CCCafe Post# 4539 of 16788
From: zengrifter
Date: Sat Mar 24, 2001 2:05 pm
Subject: Precise v. approximate indices
SBA's Karel Janecek, Zengrifter, and the MegalaDon
debating the issue of "precision" indices
Karel:
I never claimed that it is crucial to
have precise indices to half integer or so. PS: For
your information, the new version of SBA ready in a
week or two supports all the wonging-in, wonging-out,
and sitting-out features, which were used in the
rabbit, perfect, etc. article in fall BJF (Don
Schlesinger)
ZenGrifter:
Index precision even within 2-3 integers is NOT needed, even if RA. The wonging-in/out sim will not accomplish the type of betting sim needed for today's game, consolidation betting, card eating, psuedo-progressions, etc. -EXCEPT- to the extent that the relevant sim data has primarily already been gleaned and published (as note by you)and assimilated ("exit @ -1, enter @ +1," etc.)
Karel:
Well, then we are in disagreement again.
An error of 2 for index value is already
significant, especially for positive indices (with high bet
out). Of course, if this error is just in one index it
is not major. The error is usually significant if
you have all indices (randomly) with error 2.
In any case, it is useful to have precise simulated
indices rather than approximated ones, even though (or
rather even more because of ) in practical play one does
approximate estimates of the current TC. The point is that
the errors otherwise cumulate. For example, if
simulated indices are 1 TC off, and the player makes
estimate errors in actual play plus or minus 2 TCs, the
total errors range is between -1 TC up to 3 TC. In
relative terms, being off 1 TC all the time vs. being
precise is smaller difference compared to being off -1 TC
up to 3 TC vs. -2 TC up to 2 TC. The reason is that
the error cost is NOT A LINEAR FUNCTION of the error.
Being 3TC off 25% of time is quite a bit worse than
being 1TC off 75% of time, or being off 2 TC 37.5% of
time.
As a conclusion: Given that, in reality,
people need to estimate the current index with some
random errors, it is even more important to have the
simulated index (the starting benchmark) precise (or close
to precise).
Observe that this may seem quite contradictory to general intuition.
ZenGrifter:
The worst that I would imagine a sufficiently-skilled player being off in TC estimation is by 1 approximately 25% of the time - so the 'compunding effect' as you call it would be a fraction of your estimate.
One element that I beleive you have inadvertantly obscurred is the sub-issue of 'approximated' (algebraic-algorithm?) indices v. simulated. My positition being that algebraic indices rounded to within 2 integers, will perform equally, for all intent and purpose within 2-3/100s of 1% of the gain available from the most extensively crafted, precision indices (non-composition-dependent), floored, truncated, or otherwise. This is why index generation by the likes of SmartCards, NeUltra, BJ678, BJSrat, etc. is more than completely adequate.
In any event, I would not refer to index rounding within 2 or 3 integers to be an error, and I WANT TO accept the studies of ASnyder, JAustin, GeoC, and others that reveal that 'extreme rounding' does NOT lose significant power : -1/0/+1 = 0, +2/+3+4 = +3, etc. ASnyder's ZEN indices are rounded even more extremely: indices even more: -10,-5,0,+5,+10,+15. According to Snyder it was JAustin using SBA who found that... GeoC's grossly rounded indices underperformed precise indices by 2/100s of a percent, requiring over 2 billion hands to be conclusive. Therefore one would not play enough hands in a lifetime to gain a conclusive increased gain from the precision-simulated variety. ZG
Karel writes -
If all indices are 2 TC
(Hi-Lo) off, the loss will be most certainly larger. If
they are randomly off with maximum of 2 TCs
error, then your number may well be correct for
some setup. However, 3% difference is still
significant! This 3% is extra profit for free, with no
extra effort. A simple calculation shows that for $60
per hours expectation bettor, this extra profit
covers the cost of a simulator like SBA after just
a week of playing, purely from the optimization of
indices. (And there is much more.)
It is clear that for a decent player, it is just irrational
to use approximated indices when precise indices are
readily available.
Don Schlesinger writes -
Here's what I don't understand.
In any endeavor in life, there is a cost for being
"sloppy" and imprecise. Either a player wants precision,
or he doesn't.
