improving point count
David Babcock
in which the relevant cards in a suit add to 20 and the values are all
whole numbers. The hypothesis is that such a count might result in
improving the point-count component of hand evaluation with no more
mental gymnastics than remembering the new scheme and then dividing by
2 to get back to familiar territory. ???
David
Hasse
I think you are an the wrong course. Distribution and whether honor
are in the long or shorts suits are of more importance than autotmatic
calculations. Learning that value of the hand takes time and great
effort. Avoiding autonmatic valuation is the first step.
Reint
The problem of obtaining a reliable work-type point count is one of
the problems for people developping bridge computer programs. I do
know this topic well (maybe someone on the forum happens to be an
expert in this field). I am however quite certain that this cannot be
simple integer arithmetic. The valuation of the hand will (have to)
depend strongly on the auction and will therefore have to require
(formalised) rules for each update. Complicated stuff...
Simple point count arithmatic (like Milton work count) only works on
average. However matches are won (or lost!) on exactly those plays
that are non-average.
Reint.
David Babcock
> I think you are on the wrong course. Distribution and whether honor
> are in the long or shorts suits are of more importance than autotmatic
> calculations. Learning that value of the hand takes time and great
Indeed. That's why I said point count *component* of hand
evaluation. I don't mean to dismiss other factors; this just happens
to be the one I posted about.
David
Anton van Uitert
more limited by these flaws than by the way points are assigned to
highcards.
Therefore, imo, an optimal point-count of high cards and intermediates
will not improve things significantly because the flaws remain.
I think Point count is ok as it is. It is a huge advantage that by
virtue of its simplicity, it does not pretend to be exact by using
complicated hocus-pocus formulas and refinements.
Therefore it is suitable for the very beginners. Otoh it stimulates
imagination and intelligent hand evaluation for players who are at
allmost intermediate level and upwards.
Anton
"David Babcock" <dp...@fastmail.fm> schreef in bericht
news:43fa6629-0d29-445d...@n33g2000pri.googlegroups.com...
David Babcock
> Point count is useful and it has well known flaws. The accuracy is far
> more limited by these flaws than by the way points are assigned to
> highcards.
>
> Therefore, imo, an optimal point-count of high cards and intermediates
> will not improve things significantly because the flaws remain.
I have no disagreement with any of that. Improving things marginally,
even just barely perceptibly, would be highly satisfactory here. :-)
David
Steve Foster
How about:
A 8
K 6
Q 3
J 2
10 1
If you think that undervalues Aces, and overvalues Kings:
A 10
K 4
Q 3
J 2
10 1
If you didn't want to include 10s (though it seems obvious to do so):
A 10
K 5
Q 3
J 2
--
Steve Foster
brsri...@gmail.com
A = 6 p.
K = 4 p.
Q = 2 p.
J = 1 p.
void = 6 p.
singleton = 4 p.
doubleton = 2 p.
working void + super-fit or double-fit = 12 p.
working singleton + super-fit or double-fit = 8 p.
working doubleton + super-fit or double-fit = 4 p.
ace alone in side suite + double-fit = 4 p.
ace alone in side suite opposite partner's singleton in the same suite
+ super-fit = 2 p.
singleton king, queen or jack are worth nothing but you can count the
singleton alone = 4 p.
48 p. + a fit needed for 13 tricks
44 p. + a fit needed for 12 tricks
40 p. + a fit needed for 11 tricks
36 p. + a fit needed for 10 tricks
etc.
Minimum opening bid = 18 p.
Inviting game after a simple two level major suite raise with = 22 p.
Bidding game after a simple two level major suit raise with = 26 p.
Simple enough ?
You can cut the size of those numbers x/2 and you will get:
A= 3
K = 2
Q = 1
J = 0.5
but you will loose the whole number requirement thing. However one
will notice that despite the 0.5 p. given to the jack it is a lot
easier to count by cutting those relatively high numbers to half of
its initial size, as for instance from 6 to 3 for the ace.
Boris
vsp...@hotmail.com
<AntonEnCaroline.dontwants...@planet.nl> wrote:
> I think Point count is ok as it is. It is a huge advantage that by
> virtue of its simplicity, it does not pretend to be exact by using
> complicated hocus-pocus formulas and refinements.
>
> Therefore it is suitable for the very beginners. Otoh it stimulates
> imagination and intelligent hand evaluation for players who are at
> allmost intermediate level and upwards.
>
> Anton
>
Suitable for beginners and most human players. They are
only temporary estimates of trick taking potential to be
reevaluated with every new call. Just not worth the effort.
Now if you're writing a program for a computer, then the
added accuracy would be worth it, since the computer
needs to expend no extra energy for the computations.
Stig Holmquist
In the last two issues of the ACBL Bulletin there are two articles by
M.Bergen about his "Adjust-3" count. He claims the computer count
4.5-3.1.5-0.75-0.25 is most accurate.
He fails to give credit to Alex Martelli who introduced this count six
years ago at RGB under the acronym BUMRAP.
Bergen seems to think that counting fractions is too hard on ordinary
player, so he proposed insted a cumbersome six step method to account
for the T card but keeping the Work count.
I think there is a simpler method using a slightly modified form of
the BUMRAP count. Specifically I propose keeping that count for AK anQ
but upgrading the J to 1 and counting the T as 0.5 by not counting the
first T but only each additional one and also deducting 0.5 for no T,
thus maintaining a 'biblical' total of 40 for the deck.
