relative values of patterns

[jogs]

relative values of patterns

Has anyone ever reported a study of the relative values of patterns?
The five most frequent patterns are:
4432 21.6%
5332 15.5
5431 13
5422 10.6
4333 10.5

I feel confident that 5431 plays better than 5422. About 30 years ago playing Flannery I wanted 4522 to be 12-16 and 4531 to be 11-15. My partner wouldn't have it. Since I had no proof that my suspicions were correct, I dropped the issue. Still I bid 2D with a lower range when 4531 than 4522.

jogs

[Player]

Like your partner I would not have Flannery either. There are far better used for 2D.

[rhm]

Post by jogs
relative values of patterns
Has anyone ever reported a study of the relative values of patterns?
4432 21.6%
5332 15.5
5431 13
5422 10.6
4333 10.5
I feel confident that 5431 plays better than 5422. About 30 years ago playing Flannery I wanted 4522 to be 12-16 and 4531 to be 11-15. My partner wouldn't have it. Since I had no proof that my suspicions were correct, I dropped the issue. Still I bid 2D with a lower range when 4531 than 4522.
jogs

http://bridge.thomasoandrews.com/valuations/original.html#patterns

[jogs]

Post by rhm
http://bridge.thomasoandrews.com/valuations/original.html#patterns

It will take awhile to absorb this. It seems like Andrews is evaluating trick potential from only one hand. There's an auction. Shouldn't there be an attempt to evaluate trick potential from my hand and the partial info from pard's hand?

[rhm]

Post by jogs

Post by rhm
http://bridge.thomasoandrews.com/valuations/original.html#patterns

It will take awhile to absorb this. It seems like Andrews is evaluating trick potential from only one hand. There's an auction. Shouldn't there be an attempt to evaluate trick potential from my hand and the partial info from pard's hand?

It is a triviality that the value of a hand and the value of its feature changes according to what partner holds.
So if you claim in your opening statement "I am confident that 5431 plays better than 5422" that is common wisdom if nothing is known about partners hand in general
It is not hard to show exceptions where 5422 plays better than 5431.
Also distribution has in general much more of an impact in suit contracts than at notrumps.

[f***@googlemail.com]

Post by jogs

Post by rhm
http://bridge.thomasoandrews.com/valuations/original.html#patterns

It will take awhile to absorb this. It seems like Andrews is evaluating trick potential from only one hand. There's an auction. Shouldn't there be an attempt to evaluate trick potential from my hand and the partial info from pard's hand?

So you start by saying that you are confident that a 5431 plays better than a 5422. That is probably true on average although it's easy to construct examples where it isn't the case.

So someone else points at some analysis looking at the value of one hand, and you object to it on the basis that it is looking only at one hand. That is where you started.

I don't understand why you seem to think it is an original idea to suggest that the value of the hand changes over the auction. Yes, if I start with 54 in the majors I'll be happier with 5431 or 5413 than with 5422. If the auction uncontested continues 1S-2C (game forcing) -2H -3C I would much rather have a 5422 than a 5431. if it continues 1S-2D-2H-3D I'd rather be 5431 than 5422.

If the auction starts 1S - 2NT (jacoby) I'd initially rather be either 5431 or 5413. If partner later shows a singleton club then 5422 or 5413 are both better than 5431. So what? This is why most systems have such an emphasis on showing shape.


As rhm says, this is a triviality.

[jogs]

Post by f***@googlemail.com
So you start by saying that you are confident that a 5431 plays better than a 5422. That is probably true on average although it's easy to construct examples where it isn't the case.
So someone else points at some analysis looking at the value of one hand, and you object to it on the basis that it is looking only at one hand. That is where you started.
I don't understand why you seem to think it is an original idea to suggest that the value of the hand changes over the auction.