Grifter makes it sound like we need to choose: either do all the "real-world"
things he is so enamored of (pseuso-progression, card-eating, etc.) or use precise indices.
Heaven forbid we should strive for "all of the above."
Makes no sense to me. You either want to play the game to the
maximum of your ability or you don't. Why try to convince
us, Grift, that sloppy is better? We should use the
most precise indices we can generate. And, those
should be the most precise RA indices, simply because RA
is better.
Yet, that set is still just ONE set of indices! Where's the problem?
Too much for your brain to assimilate? Fine. Settle for less. But,
don't try to convince others, who wish to make the effort, to settle for less.
Because, you can argue until you're blue in the face, but sloppy isn't going to
come out better than precise.
The whole thread is kind of pointless, don't you think, Grift?
ZenGrifter writes (to Karel)-
Some apparent errors in your analysis include the impossibility of obtaining precise/total dependent (a misnomer in any event) without specifying #decks and pene-level. Conversely your precise indices are only good for one game each - Bellagio 2D,s17,DAS,65%; or Silver Legacy 2D,h17,noDAS,75%; or Caesars 2D,s17,DAS,LS,55%, etc.
Further, because the borders for the hit-stand-double decisions are quite WIDE, as Snyder (and your predecessor John Imming/UBE) reveals, for what is truly a 'COIN TOSS' decision, reliance on so-called precise indices can actually be a slight detriment to performance - the wide-border/coin-toss decisions are an excellent opportunity to call into play 'intuitive' and meta-conscious awareness, which is why a human proficient counter can out play a computer programmed to play the same count system - the computer wouldn't notice that three 5s and three 4s played out in the first 1/2D of 2Ds, so 16v 10 is a stand @ -1, and other key card considerations - even if the conscious mind didn't register the card-depletion event, we all know that the brain didn't miss it, so these wide-border decision moments are a great opportunity for enhanced play that would otherwise be missed by over-reliance on so-called 'precision indices.'
ZenGrifter writes (to Don)
DS, replace 'sloppy' with 'fuzzy' and I will opt for fuzzy.
zg
From: zengrifter
Date: Sat Mar 24, 2001 2:05 pm
Subject: Precise v. approximate indices
SBA's Karel Janecek, Zengrifter, and the MegalaDon
debating the issue of "precision" indices
Karel:
I never claimed that it is crucial to
have precise indices to half integer or so. PS: For
your information, the new version of SBA ready in a
week or two supports all the wonging-in, wonging-out,
and sitting-out features, which were used in the
rabbit, perfect, etc. article in fall BJF (Don
Schlesinger)
ZenGrifter:
Index precision even within 2-3 integers is NOT needed, even if RA. The wonging-in/out sim will not accomplish the type of betting sim needed for today's game, consolidation betting, card eating, psuedo-progressions, etc. -EXCEPT- to the extent that the relevant sim data has primarily already been gleaned and published (as note by you)and assimilated ("exit @ -1, enter @ +1," etc.)
Karel:
Well, then we are in disagreement again.
An error of 2 for index value is already
significant, especially for positive indices (with high bet
out). Of course, if this error is just in one index it
is not major. The error is usually significant if
you have all indices (randomly) with error 2.
In any case, it is useful to have precise simulated
indices rather than approximated ones, even though (or
rather even more because of ) in practical play one does
approximate estimates of the current TC. The point is that
the errors otherwise cumulate. For example, if
simulated indices are 1 TC off, and the player makes
estimate errors in actual play plus or minus 2 TCs, the
total errors range is between -1 TC up to 3 TC. In
relative terms, being off 1 TC all the time vs. being
precise is smaller difference compared to being off -1 TC
up to 3 TC vs. -2 TC up to 2 TC. The reason is that
the error cost is NOT A LINEAR FUNCTION of the error.
Being 3TC off 25% of time is quite a bit worse than
being 1TC off 75% of time, or being off 2 TC 37.5% of
time.
As a conclusion: Given that, in reality,
people need to estimate the current index with some
random errors, it is even more important to have the
simulated index (the starting benchmark) precise (or close
to precise).
Observe that this may seem quite contradictory to general intuition.