I tried this out on a sheet of computer print outs made with DMPro at
my local club. I counted manually the points for N and got 357 as also
shown on the print out. The range per deal was 3 to 18. I then counted
using my modified version of BUMRAP and got 365.5 and 375.75 with the
full BUMRAP count. Thus I feel the revised BUMRAP is as good as the
original. I made no attempt to compare it with the Bergen new count,
which I find to bee too convoluted. My simple count yielded the same
results as Bergen got with his illustrations.
No selected unique hand should be used to prove or disprove any
particular method. Only a statistical simulation of at least 1000
deals should be used to compare the four methods above.
I expect my regular detractors will start howling, but I'll not
respond to unidentified posts.
Stig Holmquist
brsri...@gmail.com
Ah, I forgot to tell you.
This hand S xx H xx D Kxxx C xxxxx is not going to be worth much in
4S, I rate this hand to be worth only 4 ( 2 ) of my points, don't give
any points to the short suits if you are short in the trump suit
itself.
Boris
Cheers
Charles Brenner
Evan Bailey is well worth reading. His analysis of point count is at
http://members.cox.net/4evanb/bigclub/appnd1.htm. Quotable line --
"What is done here needs to be done. I believe I do it [analyze the
double-dummy library] correctly. If you started using a reasonable
count, the most improvement you can expect is 1%."
1% in what sense? You'll have to read it.
Charles
Andrew
There are many writers who have already done serious work on the topic
of improving point count.
Thomas Andrews
Tysen Streib
Kurt Schneider
Alex Martelli
Zar Petkov
Virtually everything they have written can be found with a Google
search.
Andrew
olivier....@gmail.com
I can't remember who issued this one a few years ago based on computer
simulations,
but it was very simple:
A : 11 points
K: 7
Q: 4
J: 2
T: 1
The rationale was that the correlation between the count and the
number of tricks
was much higher than the standard 4321 count.
ttw...@att.net
some years ago that purported to show that a 5,4,3,2,1 (A,K,Q,J,10)
was more accurate for Notrump than the Work count.
Goren's version of the Work count seems accurate enough if one
remembers to adjust (maybe drastically) in the light of the bidding.
Combining a simple point count with a simple LTC doesn't do badly.
Stig Holmquist
wrote:
This scale corresponds to 4.4--2.8--1.2--0.8 --0.4 and does not look
like an improvement.
Stig
Michael Angelo Ravera
> On Wed, 12 Nov 2008 13:30:35 -0800 (PST), olivier.couvr...@gmail.com
>
> - Show quoted text -
A=4.0, K=2.8 Q=1.8 J=1.0 T=0.4
To total 25 per suit, you approximate quite well with
10, 7, 4.5, 2.5, 1
The best approximation is to use the Work count and take off a little
when you don't have a ten for every two kings and queens and to add a
little bit when you have 10s that don't have two kings and queens to
go with them.
Using only whole numbers, you could use:
A=20 K=14 Q=9 J=5 T=2 and then divide by 5.
However, the improvements will be slight at best and are only a
starting point and will vary depending upon other information
available to you.
alvin
Olivier:
Although this count does not add to 20 (rarher to 25), it is
reasonable for a start, though the A to Q ratio seems a little high. I
understand that (using decimals) it is closer to 4.5 to 1.9, (or 2.3
to 1 rather than 2.67 to 1). And adding to 25 gives a total HCP of
100, which is easy to work with.
What are the point counts needed for game and slam in this count?
Alvin P. Bluthman
apblu...@aol.com
this...@yahoo.com
>The hypothesis is that such a count might result in
> improving the point-count component of hand evaluation with no more
> mental gymnastics than remembering the new scheme and then dividing by
> 2 to get back to familiar territory.
Since "everyone knows" with the current point count that aces and tens
are undervalued and jacks are overvalued, then certainly a point count
that knows that is an improvement. Let's get rid of the rote
adjustments. Players should only need to do extra work on the
complicated adjustments.
Douglas Newlands
> On Nov 12, 3:53 am, David Babcock <d...@fastmail.fm> wrote:
>> I would be interested in anyone's thoughts on a Work-type point count
>> in which the relevant cards in a suit add to 20 and the values are all
>> whole numbers. The hypothesis is that such a count might result in
>> improving the point-count component of hand evaluation with no more
>> mental gymnastics than remembering the new scheme and then dividing by
>> 2 to get back to familiar territory. ???
>>
>> David
>
> There are many writers who have already done serious work on the topic
> of improving point count.
> Thomas Andrews
> Tysen Streib
> Kurt Schneider
> Alex Martelli
> Zar Petkov
While I am sure all of the above enjoyed playing with this (and no irony
is intended!), I think the unwary should not commit to this without
finding out what "over-fitting" is and what "the no free lunch theorem"
(Shaffer I think) has to say.
A contrasting approach of having multiple, different, simple classifiers
(ensemble classifiers) e.g. decision stumps, might yield better outcomes
and there is plenty of literature on such systems. Any good machine
learning text (Mitchell?) will explain what these are.
regards,
Douglas
(don't have access to my bib files here just now or I would be a bit
more specific about refs)
Thomas Andrews
This is Richard Cowan's result: http://www.maths.usyd.edu.au/u/richardc/bridge.html
(Cowan used to have the entire article on his web site, but it doesn't
appear to be any more.)
It relates to notrump, and only in the case where both hands are
balanced (4333 or 4432.)
I find the result highly suspect after reading the article a few years
ago.
My reading of his article was that he is strictly counting only the
trick-taking value of a card. So Axx is only worth one trick. But in
reality, Axx has its one trick as value, plus the value of stopping
the opponents from taking their tricks. This advantage is quite
huge. The difference betwen Axx and xxx can be 1 trick, or, if
partner has xx, it could be 4, 5, or 6 tricks.