"What decision do I make now?" : Math will not get you there

[emilyisqueen]

The Poker Rock says:

If you have A8o on the button, you fold to that raise.

Then when he raises later and you have AA on the button, you can layeth the smacketh down on him.

The Poker Rock agrees with Utah's roody poo theory about historical data being meaningless. The Millions (and Millions) of The Poker Rock's fans know you have to make your decisions in the now, regardless of past play.

[Utah]

[ QUOTE ]


utah:

Quote:
One of the biggest problems with nets is that you often need lots of data. What makes you think that we dont train it with historical data?


Andrew Prock :

Quote:
I guess that makes this thread rather silly then.

[/ QUOTE ]

I dont get your post at all????? Are you suggesting that my statements contradict each other? If so, you are wrong
[img]/images/graemlins/wink.gif[/img]

[Cyrus]

[ QUOTE ]
I dont get your post at all????? Are you suggesting that my statements contradict each other?

[/ QUOTE ]Dear Utah,

I had high hopes for this thread. I was expecting some non-math, revelatory method of getting the answer to the poker player's question What I Do Now, as you put it. Instead, I understand that it's all a matter of confusing the nomenclature!

You claimed that the math or our past history won't help us there but your method involves, among other things, how the other players perceive us (=past history) and neural networking (=math).

Still, it's been educational, honestly.

Take care.

--Cyrus

[Utah]

[ QUOTE ]
You claimed that the math or our past history won't help us there but your method involves, among other things, how the other players perceive us (=past history) and neural networking (=math).

[/ QUOTE ]

I think this is a tricky conversation and we are still not on the same page. The conversation is really suited for a round table discussion and not an internet forum. Let me see if I can summarize and clarify. A problem can be one of 2 things:

1) a problem with a knowable solution. e.g., 2 player game with one round and known factors.
2) a problem without a knowable solution. e.g., which 3^169 UTG preflop strategies do I choose?

For type 1, I dont need to historically measure my game as the problem can be solved directly. I only need to know the game and the factors. For example, if I need to solve 1567+1656 I can do so directly. No objective measures of my past work in math are neccessary. I dont need to know about the times I calculated correctly or incorrectly. I simply solve the problem in the here and now.

For type 2, there are no answers. Thus, there can be no true objective measurements because what can you measure against. All the hands ever played in the history of poker cannot solve the simple problem I laid out above. So, I cannot ever definitively say, it was correct to raise 10,10 in a particular hand. So, we solve with broad based rules and fuzzy logic. But, again, we are solving in the here and now and not with objective measurements.

To the question of neural nets. Well, the math question is a bit silly and completely unimportant. However, please realize that we are not solving the problem with the neural nets. Rather, we are simply providing information. There is simply no math+technology that exists today that can solve the simple UTG problem that I showed above. That is what I meant by math not being able to get you there.

Today, I presented the latest version of the software to a partner and one of our investors/advisors. We spent 2 hours passionately discussing the issues above.

It is not a question on nomenclature. It is a question of approach. btw - both the partner and investor were blown away by what they saw. They were in awe.

Here is the first slide of the presentation. It is simple but effective for discussing the problem. Nothing else matters.

The player before you just bet $100. There is $650 in the pot and one player to act behind you.

Do you: Check / Raise / Fold?

If you call you may be throwing away $100. If you fail to raise you could be throwing away $100. Fold and you may be throwing away $650

[Andrew Prock]

[ QUOTE ]
There is simply no math+technology that exists today that can solve the simple UTG problem that I showed above. That is what I meant by math not being able to get you there.

[/ QUOTE ]

This all depends on exactly what you mean by "solve". On a practical level, you can use applied mathematics to solve the problem. The quality of the solution with depend on the model that you use, the validity of your assumptions, and the method of solution.

Will it be univerally applicable? No. No mathematical solution ever is. In fact, when you compare any mathematical solution to real world problems you'll find out that the solutions only ever serve as an approximation to the true answer.

The real world is messy like that.

But applied mathematics is perfect for generating approximate solutions to these sorts of problems, and those solutions tend to be very useful.

- Andrew

[Utah]

Hi Andrew,

I agree with almost everything you said and you said much better part of what I was trying to say. The real world is very messy.

The only comment I would add is that I think there are poker problems that are next to impossible to solve with any real accuracy even with applied mathematics and approximation techniques.

[toms2866]

I'm curious to what degree your software's method incorporates traditional poker math alongside the AI approach. It seems that certain poker math could be used to bound potential AI decisions or to provide inputs to the AI model. For example, the "conditions" being fed to the AI model might include immediate pot odds, all-in equities, maximum implied odds, etc. An example of bounding AI decisions might include targeting bluffing frequency so that the prob of bluffing is equal to the pot odds offered to opponents.

[cliff]

FWIW,

I have been around a lot of investor backed startup companies in my career, and every one has a pitch from its CEO that sounds just like this. Differentiate yourself strongly from all competitors, stay vague on your approach with a few colorful talking points, etc. I guess the positive side is that a very few of them actually deliver at the end of the day so we will see.

[Utah]

You are exactly right on the approach. However, you crash any current math program on the market. The best you can do is a 1000x1000 sparse matrix. That isnt very big.

[Utah]

[ QUOTE ]
FWIW,

I have been around a lot of investor backed startup companies in my career, and every one has a pitch from its CEO that sounds just like this. Differentiate yourself strongly from all competitors, stay vague on your approach with a few colorful talking points, etc. I guess the positive side is that a very few of them actually deliver at the end of the day so we will see.

[/ QUOTE ]The product is essentially done and there is not more work left on the poker aspects. However, we are not home free and we could fall down on the other things that are needed to take this to market (if we take it to market as opposed to using for private team play. note:no collusion). We have focused on poker so much that we didnt think of the other things. One of our advisors asked us a lot of questions we did not have answers to.