The object of the game is to win tricks. The initial value of one's hand is determined from summing the HCP of the four suits in my hand. This HCP total is a tool for measuring our tricks. During the auction why are most systems still revising the HCP count?
While trick estimates during an auction is provisional and continually changing, many (too many) systems seem to think hand valuation is about revising point count. While other systems are reevaluating the hand, they should be reevaluating the fit of partnership hands to improve the estimates of partnership tricks. All 2X13 cards of the partnership plays a role in estimating tricks. My approach is different from theirs.
Virtually every system ignores Mike Lawrence's contribution to the theory. Lawrence claims his short suit totals(SST) is a better estimator than trumps. SST augments trumps in estimating tricks. The two parameters together is better than either separately. The second suit is a source for additional tricks. All four suits plays a role in generating tricks. All 26 cards of partnership plays a role in generating tricks.
The original idea is to think in terms of partnership tricks, not partnership high card points.

[t***@att.net]

Perhaps the top 5 opening patterns combined with the top 5 responding patterns, correctly weighted, might be of interest. These 5 patterns if treated independently would combine with the same 5 patterns in the Other Hand to cover about 50% of all patterns.

The bidding could be treated as a re-weighting of the pattern probabilities as it progresses.

[KWSchneider]

Post by jogs
relative values of patterns
Has anyone ever reported a study of the relative values of patterns?
4432 21.6%
5332 15.5
5431 13
5422 10.6
4333 10.5
I feel confident that 5431 plays better than 5422. About 30 years ago play=
ing Flannery I wanted 4522 to be 12-16 and 4531 to be 11-15. My partner wo=
uldn't have it. Since I had no proof that my suspicions were correct, I dr=
opped the issue. Still I bid 2D with a lower range when 4531 than 4522.

Sorry for exploring "trivial" issues, but jog's point is something at which I've been looking for years.
FWIIW, I've done the shape analysis (DD for 44 and for 53 - 100K simulations each, I also have SD sims as well). It depends on how granular you would like to go. For example, you can get progressively more granular, as follows (one hand, trump contracts):

Any pattern -> 3.14 tricks
4trump, any pattern -> 3.96 tricks
4=432 -> 3.86 tricks (note: below average shape for 4trump)
Others:
4=333 -> 3.62
4=441 -> 4.16
4=522 -> 4.00
4=531 -> 4.19
4=540 -> 4.59
4=621 -> 4.24
4=630 -> 4.58

Then you get into the fit conversation:
4any vs 4any -> 8.10 (vs 7.92 additive), so 0.18trick fit benefit for 8combined trump
4=432 vs 4any -> 7.89, bad 44 fit relatively speaking
4=432 vs any 4=432 pattern -> 8.21
4=432 vs specific shapes, ie 4=432, 4=234, 4=324, etc. For example: 4=432 vs 4=432/4=4=342 -> 8.12, 4=432 vs 4=423/4=243 ->8.31, 4=324/4=234 ->8.19

Therefore an average hand with a 4=432 pattern can expect his average partner with a 4card trump fit to deliver 4.03 tricks. If you can narrow partner's pattern to any 4=432, this is worth an additional 0.3tricks, and if you know that the shape is precisely 4=234 opposite your 4=432, you can add an additional 0.1tricks. Ultimately, opposite a 4=432 hand, you can expect a low of 3.72 tricks with a 4=333 dummy, to a high of 5.41 tricks with a 4=360. More simply stated, shape can provide a 1.7trick range, all else equal.

For 5trump...
5trump, any pattern -> 5.22 tricks
5=332 -> 4.86 tricks
Others:
5=422 -> 5.10
5=431 -> 5.30 (answer to jog's question -> approximately 1/2 pt)
5=440 -> 5.84
5=521 -> 5.59
5=530 -> 5.95
5=611 -> 6.00
5=620 -> 6.19

Again, these results are for DD. For SD, there is a measurable difference. Anyone interested in more detail can provide me an email address and I can contact directly offline.

