA-QQ more than 3% of the time?

[RIIT]

[ QUOTE ]
[ QUOTE ]
The short answer is I don't know. What I'm most interested in is the win,lose,tie river chances for the hero. The tool I use deals 2 random hands and a random river ending and measures the hero's hand against those 2 hands for a win a loss or a tie and sums the results over many trials to get a statistical sampling of the current situation.

[/ QUOTE ]

I find this interesting coming from an engineering background. If you don't know what hand range your equivalent win/lose/tie equities are associated with, how do you verify that your model is valid? IOW, how do you know that the equities calculated would represent a realistic hand range for the villain here?

[/ QUOTE ]

I'll temporarily not answer here for a moment and echo this question back to you:

If you can't/won't produce a "certainty" or "error" number for your "guess" about the opponent hand range, how can you know your conclusions are valid?

What happens with most of our human analysis is that we say "oh any good player would play only this range" and we then move directly to the statistical analysis and never give any thought to margin of error on the "guess". So all poker analysis that begins with a hand range "guess" is "uncertain" at best. And even if either of us is willing to say "I'm 70% certain this range is correct" then how do we factor that uncertainty into the calculation; and then how certain are we that our 70% certain guess is correct (and on and on).

So what we end up doing is treating our original "guess" work about hand ranges as 100% certain (because it makes our job easier) in which case the conclusions are 100% certain for sure.

And now I'll try to answer you:

The win/lose/tie numbers are accurate for the sample taken. I make this statment on the basis that it's rather straightforward to do a Monte Carlo sampling of a probability field (I hope you agree).

Your question assumes that we can't measure healthy win/lose/tie numbers without beginning with a "guess" for a hand range. But we can do exactly this as long as we sample the probability field correctly and we do that in this case by putting the villain on ATC a total of 2 times (because 2 opponent hands were dealt; we'd do it 3 times if 3 opponents had been dealt and so on).

RIIT

[Guaran]

[ QUOTE ]

Is that the way it works? It has nothing to do with the frequency you are raising and the frequency you are calling? Just simply that you've decided that the play will work 97% of the time? Even though unless you have a redicously loose min 3-betting range, you have AA-QQ more than 3% of the time?

[/ QUOTE ]

When you are isolating a single hand for discussion, unless you have a specific hand to profile on, it doesn't have any connection to raising frequency.

The point is I reraised 3-handed, and it decided to push. I can see arguments for both ways on if it was the correct decision; but personally don't see it effective in the long run because of the possibility of the opponent calling and you being out. It being a winner take all format & shorthanded the decision is a risk that it decided to take.

[DevinLake]

Well, I'm not a math guy. So, I'll admit I don't know what you mean when you say you put the villain on ATC a total of 2 times, or that you are sampling a probablility field, or any of that.

But maybe you could explain in laymen's terms, how sampling a probability field goes about producing realistic numbers as they relate to a real life poker situation?

Or maybe you could point me to some good sources to learn this stuff?

Now, to address you're question. We do consider and account for the fact that our information is incomplete and assigning hand ranges is not going to be 100%. In SNGs we do ICM calculations were we assign a min edge required. This edge can account for, among other things, error in our hand range estimates. You can certainly do the same for a cash game scenario.

However, sometimes we can look a situation in a very simple way. Like what hand range would this villain need for this shove to be +EV? If we then find that hand range, and see that it very unrealistic, we could determine that that play would not be +EV.

That's where my question comes from, how do you know that the equities that your sampling of probablility fields yeilds results that are in the realm of reasonability?

I'm not trying to be overly critical or anything, but like I said in my last post, I have an engineering background. So, I have a difficult time accepting a solution to a problem that has no physical meaning. So, I'm just trying to get a grasp on how you are getting to these equities and how I/we can be reasonable sure they make sense.

I am in no way saying your method is invalid, because I really don't understand what you are doing and how it pertains to a real life situation.

[DevinLake]

[ QUOTE ]
[ QUOTE ]

Is that the way it works? It has nothing to do with the frequency you are raising and the frequency you are calling? Just simply that you've decided that the play will work 97% of the time? Even though unless you have a redicously loose min 3-betting range, you have AA-QQ more than 3% of the time?

