Norm Wattenberger: Comparing strategies

[Norm Wattenberger]

In response to the Zen vs. Zen post on this page, on another site there is a post disparaging SCORE that begins ?SCORE is not a realistic comparison of how systems perform in real world games?.? The poster continues by claiming that SCORE doesn?t relate to the real world because it uses impractical bet sizes. IMO, the poster misunderstands the power of SCORE. In fact, this is SCORE?s strong point. Have you ever noticed that somehow every Blackjack book claims that the system described in that book sims better than other systems? They can?t all be the best. The reason is they use the obsolete methodology suggested by the poster. That you ?just set up a standard computer sim with whatever spread you use, and run off 100 million hands. Forget about the perfect SCORE bets.? That?s what these books have done over the years and it is why their comparisons are so poor.

Let me try to show this graphically. Below is a chart of the six deck EVs for Hi-Opt II (no side count) and HiLo for different penetrations using practically sized bets. You will see that HiOpt II beats HiLo for high penetrations and low penetrations but HiLo wins for medium penetrations. So if I?m writing a book on Hi-Opt II, I use 5/6 penetration for my example and if writing a book for Hi-Lo, I use 4/6 penetration. And I can ?prove? that my system is better. Using this obsolete methodology, you can pick and chose your examples to prove most anything you wish. Hence the contradictory claims. Clearly if we want to fairly compare two strategies, we need a more accurate methodology.



Now let us take the same two strategies, but use the SCORE comparison methodology. The below chart shows the SCOREs for the same penetrations. The curves are smooth and Hi-Opt II always wins.



Picking the penetration point is just one method of fudging sims. You can also pick betting ramps that unfairly disadvantage one system. But, it?s far more difficult to fiddle with the examples in order to show your system better if it isn?t when using SCOREs. Why does this methodology work better? Because it forces the strategies to be tested with bets tuned to each strategy and because it takes into account the standard deviation as the unit sizes change. In other words, it normalizes the two strategies so one is not unfairly disadvantaged.

Conclusion I: Be wary of strategy comparisons that do not use SCORE methodology.

One other note. The poster suggests that SCOREs are not practical in the real world because they require odd bets like $167. This is true in the pure sense. It is also why the tables in Blackjack Attack 3e provide both optimal and practical bets. The optimal bets are actually calculated to the penny. The practical bets are rounded to reasonable chip sizes. For example, 62/104 S17 LS 1-4 spread, the optimal bets are 24, 46, 90 & 97 dollars and some cents. Clearly unusable. The practical bets are 25, 50 and 100, eminently usable. So how does this affect SCORE? 30.87 for optimal and 30.75 for practical. There is barely a difference. That is because CVCX makes numerous passes and uses a feedback mechanism to ensure the best possible SCORE even when using practical bet sizes.

Conclusion II: Don?t be scared off by comments that SCORE can?t be used in the real world.

[Don Schlesinger]

> In response to the Zen vs. Zen post on this
> page, on another site there is a post
> disparaging SCORE that begins ?SCORE is not
> a realistic comparison of how systems
> perform in real world games?.?

Bet you can't find that article anymore! :-)

Don

[Don Schlesinger]

Apparently, Norm was thinking of a different article and author than I was. One such article, stating similar, mistaken (ignorant?), ideas, was deleted. I guess the other one remains as a testimony to some people's stupidity in this area.

The SCORE concept has been embraced by virtually every known blackjack authority and theoretical mathematician. Decrying the concept not only shows lack of understanding, ignorance, or both, but also a certain mean-spiritedness for which the author has become well known in recent years.

Bottom line: Who cares what the article says? We take it from where it comes and move on.

Don

[rich garcia]

>

In response to the Zen vs. Zen post on this
> page, on another site there is a post
> disparaging SCORE that begins ?SCORE is not
> a realistic comparison of how systems
> perform in real world games?.? The poster
> continues by claiming that SCORE doesn?t
> relate to the real world because it uses
> impractical bet sizes. IMO, the poster
> misunderstands the power of SCORE. In fact,
> this is SCORE?s strong point. Have you ever
> noticed that somehow every Blackjack book
> claims that the system described in that
> book sims better than other systems? They
> can?t all be the best. The reason is they
> use the obsolete methodology suggested by
> the poster. That you ?just set up a standard
> computer sim with whatever spread you use,
> and run off 100 million hands. Forget about
> the perfect SCORE bets.? That?s what these
> books have done over the years and it is why
> their comparisons are so poor.