Kurt

[jogs]

Post by KWSchneider

Post by jogs
relative values of patterns
Has anyone ever reported a study of the relative values of patterns?
4432 21.6%
5332 15.5
5431 13
5422 10.6
4333 10.5
I feel confident that 5431 plays better than 5422. About 30 years ago play=
ing Flannery I wanted 4522 to be 12-16 and 4531 to be 11-15. My partner wo=
uldn't have it. Since I had no proof that my suspicions were correct, I dr=
opped the issue. Still I bid 2D with a lower range when 4531 than 4522.

Sorry for exploring "trivial" issues, but jog's point is something at which I've been looking for years.
Any pattern -> 3.14 tricks
4trump, any pattern -> 3.96 tricks
4=432 -> 3.86 tricks (note: below average shape for 4trump)
4=333 -> 3.62
4=441 -> 4.16
4=522 -> 4.00
4=531 -> 4.19
4=540 -> 4.59
4=621 -> 4.24
4=630 -> 4.58
4any vs 4any -> 8.10 (vs 7.92 additive), so 0.18trick fit benefit for 8combined trump
4=432 vs 4any -> 7.89, bad 44 fit relatively speaking
4=432 vs any 4=432 pattern -> 8.21
4=432 vs specific shapes, ie 4=432, 4=234, 4=324, etc. For example: 4=432 vs 4=432/4=4=342 -> 8.12, 4=432 vs 4=423/4=243 ->8.31, 4=324/4=234 ->8.19
Therefore an average hand with a 4=432 pattern can expect his average partner with a 4card trump fit to deliver 4.03 tricks. If you can narrow partner's pattern to any 4=432, this is worth an additional 0.3tricks, and if you know that the shape is precisely 4=234 opposite your 4=432, you can add an additional 0.1tricks. Ultimately, opposite a 4=432 hand, you can expect a low of 3.72 tricks with a 4=333 dummy, to a high of 5.41 tricks with a 4=360. More simply stated, shape can provide a 1.7trick range, all else equal.
For 5trump...
5trump, any pattern -> 5.22 tricks
5=332 -> 4.86 tricks
5=422 -> 5.10
5=431 -> 5.30 (answer to jog's question -> approximately 1/2 pt)
5=440 -> 5.84
5=521 -> 5.59
5=530 -> 5.95
5=611 -> 6.00
5=620 -> 6.19
Again, these results are for DD. For SD, there is a measurable difference. Anyone interested in more detail can provide me an email address and I can contact directly offline.
Kurt

Thank you, Kurt you are providing numbers which are closer to what my intuition was expecting. It is all about the underlying assumptions. Andrews used assumptions which probably wasn't what I was would have used.

[KWSchneider]

Thank you, Kurt you are providing numbers which are closer to what my intui=
tion was expecting. It is all about the underlying assumptions. Andrews u=
sed assumptions which probably wasn't what I was would have used.

AFAI am concerned, Thomas Andrews is the progenitor of bridge simulation - the godfather. I followed his postings from when he first published in the mid-late '90's and was emulating his work in 1999. It was only after his musings about Binky points, that I realized "basic" simulation (combined results for two hands) was of no value. It was not properly scalable or usable at the table.

So the Binky method was the right one. But his SUIT Binky points had a problem in the method of derivation. Andrews never differentiated a trump suit - he simply stipulated that the trump suit was whatever strain yielded the most tricks for a deal. That is why Binky points are not accurate for valuations in a trump contract.

To solve this, I separated trump and offsuit completely, by varying each independently of the other. I simulated hundreds of different scenarios, involving millions of separate deals that separated trump length and honor holdings, from offsuit length and honor holdings. The resultant matrix was much larger and more complex to invert than Andrews, but after solving, it provided:

1) A "base" trick value of every shape for each trump length (I limited hands to maximum 7card long suits to keep this manageable and because this represented over 99% of the possibilities), ranging from 0-7trump. This yielded 70 base values ranging from 0.08 (0=751) to 8.31 (7=510).
2) A normalized adjustment for trump honor holdings. I only considered AKQJT. This yielded 176 values ranging from -2.12 for xxxxxx to 3.02 for AKQJ.
3) A normalized adjustment for offsuit honor holdings. Also AKQJT only. This yielded 176 values ranging from -1.43 for xxxxxxx to 2.24 for AKQ.