[/ QUOTE ]

When you are isolating a single hand for discussion, unless you have a specific hand to profile on, it doesn't have any connection to raising frequency.

The point is I reraised 3-handed, and it decided to push. I can see arguments for both ways on if it was the correct decision; but personally don't see it effective in the long run because of the possibility of the opponent calling and you being out. It being a winner take all format & shorthanded the decision is a risk that it decided to take.

[/ QUOTE ]

What it sounds like you are saying is that, button's decision to raise is random, your decision to re-raise is random, and then button's decision to shove is random because we don't know exactly what each other have. So, we have no way of knowing if they will call or not?

[Guaran]

Well to be a good player and not be able to be profiled, yes, every decision needs to be calculated but random. You should have a reason for doing something but no one else should be able to figure out what you have by your actions. My min-raise at the time had a specific meaning, and if the same situation came up, depending on who I am playing I will play that differently every time.

[RIIT]

[ QUOTE ]
Well, I'm not a math guy. So, I'll admit I don't know what you mean when you say you put the villain on ATC a total of 2 times, or that you are sampling a probablility field, or any of that.

[/ QUOTE ]

In programmatic terms you sample a 3 handed river ending many times: one of the hands is the bot A9o, the other 2 hands are random, the board is random. Note that this is NOT intended to describe how Guaran is playing; it's merely intended to describe the statistical truth of an infinite number of random river trials. So to state this as simply as I can, it is the answer to the question: "What's the best A9o can do in a no-foldem game against 2 hands?"

[ QUOTE ]
But maybe you could explain in laymen's terms, how sampling a probability field goes about producing realistic numbers as they relate to a real life poker situation?

[/ QUOTE ]

In laymen's terms, it's an analysis of the situation against 2 opponents who refuse to fold for any reason. Yes, HTC folded but the idea is the same because he was dealt and that makes the game different than if he was not dealt in that the bot's combined opponent has 2 chances to get a winning hand. (i.e. it's an analysis of 2 opponents who have no folding range)

[ QUOTE ]
Or maybe you could point me to some good sources to learn this stuff?

[/ QUOTE ]

I would but I'm not allowed to post links.

[ QUOTE ]
Now, to address you're question. We do consider and account for the fact that our information is incomplete and assigning hand ranges is not going to be 100%. In SNGs we do ICM calculations were we assign a min edge required. This edge can account for, among other things, error in our hand range estimates. You can certainly do the same for a cash game scenario.

[/ QUOTE ]

Ok fair enough.

[ QUOTE ]
However, sometimes we can look a situation in a very simple way. Like what hand range would this villain need for this shove to be +EV? If we then find that hand range, and see that it very unrealistic, we could determine that that play would not be +EV.

[/ QUOTE ]

Yes understood. This is a different kind of guesswork (which I happen to like). You do the calc only to find that your push only profits against the bottom 5th of hands at which point you think ("I seriously doubt that to be the case since the guy bet a 4th of his stack UTG")

[ QUOTE ]
That's where my question comes from, how do you know that the equities that your sampling of probablility fields yeilds results that are in the realm of reasonability?

[/ QUOTE ]

Well if I sit down with total strangers I've got to have some type of starting point and so I prefer to begin with the idea of "what is the best I can do with this hand if everybody refuses to fold" (which seems like a decent starting point); the rest is simply measuring known frequency and quality. This starting strategy generally brings me into the table much tighter than most everyone to begin with.

[ QUOTE ]
I'm not trying to be overly critical or anything, but like I said in my last post, I have an engineering background. So, I have a difficult time accepting a solution to a problem that has no physical meaning.

[/ QUOTE ]

Well I hope I helped clear this up. My starting point against strangers is no-foldem numbers against the total number of hands dealt. And yes this is a starting "guess".

[ QUOTE ]
So, I'm just trying to get a grasp on how you are getting to these equities and how I/we can be reasonable sure they make sense.

[/ QUOTE ]

I hope I answered your question(s).

[ QUOTE ]
I am in no way saying your method is invalid, because I really don't understand what you are doing and how it pertains to a real life situation.

[/ QUOTE ]

Apologies for not being able to explain it better, faster.

RIIT