> Let me try to show this graphically. Below
> is a chart of the six deck EVs for Hi-Opt II
> (no side count) and HiLo for different
> penetrations using practically sized bets.
> You will see that HiOpt II beats HiLo for
> high penetrations and low penetrations but
> HiLo wins for medium penetrations. So if I?m
> writing a book on Hi-Opt II, I use 5/6
> penetration for my example and if writing a
> book for Hi-Lo, I use 4/6 penetration. And I
> can ?prove? that my system is better. Using
> this obsolete methodology, you can pick and
> chose your examples to prove most anything
> you wish. Hence the contradictory claims.
> Clearly if we want to fairly compare two
> strategies, we need a more accurate
> methodology.

> Now let us take the same two
> strategies, but use the SCORE comparison
> methodology. The below chart shows the
> SCOREs for the same penetrations. The curves
> are smooth and Hi-Opt II always wins.

> Picking the penetration point is just
> one method of fudging sims. You can also
> pick betting ramps that unfairly
> disadvantage one system. But, it?s far more
> difficult to fiddle with the examples in
> order to show your system better if it isn?t
> when using SCOREs. Why does this methodology
> work better? Because it forces the
> strategies to be tested with bets tuned to
> each strategy and because it takes into
> account the standard deviation as the unit
> sizes change. In other words, it normalizes
> the two strategies so one is not unfairly
> disadvantaged.

> Conclusion I: Be wary of strategy
> comparisons that do not use SCORE
> methodology.

> One other note. The poster suggests that
> SCOREs are not practical in the real world
> because they require odd bets like $167.
> This is true in the pure sense. It is also
> why the tables in Blackjack Attack 3e
> provide both optimal and practical bets. The
> optimal bets are actually calculated to the
> penny. The practical bets are rounded to
> reasonable chip sizes. For example, 62/104
> S17 LS 1-4 spread, the optimal bets are 24,
> 46, 90 & 97 dollars and some cents.
> Clearly unusable. The practical bets are 25,
> 50 and 100, eminently usable. So how does
> this affect SCORE? 30.87 for optimal and
> 30.75 for practical. There is barely a
> difference. That is because CVCX makes
> numerous passes and uses a feedback
> mechanism to ensure the best possible SCORE
> even when using practical bet sizes.

> Conclusion II: Don?t be scared off by
> comments that SCORE can?t be used in the
> real world.

[Norm Wattenberger]

You don't see them? If you have a slow connection, it might take awhile for them to display.

[Zenfighter]

[Cacarulo]

> In response to the Zen vs. Zen post on this
> page, on another site there is a post
> disparaging SCORE that begins ?SCORE is not
> a realistic comparison of how systems
> perform in real world games?.?

Are you talking about the "Zen vs Halves" post or is there a "Zen vs Zen" post?

I would like to add to your excellent post a couple of things. First, we should do two types of comparisons: One for PLAY-ALL and another for WONGING (specially with shoes). In the case of Hi-Opt II (with no side of Aces) vs Hi-Lo you will see that Hi-Lo performs better when using an optimal wonging point. It's not the case when using for example a TC of -1 as the exit point.

Another point is the number of playing indices. Catch-22 is a good number but be careful when different pens are compared. Indices may vary.
So, for doing a fair comparison we also need the correct indices for the specific penetration.

Sincerely,
Cac

[Norm Wattenberger]

> Are you talking about the "Zen vs
> Halves" post or is there a "Zen vs
> Zen" post?

Zen vs. Zen

> I would like to add to your excellent post a
> couple of things. First, we should do two
> types of comparisons: One for PLAY-ALL and
> another for WONGING (specially with shoes).
> In the case of Hi-Opt II (with no side of
> Aces) vs Hi-Lo you will see that Hi-Lo
> performs better when using an optimal
> wonging point. It's not the case when using
> for example a TC of -1 as the exit point.

> Another point is the number of playing
> indices. Catch-22 is a good number but be
> careful when different pens are compared.
> Indices may vary.
> So, for doing a fair comparison we also need
> the correct indices for the specific
> penetration.

Agree. At one time I was thinking of creating a kind of Super-SCORE. The idea is to specify the games most likely to be played, SCORE them and weight them by frequency to form a metric that includes many scenarios. I'd still like to do this at some point. Within CVCX it wouldn't be that hard since 2,000 sim scenraio results exist in one file. So it would be easy to examine different rules, decks and penetrations for one calculation. The larger problem is defining a way to specify the sims that take part in the calculation.