By parsing opener's hand and adding these values together based on his/her designated trump suit, a surprisingly accurate estimate of the trick taking potential in that suit was easily available.

But there still was a second problem (which Andrews touches upon briefly). DD is inaccurate - it yields impossible values for hands at the extremes of valuation - like yarboroughs and very strong hands.

So to resolve this, I replicated everything using SD and named it the "UNIFIED" point count, because it allows a pair to estimate their tricks in ALL strains, including their opponent's. This yields a direct calculation of total tricks to implement LoTT, effectively "unifying" valuation and LoTT. While it took much more time (5 years), the results were nonetheless gratifying. Instead of yielding negative values for 3=433 and 4=333 yarboroughs, it yielded 0 tricks for the former and 0.42 tricks for the later. Instead of yielding 6.29 tricks for declarer with a random pair of hands, it yielded 6.91 tricks, a more explainable result, considering 6.5 would be the absolute minimum.

Unfortunately, my job ate into my time over the past 4-5 years and I dropped my work on SD valuation. But now,10 years after starting my SD work, I've resurrected it and I'm currently attempting to finish off the last aspect of SD valuation, namely "the fit". Because the Binky/Unified method works on a hand individually (it's strength), the value of two hands can be aggregated by adding the two Unified values together. But because the Unified method only considers a hand in the absence of all others, there is an additional, minor fit adjustment required that considers the trump fit and the interaction of the shapes of the hands. Correlating this data is my current goal, after which I'll publish the results and the algorithm for table use.

Kurt

[jogs]

On Friday, August 26, 2016 at 1:09:07 PM UTC-7, KWSchneider wrote:
.

Post by KWSchneider
AFAI am concerned, Thomas Andrews is the progenitor of bridge simulation - the godfather. I followed his postings from when he first published in the mid-late '90's and was emulating his work in 1999. It was only after his musings about Binky points, that I realized "basic" simulation (combined results for two hands) was of no value. It was not properly scalable or usable at the table.
So the Binky method was the right one. But his SUIT Binky points had a problem in the method of derivation. Andrews never differentiated a trump suit - he simply stipulated that the trump suit was whatever strain yielded the most tricks for a deal. That is why Binky points are not accurate for valuations in a trump contract.

Isn't Binky points just work count points plus controls? Controls aren't very important in low level contracts. When we have 9+ combined trumps, controls is 'every thing' for slams. 10 1/2(1/2 is the queen of trumps or the 10th trump) makes slam most of the time.

[KWSchneider]

Post by jogs
Isn't Binky points just work count points plus controls?

Not at all - http://bridge.thomasoandrews.com/valuations/additive.html.

Basically you run a restricted DD (or SD) simulation for each holding (trump, offsuit and trump/shape separately) to determine the trick impact associated with that holding. For that specific holding you create a row in a matrix that represents the frequency (probability) of that variable occurring opposite each other holding. Once you fill in this information for every holding, you have a large YxY matrix. If you add the identity matrix to this matrix, and invert the result, you can multiply this matrix by the trick information to yield Binky points, or in my case, Unified Points.

Kurt

[rhm]

Post by KWSchneider

Post by jogs
Isn't Binky points just work count points plus controls?

