Gorilla Player: question for Don, et al..

[Gorilla Player]

> The complexity of the interaction of various
> rank depletions on strategic decisions (wow,
> that was a mouthful!). single-parameter
> count systems just aren't equipped to handle
> this; they need multiple side counts to be
> able to improve PE.

> Don

Does that mean that simply because there is no way to perfectly predict the order the cards will come out as they are dealt, there is no way to do _perfect_ BS deviations? IE even if you know that there are 4 aces, 1 seven, and 4 tens, doubling on a 10 could always get that 7 and lose. Or is that the wrong idea.

IE now that I understand what a PE of 0.0 means, it would be nice to know what 1.0 means and exactly what would be required to reach that. IE is PE 1.0 something that requires a perfect crystal ball to work, or is there a real algorithm that doesn't require information about remaining undealt cards that can reach PE=1.0???

Note that when I said "information about undealt cards" I was talking about _order_ not card counts, of course...

Sorry to beat this dead horse, but inquiring minds want to know. Or at least this inquiring mind.

[Don Schlesinger]

> Does that mean that simply because there is
> no way to perfectly predict the order the
> cards will come out as they are dealt, there
> is no way to do _perfect_ BS deviations?

No, it doesn't mean that at all.

> even if you know that there are 4 aces, 1
> seven, and 4 tens, doubling on a 10 could
> always get that 7 and lose. Or is that the
> wrong idea.

Wrong idea.

> IE now that I understand what a PE of 0.0
> means, it would be nice to know what 1.0
> means and exactly what would be required to
> reach that.

Nothing more than perfect knowledge of every card remaining to be played. Once you know that, you know what the best play is for the hand in front of you. You don't have to know the order of the cards; you simply have to know that, given the subset of cards remaining, one strategy will be superior to all others. That's the play you're looking to make. It doesn't mean that you will not wish you had made a different play; it means that you will not wish you had made a different play less frequently than if you had made any other play. :-)

> IE is PE 1.0 something that
> requires a perfect crystal ball to work, or
> is there a real algorithm that doesn't
> require information about remaining undealt
> cards that can reach PE=1.0???

See above.

> Note that when I said "information
> about undealt cards" I was talking
> about _order_ not card counts, of course...

Order plays no role whatsoever, other than in the combinatorial analysis that determines e.v. for the various plays in the first place.

Don

[Cacarulo]

> But if I tell you A is .51 and B is .55, you
> would conclude A is better. But how _much_
> better?

Ok. All the answers that you've received so far were excellent and I have nothing to add except for this part.
If you want an answer for "how much better a system is" you need to compare the squares of the correlations. It's somehow similar to how much better is a system with a DI of 7.0 compared to another with a DI of 7.1.
In this case the answer would be 7.1?/7.0? = 1.0288 or approximately 2.9% better.

Now, what happen in your example:
You have A = 0.55 and B = 0.51. In this case 0.55?/0.51? = 1.1630 or 16.3% better.
Note that this is different than doing 0.55/0.51=1.0784 or 7.84%. The latter comparison is totally wrong.

Hope this helps.

Sincerely,
Cac

[Gorilla Player]

> No, it doesn't mean that at all.

> Wrong idea.

> Nothing more than perfect knowledge of every
> card remaining to be played. Once you know
> that, you know what the best play is for the
> hand in front of you. You don't have to know
> the order of the cards; you simply have to
> know that, given the subset of cards
> remaining, one strategy will be superior to
> all others. That's the play you're looking
> to make. It doesn't mean that you will not
> wish you had made a different play; it means
> that you will not wish you had made a
> different play less frequently than if you
> had made any other play. :-)

> See above.

> Order plays no role whatsoever, other than
> in the combinatorial analysis that
> determines e.v. for the various plays in the
> first place.

> Don

OK. So to recap. PE=0 is for perfect BS player, every hand, no counting. PE=1 is for hitting 100% of the BS deviations that are correct, but only based on the combination of remaining cards, not knowing anything about the order. Which also means that I can actually do better than 1.0 if I am at home, by turning the remaining cards over and looking at the actual order.

Now I at least understand the min and max numbers, which gives some context. Which is what I was looking for. IE it is often said that BS is played 80% of the time. so HiLo gets 50% of the remaining 20% where deviations are correct. The best PE I've seen was about .67 which means that on a real comparison, we'd get

Hilo = correct 90.2% of the time,
Uston's APC = correct 93.4% of the time?

That was what I was trying to understand, ie a multi-level count that most say is too complicated (Uston's) gains 3.2% more correct plays than Hi-Lo which I consider to be a very easy to use count.

Now, I only hope you don't say "GP, you've missed it again.."



I really am pretty good at math. But I've always considered statistics and probability to be "voodoo math".

Statistician: (proper noun): A person that can use a set of numbers to show or prove anything he wants.

A CS department chairman I worked for almost 20 years ago was a statistician. He made the above point many times in debunking reports that came out of the ed-psych department at the same university...

