Myoo: Recursion problems in strategy generation

[Norm Wattenberger]

Hi T-Hop. I don't think that will be accurate. But, many subsets instead of an average at each penetration might work better. The exact method of TC calculation might also be more critical. In my mind, average or representative subsets fail because counting is inherently non-linear. Even with exact TC calculation. With human TC calculation, the problem is worse.

> I'm working on a hybrid approach that uses simulation
> to generate an average deck composition for each
> count. I have that working, the next step will be to
> use EORs to estimate the standard error. After that I
> will do what you think is impossible with current
> technology and compare the results. ;-)

[kc]

> I'm working on a hybrid approach that uses simulation
> to generate an average deck composition for each
> count. I have that working, the next step will be to
> use EORs to estimate the standard error. After that I
> will do what you think is impossible with current
> technology and compare the results. ;-)

I have a program that computes shoe comp for Hi-Lo or KO for any running count/pen combination. It goes through all of the possible count subsets and computes the probability of drawing each rank using a weighted average of each subset. It also allows for the input of any number of specific removals of each rank. It's interesting that the only time that the probability of drawing a zero tagged card when there are no specific removals and running count is not zero = 1/13 is when pen is exactly 1/2 shoe. When running count equals 0 and there are no specific removals, the probability of drawing a zero tagged card = 1/13 only for a full shoe or 1/2 shoe. It's good for analyzing insurance because all that's needed for that decision is the probability of drawing a ten. It can only do Hi-Lo or KO. Any level one count could be added relatively easily since the subsets could be identified in the same way used for Hi-Lo and KO.

[T. Hopper]

> I have a program that computes shoe comp for Hi-Lo or
> KO for any running count/pen combination. It goes
> through all of the possible count subsets and computes
> the probability of drawing each rank using a weighted
> average of each subset. It also allows for the input
> of any number of specific removals of each rank. It's
> interesting that the only time that the probability of
> drawing a zero tagged card when there are no specific
> removals and running count is not zero = 1/13 is when
> pen is exactly 1/2 shoe. When running count equals 0
> and there are no specific removals, the probability of
> drawing a zero tagged card = 1/13 only for a full shoe
> or 1/2 shoe. It's good for analyzing insurance because
> all that's needed for that decision is the probability
> of drawing a ten. It can only do Hi-Lo or KO. Any
> level one count could be added relatively easily since
> the subsets could be identified in the same way used
> for Hi-Lo and KO.

I use simulation to generate the subsets because the composition of the hand can change based on playing strategy. For example, if you double on 6,5 vs. 10, you are less likely to later face a hit/stand decision with 5,5,6 vs. 10.

[kc]

> I use simulation to generate the subsets because the
> composition of the hand can change based on playing
> strategy. For example, if you double on 6,5 vs. 10,
> you are less likely to later face a hit/stand decision
> with 5,5,6 vs. 10.

Another approach is to specfically remove 10-10-6 and generate an index for 10-6 vs 10, remove 5-5-6-10 and generate an index for 5-5-6 vs 10, etc. Since 10-6 vs 10 occurs more it would have the greater weight in an overall index for 16 vs 10. A sim could tell you how much weight to give to each or possibly the weight could be calculated by paying attention to the order that the cards are dealt. Each pen would have its own index as well. A CA would be needed to weight the EV. At the end a graph could be constructed of index vs pen. It might get complicated, but it may be doable.

[Don Schlesinger]

> Another approach is to specfically remove 10-10-6 and
> generate an index for 10-6 vs 10, remove 5-5-6-10 and
> generate an index for 5-5-6 vs 10, etc. Since 10-6 vs
> 10 occurs more it would have the greater weight in an
> overall index for 16 vs 10. A sim could tell you how
> much weight to give to each or possibly the weight
> could be calculated by paying attention to the order
> that the cards are dealt. Each pen would have its own
> index as well. A CA would be needed to weight the EV.
> At the end a graph could be constructed of index vs
> pen. It might get complicated, but it may be doable.

I think that it has been demonstrated in the past that indices change very little according to varying penetrations -- surely by not more than one. I think that trying to generate indices by different levels of penetration and then averaging them is fine, but I think that the final blended value will differ very little from any of its component parts.

Don

[Norm Wattenberger]

Unbalanced strategies are more penetration sensitive.

[kc]

> I think that it has been demonstrated in the past that
> indices change very little according to varying
> penetrations -- surely by not more than one. I think
> that trying to generate indices by different levels of
> penetration and then averaging them is fine, but I
> think that the final blended value will differ very
> little from any of its component parts.

> Don

This would be a mathematical solution to deriving indices. To ensure correctness of the method it would be best to consider pen. It is neither here nor there how pen influences the indices, just a byproduct of the method.

[Zenfighter]

It seems to me, that for the vast majority of the CC?s (Hi-lo?s ones), obtaining more
precise indices (and memorizing them all), for different penetrations, will be a clearly waste
of time and mental energy, I am afraid. The reason mainly, the low playing efficiency of the
Hi-lo indices to extract noticeable strategic gains, during the first decks of the shoe.

E.g. 15 v T, a very Illustrious one, show us a cc = .55

With, let?s say, two decks dealt out, the playing efficiency sinks dramatically. So what?

The ?Hot dog? issue strikes again, Don. A recurrent problem, isn?t it? :-)

Regards,

Zf