Not at all - http://bridge.thomasoandrews.com/valuations/additive.html.
Basically you run a restricted DD (or SD) simulation for each holding (trump, offsuit and trump/shape separately) to determine the trick impact associated with that holding. For that specific holding you create a row in a matrix that represents the frequency (probability) of that variable occurring opposite each other holding. Once you fill in this information for every holding, you have a large YxY matrix. If you add the identity matrix to this matrix, and invert the result, you can multiply this matrix by the trick information to yield Binky points, or in my case, Unified Points.
Kurt
--
Posted by Mimo Usenet Browser v0.2.5
http://www.mimousenet.com/mimo/post

Kurt,

I would be interested should your result come to different conclusions than Andrews. (Would be nice if you tabulate your results similar to Andrew, which would make it easy to compare.
I am interested in the practical aspect of hand evaluation.
What is the value of a hand assuming you know little about partners and opponents. I understand there are big variations.
I am less interested in predicting the trick result for 2 hands combined, but on average it is assumed it would be additive.
I have analyzed Thomas Andrews work in some detail and I use a primitive variation of Binky points, but which is usable for me at the table.
I understand valuing suit holdings according to whether they are trumps or in a side suit makes sense, but complicates the matter further.
The question, which is less clear to me, is how much you would gain at the table by this additional complication .

Once I find a fit and a trump suit I tend to switch to my version of NLTC.
This sort of evaluation also gives inside whether a hand is more suitable for notrump or for suit play.
For example AJxx AJxx xxx xx is a better hand than KQJx KQJx xxx xx, particularly when it comes to suit play. At notrumps these 2 hands may be close.

[KWSchneider]

Kurt,
I would be interested should your result come to different conclusions than=
Andrews. (Would be nice if you tabulate your results similar to Andrew, wh=
ich would make it easy to compare.=20
I am interested in the practical aspect of hand evaluation.=20
What is the value of a hand assuming you know little about partners and opp=
onents. I understand there are big variations. =20
I am less interested in predicting the trick result for 2 hands combined, b=
ut on average it is assumed it would be additive. =20
I have analyzed Thomas Andrews work in some detail and I use a primitive va=
riation of Binky points, but which is usable for me at the table.
I understand valuing suit holdings according to whether they are trumps or =
in a side suit makes sense, but complicates the matter further.
The question, which is less clear to me, is how much you would gain at the =
table by this additional complication . =20
Once I find a fit and a trump suit I tend to switch to my version of NLTC.
This sort of evaluation also gives inside whether a hand is more suitable f=
or notrump or for suit play.=20
For example AJxx AJxx xxx xx is a better hand than KQJx KQJx xxx xx, partic=
ularly when it comes to suit play. At notrumps these 2 hands may be close. =

Hi Rainer,

I have written an app that compares multiple methods (K&R, Goren, jogs, Binky, Unified [DD version], NLTC, and soon Ronald Kalf's method, once I complete the algorithm) against DD results for correlation purposes. I can also include SD results but that will skew against Binky, since it was derived specifically from DD data. Unfortunately, in order to compare with actual DD data, you have to determine the Binky and Unified values for both declarer and dummy, and then add them together - which hurts Binky.

Using this additive method over an unrestricted set of 10,000 hands (deal and play), results from Gretl show:

1) DD = 0.9998*Unified-.003 (essentially DD=UN), ADJ R2 = 0.87, SE of Regression = 1.0
2) DD = 1.009*Binky-2.20, ADJ R2 = 0.46, SE = 2.0
3) DD = 0.32 *KR-.559, ADJ R2 = 0.43, SE = 2.1

Observations:
a) Unified passes through (0,0) - the others have a 1.5 to 2 trick negative offset
b) Unified R2 is significantly better than both Binky and K&R
c) Binky is better than K&R
d) Both Binky and Unified have the expected slope of 1.0 against tricks ( 1 trick = 1 trick)
e) K&R has a 0.32 slope -> 3 K&R points = 1 trick

If you'd like to see them simply compared against each other for opener only (but not against DD results), I can do that as well. Let me know privately (same email as when we corresponded 5 years ago) what specifically you would like to see.