I would add that clearly if I know the order as well as the composition, I could play even better. But of course except for "mindplay" no one has that information (the house at the Flamingo has it of course, which is a bit troubling).

[Gorilla Player]

> Ok. All the answers that you've received so
> far were excellent and I have nothing to add
> except for this part.
> If you want an answer for "how much
> better a system is" you need to compare
> the squares of the correlations. It's
> somehow similar to how much better is a
> system with a DI of 7.0 compared to another
> with a DI of 7.1.
> In this case the answer would be 7.1?/7.0? =
> 1.0288 or approximately 2.9% better.

> Now, what happen in your example:
> You have A = 0.55 and B = 0.51. In this case
> 0.55?/0.51? = 1.1630 or 16.3% better.
> Note that this is different than doing
> 0.55/0.51=1.0784 or 7.84%. The latter
> comparison is totally wrong.

> Hope this helps.

It does except for that "2.9% better". I'm not sure what that actually means. IE when I play BJ, the only real measure I use is $. If a system wins more money, that can be estimated and/or measured... That's why I've run so many CVCX runs to answer some of these with simmed results. But sims are sims, and a "closed solution" is sometimes nice to have as well, if it is possible of course.

> Sincerely,
> Cac

[Cacarulo]

> It does except for that "2.9%
> better". I'm not sure what that
> actually means. IE when I play BJ, the only
> real measure I use is $. If a system wins
> more money, that can be estimated and/or
> measured... That's why I've run so many CVCX
> runs to answer some of these with simmed
> results. But sims are sims, and a
> "closed solution" is sometimes
> nice to have as well, if it is possible of
> course.

The 2.9% means that for the same game conditions one system wins 2.9% more $/100 rounds than the other. Actually DI? is also known as the SCORE.

The 16.3% means that for the same game conditions one system is able to predict 16.3% more betting opportunities than the other.

Sincerely,
Cac

[Don Schlesinger]

> The 16.3% means that for the same game
> conditions one system is able to predict
> 16.3% more betting opportunities than the
> other.

I learned something. I didn't realize we needed to square the PEs to get this information. Thanks for jumping in!

Don

[Don Schlesinger]

> OK. So to recap. PE=0 is for perfect BS
> player, every hand, no counting. PE=1 is for
> hitting 100% of the BS deviations that are
> correct, but only based on the combination
> of remaining cards, not knowing anything
> about the order. Which also means that I can
> actually do better than 1.0 if I am at home,
> by turning the remaining cards over and
> looking at the actual order.

Sounds good to me.

> Now I at least understand the min and max
> numbers, which gives some context. Which is
> what I was looking for. IE it is often said
> that BS is played 80% of the time. so HiLo
> gets 50% of the remaining 20% where
> deviations are correct. The best PE I've
> seen was about .67 which means that on a
> real comparison, we'd get

> Hilo = correct 90.2% of the time,
> Uston's APC = correct 93.4% of the time?

Yup.

> That was what I was trying to understand, ie
> a multi-level count that most say is too
> complicated (Uston's) gains 3.2% more
> correct plays than Hi-Lo which I consider to
> be a very easy to use count.

Right.

> Now, I only hope you don't say "GP,
> you've missed it again.."

No, you're fine.

> Statistician: (proper noun): A person that
> can use a set of numbers to show or prove
> anything he wants.

And my favorite quote: "Statistics can be made to support many things -- mostly statisticians!"

> I would add that clearly if I know the order
> as well as the composition, I could play
> even better. But of course except for
> "mindplay" no one has that
> information (the house at the Flamingo has
> it of course, which is a bit troubling).

Right again. If you knew the order, many of the plays that the count was telling you to make would turn out to be wrong.

Don

[gorilla player]

> Sounds good to me.

> Yup.

> Right.

> No, you're fine.

> And my favorite quote: "Statistics can
> be made to support many things -- mostly
> statisticians!"

> Right again. If you knew the order, many of
> the plays that the count was telling you to
> make would turn out to be wrong.

> Don

OK. Thanks for your patience. It was a long road.

[Cacarulo]

> I learned something. I didn't realize we
> needed to square the PEs to get this
> information. Thanks for jumping in!

If I remember correctly this was a property of the correlation coefficient.

Cac

[Norm Wattenberger]

> But sims are sims, and a
> "closed solution" is sometimes
> nice to have as well, if it is possible of
> course.

There is no closed solution. That's why we have sims. PE, BC, IC are very interesting numbers, but they are only an estimate of a strategy's effectiveness.

[ES]

What is (are) the best measure(s) of a strategy's effectiveness? What is a "closed solution?"

[Norm Wattenberger]

> What is (are) the best measure(s) of a
> strategy's effectiveness?

SCORE as discussed somewhere in the thread

> What is a "closed solution?"

Generally a formula that provides an exact answer. But BJ is an intractable math problem.