[catch22]

I need a way, from a computer program, to determine certain characteristics of a hand. In particular,

1. Does a hand contain scattered or concentrated values. KQ2 54 52 AQJT32 vs Q32 K2 Q2 AJT532
2. Is a suit a good quality or poor quality. e.g. Q97432 vs AQ9765

I could have a good stab at defining this for myself, but would prefer to use existing evaluation methods if they do exist

Does anyone know if there are any algorithms to analyse these types of characteristics and if they exist are they written up anywhere.

[matmat]

i think that the kaplan-rubens hand evaluator does something like this:
http://www.gg.caltec...u/~jeff/knr.cgi

i believe it was originally written up in the Bridge World.
a little more info on the method is here:
http://www.gg.caltec...u/~jeff/knr.txt

[H_KARLUK]

Quite simple.

Thinking process is different than any human. Software using a simulation. Unknowns constructed consistent. Each analyze double dummy.

Here's an example of 2008 World Computer Bridge Championship :

Contract 6 ♥ , ♠ Q led

---

WBridge5 won at hand. ♥ played to Ace. Then both hands low ♥. Duck !
Now ♦ ten. ♦ Ace, ♥.

5 cards ending :

Contract 6 ♥ , ♠ Q led

---

♠A squeezed W. ♦ discard bad, obligatory ♣ thrown.
Now time to squeeze East. ♦KQ ! Must keep ♠ knave, already pitched ♣. ♣AT good.
+1430

You can try http://www.automaton.../en/OddsTbl.htm

[matmat]

H_KARLUK, on Dec 30 2008, 10:19 AM, said:

Quite simple.

Thinking process is different than any human. Software using a simulation. Unknowns constructed consistent. Each analyze double dummy.

Irrelevant to the question asked.

[ASkolnick]

1. Yes, you can create an alfgorithm. The dealer program (by Hans Van Sternen) is the one I love. It uses a function Quality to evaluate each suit and CCCC for the overall. This uses Kaplan-Reubens evaluation.

To evaluate "scattered values" versus "concentrated", once could take the standard deviation of the combined 4 suit Quality. Where your breakpoint of what you consider "scattered" I don't know.

[helene_t]

Something that is relatively easy to do is a statistical analysis of large bridge result databases.

I made a logistic regression analysis of some 100.000 DD results in which there was no major suit fit. The optimal coefficients of the honours turned out to be
4.2
3.0
1.8
1.0

The model was
logit(p("3 NT makes")) = a #aces + b#kings + c#queens + d#jacks

Something similar could be made involving honor position and shape.

[matmat]

you could also always calculate ZARs

[TimG]

helene_t, on Dec 30 2008, 01:44 PM, said:

Something that is relatively easy to do is a statistical analysis of large bridge result databases.

I made a logistic regression analysis of some 100.000 DD results in which there was no major suit fit. The optimal coefficients of the honours turned out to be
4.2
3.0
1.8
1.0

The model was
logit(p("3 NT makes")) = a #aces + b#kings + c#queens + d#jacks

Something similar could be made involving honor position and shape.

Helene,

I wonder why you precluded a major suit fit, but not a minor suit fit. Presumably your are counting tricks and it shouldn't matter, for evaluation purposes, whether those tricks are won by minor suit cards or major suit cards.

I'd love to see how this breaks down by honor combination.

Tim

[MarkDean]

helene_t, on Dec 30 2008, 10:44 AM, said:

Something that is relatively easy to do is a statistical analysis of large bridge result databases.

I made a logistic regression analysis of some 100.000 DD results in which there was no major suit fit. The optimal coefficients of the honours turned out to be
4.2
3.0
1.8
1.0

The model was
logit(p("3 NT makes")) = a #aces + b#kings + c#queens + d#jacks

Something similar could be made involving honor position and shape.

This is an interesting idea. I wonder if Q's are found to be overvalued due to DD analysis: opposite ATx, KJx is just as good as KQx (well unless there is an entry issue).

[TimG]

MarkDean, on Dec 30 2008, 02:06 PM, said:

helene_t, on Dec 30 2008, 10:44 AM, said:

Something that is relatively easy to do is a statistical analysis of large bridge result databases.

I made a logistic regression analysis of some 100.000 DD results in which there was no major suit fit. The optimal coefficients of the honours turned out to be
4.2
3.0
1.8
1.0

The model was
logit(p("3 NT makes")) = a #aces + b#kings + c#queens + d#jacks

Something similar could be made involving honor position and shape.

This is an interesting idea. I wonder if Q's are found to be overvalued due to DD analysis: opposite ATx, KJx is just as good as KQx (well unless there is an entry issue).

Yes, it would be fascinating to see the analysis done on a large set of deals from BBO play, both on a DD and an actual result basis. Then compare the results.

[hotShot]

There is an old thread where tysen2k discusses a lot of this.
He posted most of this in rgb too.

[P_Marlowe]

helene_t, on Dec 30 2008, 01:44 PM, said:

Something that is relatively easy to do is a statistical analysis of large bridge result databases.

I made a logistic regression analysis of some 100.000 DD results in which there was no major suit fit. The optimal coefficients of the honours turned out to be
4.2
3.0
1.8
1.0

The model was
logit(p("3 NT makes")) = a #aces + b#kings + c#queens + d#jacks

Something similar could be made involving honor position and shape.

It was a good analysis, but just proved one thing:

The 4321 count is fairly good, because you also mentioned
that the likelhood of correct prediction of making 3NT
only increase by a small degree, compare with the 4321
count.

Which is not surprising: The 4321 count was developed for
balanced hand, and there was statistical / propability analysis
involved in the developement of the 4321 count.

With kind regards
Marlowe