From an opening hand perspective only, Binky and Unified compare favorably (more so for obvious trump suits like a 6or7card suit in Binky's case). As the trump suit length gets shorter, Binky over-estimates the value of the hand significantly (generally because there is a longer suit), making the additive results more inaccurate as well.

While this study uses the "exact" Binky and Unified values from tabular results, I also have a simplified model of the Unified Count usable at the table, only requiring weakening the value for offsuit honors, and adding a trump length component. When you consider other "popular" methods like K&R, this is far, far simpler and much more accurate.

Kurt

[jogs]

On Monday, August 29, 2016 at 9:59:29 AM UTC-7,

Post by KWSchneider
Hi Rainer,
I have written an app that compares multiple methods (K&R, Goren, jogs, Binky, Unified [DD version], NLTC, and soon Ronald Kalf's method, once I complete the algorithm) against DD results for correlation purposes. I can also include SD results but that will skew against Binky, since it was derived specifically from DD data. Unfortunately, in order to compare with actual DD data, you have to determine the Binky and Unified values for both declarer and dummy, and then add them together - which hurts Binky.
1) DD = 0.9998*Unified-.003 (essentially DD=UN), ADJ R2 = 0.87, SE of Regression = 1.0
2) DD = 1.009*Binky-2.20, ADJ R2 = 0.46, SE = 2.0
3) DD = 0.32 *KR-.559, ADJ R2 = 0.43, SE = 2.1
a) Unified passes through (0,0) - the others have a 1.5 to 2 trick negative offset
b) Unified R2 is significantly better than both Binky and K&R
c) Binky is better than K&R
d) Both Binky and Unified have the expected slope of 1.0 against tricks ( 1 trick = 1 trick)
e) K&R has a 0.32 slope -> 3 K&R points = 1 trick
If you'd like to see them simply compared against each other for opener only (but not against DD results), I can do that as well. Let me know privately (same email as when we corresponded 5 years ago) what specifically you would like to see.
From an opening hand perspective only, Binky and Unified compare favorably (more so for obvious trump suits like a 6or7card suit in Binky's case). As the trump suit length gets shorter, Binky over-estimates the value of the hand significantly (generally because there is a longer suit), making the additive results more inaccurate as well.
While this study uses the "exact" Binky and Unified values from tabular results, I also have a simplified model of the Unified Count usable at the table, only requiring weakening the value for offsuit honors, and adding a trump length component. When you consider other "popular" methods like K&R, this is far, far simpler and much more accurate.
Kurt

our tricks
E(tricks) = trumps + (HCP-20)/3

This is for the general case.

For any specific board one shifts to using the other parameters in an attempt to count the tricks. SST- short suit totals. ST- source of tricks. Can be from a second suit or long trumps. Interaction of honors and spot cards. This improves the estimates based solely on HCP.

Many of the other point count methods assume players actually know all 2X13 cards of the partnership. Even after a completed auction this is rarely true. My general formula for our tricks only assumes info from the early bids of the auction.

[KWSchneider]

For example AJxx AJxx xxx xx is a better hand than KQJx KQJx xxx xx, partic=
ularly when it comes to suit play. At notrumps these 2 hands may be close. =

10,000 simulations/Spade Contract
AJxx AJxx xxx xx (tests do not eliminate T's)
Binky = 4.34 tricks
UnifiedDD = 4.55 tricks
DD = 7.17 tricks

KQJx KQJx xxx xx ((tests do not eliminate T's)
Binky = 4.06 tricks
UnifiedDD = 4.57 tricks
DD = 7.16 tricks

DD is virtually identical. Unified predicts this. Binky does not.