ZenGrifter:
The worst that I would imagine a sufficiently-skilled player being off in TC estimation is by 1 approximately 25% of the time - so the 'compunding effect' as you call it would be a fraction of your estimate.
One element that I beleive you have inadvertantly obscurred is the sub-issue of 'approximated' (algebraic-algorithm?) indices v. simulated. My positition being that algebraic indices rounded to within 2 integers, will perform equally, for all intent and purpose within 2-3/100s of 1% of the gain available from the most extensively crafted, precision indices (non-composition-dependent), floored, truncated, or otherwise. This is why index generation by the likes of SmartCards, NeUltra, BJ678, BJSrat, etc. is more than completely adequate.
In any event, I would not refer to index rounding within 2 or 3 integers to be an error, and I WANT TO accept the studies of ASnyder, JAustin, GeoC, and others that reveal that 'extreme rounding' does NOT lose significant power : -1/0/+1 = 0, +2/+3+4 = +3, etc. ASnyder's ZEN indices are rounded even more extremely: indices even more: -10,-5,0,+5,+10,+15. According to Snyder it was JAustin using SBA who found that... GeoC's grossly rounded indices underperformed precise indices by 2/100s of a percent, requiring over 2 billion hands to be conclusive. Therefore one would not play enough hands in a lifetime to gain a conclusive increased gain from the precision-simulated variety. ZG
Karel writes -
If all indices are 2 TC
(Hi-Lo) off, the loss will be most certainly larger. If
they are randomly off with maximum of 2 TCs
error, then your number may well be correct for
some setup. However, 3% difference is still
significant! This 3% is extra profit for free, with no
extra effort. A simple calculation shows that for $60
per hours expectation bettor, this extra profit
covers the cost of a simulator like SBA after just
a week of playing, purely from the optimization of
indices. (And there is much more.)
It is clear that for a decent player, it is just irrational
to use approximated indices when precise indices are
readily available.
Don Schlesinger writes -
Here's what I don't understand.
In any endeavor in life, there is a cost for being
"sloppy" and imprecise. Either a player wants precision,
or he doesn't.
Grifter makes it sound like we need to choose: either do all the "real-world"
things he is so enamored of (pseuso-progression, card-eating, etc.) or use precise indices.
Heaven forbid we should strive for "all of the above."
Makes no sense to me. You either want to play the game to the
maximum of your ability or you don't. Why try to convince
us, Grift, that sloppy is better? We should use the
most precise indices we can generate. And, those
should be the most precise RA indices, simply because RA
is better.
Yet, that set is still just ONE set of indices! Where's the problem?
Too much for your brain to assimilate? Fine. Settle for less. But,
don't try to convince others, who wish to make the effort, to settle for less.
Because, you can argue until you're blue in the face, but sloppy isn't going to
come out better than precise.
The whole thread is kind of pointless, don't you think, Grift?
ZenGrifter writes (to Karel)-
Some apparent errors in your analysis include the impossibility of obtaining precise/total dependent (a misnomer in any event) without specifying #decks and pene-level. Conversely your precise indices are only good for one game each - Bellagio 2D,s17,DAS,65%; or Silver Legacy 2D,h17,noDAS,75%; or Caesars 2D,s17,DAS,LS,55%, etc.
Further, because the borders for the hit-stand-double decisions are quite WIDE, as Snyder (and your predecessor John Imming/UBE) reveals, for what is truly a 'COIN TOSS' decision, reliance on so-called precise indices can actually be a slight detriment to performance - the wide-border/coin-toss decisions are an excellent opportunity to call into play 'intuitive' and meta-conscious awareness, which is why a human proficient counter can out play a computer programmed to play the same count system - the computer wouldn't notice that three 5s and three 4s played out in the first 1/2D of 2Ds, so 16v 10 is a stand @ -1, and other key card considerations - even if the conscious mind didn't register the card-depletion event, we all know that the brain didn't miss it, so these wide-border decision moments are a great opportunity for enhanced play that would otherwise be missed by over-reliance on so-called 'precision indices.'
ZenGrifter writes (to Don)
DS, replace 'sloppy' with 'fuzzy' and I will opt for fuzzy.
zg