Kurt

[rhm]

Post by KWSchneider

For example AJxx AJxx xxx xx is a better hand than KQJx KQJx xxx xx, partic=
ularly when it comes to suit play. At notrumps these 2 hands may be close. =

10,000 simulations/Spade Contract
AJxx AJxx xxx xx (tests do not eliminate T's)
Binky = 4.34 tricks
UnifiedDD = 4.55 tricks
DD = 7.17 tricks
KQJx KQJx xxx xx ((tests do not eliminate T's)
Binky = 4.06 tricks
UnifiedDD = 4.57 tricks
DD = 7.16 tricks
DD is virtually identical. Unified predicts this. Binky does not.
Kurt
--
Posted by Mimo Usenet Browser v0.2.5
http://www.mimousenet.com/mimo/post

Kurt,

I agree the difference is small, maybe negligible.
But the vast majority of Bridge player think the KQJ hand is better, which it seems not to be.
Anyway if you could tabulate honor holdings of various length the same way Andrew did (He does it for all holdings down to the 9 and from singleton to 7 card suits) and show their average trick potential

a) in notrump contracts
b) as a trump suit
c) as a side suit

and state your assumptions and methodology.
Andrew did not differentiate b) and c)
That would be very interesting (at least to me) in comparison to Andrews work.

[KWSchneider]

Post by rhm
Kurt,
I agree the difference is small, maybe negligible.
But the vast majority of Bridge player think the KQJ hand is better, which it seems not to be.
Anyway if you could tabulate honor holdings of various length the same way Andrew did (He does it for all holdings down to the 9 and from singleton to 7 card suits) and show their average trick potential
a) in notrump contracts
b) as a trump suit
c) as a side suit
and state your assumptions and methodology.
Andrew did not differentiate b) and c)
That would be very interesting (at least to me) in comparison to Andrews work.

Rainer,

a) for DD notrump, my results will be identical to TAndrews (same method). I haven't completed SD.
b) for DD suit, I believe that I sent you this 4-5 years ago. I'll recheck. Maybe only for SD...
c) Same as b).

For suit contracts, I did not do down to the 9 for a number of reasons - for offsuit, the T is already virtually worthless. For trump, the 9 is valuable (~0.1 trick) but it increases the complexity of effort by adding a complete cycle of additional simulations (9, A9, K9, Q9, J9, T9, A9x, K9x, etc - around 100 more), and I had no intention of using 9's in my table method.

Kurt

[jogs]

Post by KWSchneider
For suit contracts, I did not do down to the 9 for a number of reasons - for offsuit, the T is already virtually worthless. For trump, the 9 is valuable (~0.1 trick) but it increases the complexity of effort by adding a complete cycle of additional simulations (9, A9, K9, Q9, J9, T9, A9x, K9x, etc - around 100 more), and I had no intention of using 9's in my table method.
Kurt

Value of tens and nines.
While interesting in theory, in practice this is mostly useless info.

In the short secondary suits queens and jacks are often worthless.

AQ //// xx
Instead of thinking of the value of the queen, think in terms of the AQ combination. The AQ combination has value.
Ax //// Qx
Seems doubtful that the queen would win a trick very often. The opponent with the king must lead the suit. That occurs on the opening lead or if that player can be endplayed.

[s***@gmail.com]

Post by KWSchneider

For example AJxx AJxx xxx xx is a better hand than KQJx KQJx xxx xx, partic=
ularly when it comes to suit play. At notrumps these 2 hands may be close. =

10,000 simulations/Spade Contract
AJxx AJxx xxx xx (tests do not eliminate T's)
Binky = 4.34 tricks
UnifiedDD = 4.55 tricks
DD = 7.17 tricks
KQJx KQJx xxx xx ((tests do not eliminate T's)
Binky = 4.06 tricks
UnifiedDD = 4.57 tricks
DD = 7.16 tricks

Are others also shocked that AJxx is as good as, or
even better than KQJx?
I would have thought KQJx better, especially at NT.

Is this an artifact of the methodology? For example,
AJxx might be worth more at double dummy than single dummy?

Have those acquainted with this result adjusted their
bidding to re-evaluate such holdings?

Septimus

[Steve Willner]

Post by KWSchneider
AFAI am concerned, Thomas Andrews is the progenitor of bridge
simulation - the godfather. I followed his postings from when he
first published in the mid-late '90's

I think Alex Martelli (the "M" in BUMRAP) might have preceded Thomas,
but Thomas' work on hand evaluation was far more comprehensive. Matt
Ginsberg I think predated Alex, but I don't think Matt applied his
results to hand evaluation. Later Tysen Streib did quite a lot.

The upshot of Tysen's work seemed to be that the Goren (really William
Anderson) count is very good, but you can improve it by making small
adjustments. I think the largest of those is subtracting half a point
for 4441 shape.

There has been other simulation work that's just wrong: wrong
assumptions or wrong evaluation of results. Some of that has picked up
followers for no very good reason.

Kurt: you seem to have gone beyond others. I'll be very interested in
how your results compare to Goren's. I'd expect you to be better, but
the question is how much, especially after those minor adjustments.

[Thomas Andrews]

Post by Steve Willner

Post by KWSchneider
AFAI am concerned, Thomas Andrews is the progenitor of bridge
simulation - the godfather. I followed his postings from when he
first published in the mid-late '90's

I think Alex Martelli (the "M" in BUMRAP) might have preceded Thomas,
but Thomas' work on hand evaluation was far more comprehensive. Matt
Ginsberg I think predated Alex, but I don't think Matt applied his
results to hand evaluation. Later Tysen Streib did quite a lot.

For what it's worth, no, Martelli didn't precede me. I think I get a mention in one of his first articles.

[Thomas Andrews]

Post by Thomas Andrews

Post by Steve Willner

Post by KWSchneider
AFAI am concerned, Thomas Andrews is the progenitor of bridge
simulation - the godfather. I followed his postings from when he
first published in the mid-late '90's

I think Alex Martelli (the "M" in BUMRAP) might have preceded Thomas,
but Thomas' work on hand evaluation was far more comprehensive. Matt
Ginsberg I think predated Alex, but I don't think Matt applied his
results to hand evaluation. Later Tysen Streib did quite a lot.

For what it's worth, no, Martelli didn't precede me. I think I get a mention in one of his first articles.

And yes, I'm up too late googling my name. :)

In my defense, I was looking for an old post.

--
Thomas Andrews

[a***@hotmail.com]

Post by jogs
relative values of patterns
Has anyone ever reported a study of the relative values of patterns?
4432 21.6%
5332 15.5
5431 13
5422 10.6
4333 10.5
I feel confident that 5431 plays better than 5422. About 30 years ago playing Flannery I wanted 4522 to be 12-16 and 4531 to be 11-15. My partner wouldn't have it. Since I had no proof that my suspicions were correct, I dropped the issue. Still I bid 2D with a lower range when 4531 than 4522.
jogs

S Garton Churchill gave important discourse on hand pattern in his book The Churchill Style....

His perspective was from the pov of effective bidding- as in the likelihood of finding a profitable fit. for instance 5440 > 6430 >5431 >4441 ...

regards
axman

[jogs]

Post by a***@hotmail.com
S Garton Churchill gave important discourse on hand pattern in his book The Churchill Style....
His perspective was from the pov of effective bidding- as in the likelihood of finding a profitable fit. for instance 5440 > 6430 >5431 >4441 ...
regards
axman

That was over 50 years ago. Churchill didn't have the computer power to solve questions on relative values of hand patterns.

[Sandy Barnes]

Post by jogs
relative values of patterns
Has anyone ever reported a study of the relative values of patterns?
4432 21.6%
5332 15.5
5431 13
5422 10.6
4333 10.5
I feel confident that 5431 plays better than 5422. About 30 years ago playing Flannery I wanted 4522 to be 12-16 and 4531 to be 11-15. My partner wouldn't have it. Since I had no proof that my suspicions were correct, I dropped the issue. Still I bid 2D with a lower range when 4531 than 4522.
jogs

This sparks a question, anyone use Leaping Flannery? If yes, how well does it work?