BBO Discussion Forums: Point counting method - BBO Discussion Forums

blackshoe, on 2012-June-19, 10:23, said:

I did not mean to be difficult but Siegmund had already posted a link. Tysen has also done some very interesting analysis, unfortunately he doesn't post here anymore.

I looked back at Siegmund's post before I posted that. I didn't see a link. It's there now. Perhaps I'm going blind.

I did read Tysen's original three articles from r.g.b., and saved the third one, which has the practical point count method he derived. He said he intended a fourth article, but I couldn't find it. Perhaps he never got around to writing it.

Earlier you mentioned the search for the optimal point count method. According to Tysen, Binky Points (from Thomas Andrew's original work) are that method, but Binky Points are too complicated for this tired old brain to use at the table, so (again according to Tysen), that makes TSP the optimal method (6421 plus a few adjustments, including 1 point for a suit with two honors, which is the only place where tens get any love). Do you agree? If not, is there an optimal method?

When I fitted the logistic regression model
(3NT makes) ~ #aces + #kings + #queens + #jacks
based on the GIB DD library I was chocked to see how accurate it MW is.

If all the information we have is the combined number of aces, kings, queens and jacks in the two hands, there is very little scope for improvement.

But of course one can come up with better methods if one adds more information. Shape, location of honours, collaborating honours, spots, rightsiding.

I use MW until a major suit fit has been found. And then I switch to gut feelings or some kind of modified LTC.

helene_t, on 2012-June-20, 09:21, said:

Given the size of the data set, you could probably introduce additional terms without much risk of overfitting.

For example, treat the Ace of Spade, Ace of Hearts, Ace of Diamonds, ... Jack of Clubs as separate predictors.
Add cross terms for Ace of Spades + King of Spades and the like...
Perform the same logistic regression model...

The new GeneralizedLinearModel.stepwise method in the 12b release of Statistics Tbx will automatically check which combinations are statistically significant.

(Regretfully, I don't have access to the products any more or I'd run this myself)

Statto, on 2012-June-17, 23:01, said:

Regardless of whether it is good or bad, I am always bemused when people criticise an argument without even having read the book. It is hard to take such a post seriously.

You are easily bemused!

the hog, on 2012-June-20, 12:42, said:

han, on 2012-June-20, 15:07, said:

LOL

Well if I said player x is terrible. Why? Because player Y told me so. Is that reasonable even if I have never played with or against player X or watched him play or even studied his results? Statto's comments lack the same credibility.

CSGibson, on 2012-June-18, 14:44, said:

You have more fingers and toes than I do. Even using the extra available digit only gets me to 20 1/2.

aguahombre, on 2012-June-20, 20:55, said:

Take your right hand and for 1-5 open the fingers as normal. For 6-9 open the fingers with the middle knuckle bent. For 10 open a finger on the left hand. Reapeat the steps on the right hand for 11-19. At 50 you can repeat for the left hand with a bend in the knuckle. This gets you to 100 which is probably enough unless you are using a pretty special form of point count.

Zelandakh, on 2012-June-21, 00:46, said:

When I was on Spring Break 3 years ago, I went to the library, read many books, and even checked out several books (this was when I just started learning how to play bridge). A book I read was an entire overview of Mathematics, and they had a simple way to count to 99. Your right thumb was worth 5, any other finger on your right hand was worth 1. Your left thumb was worth 50, and any other finger on your right was worth 10. This was supposed to be useful for younger kids, especially for purposes of adding or subtracting.

the hog, on 2012-June-20, 20:05, said:

But this is not at all the same. This is more like saying, player Y is terrible, here, take a look at his records, he always finished last. Of course you might say, if someone always finished last, it could be that he always plays with very bad partners and is also extremely unlucky. Still, if he asked me to form a regular partnership, I'd think twice before doing so.

We know what the Banzai method is. Point counting methods have been investigated very thoroughly and every serious statistical investigation has pointed out that the Banzai method is poor, poorer than the standard 1-2-3-4 count. Then why read a book recommending this method?

Please don't take critique on the Banzai method personally.

chasetb, on 2012-June-21, 01:15, said:

The method I gave was one I used as a kid in the days before calculators were allowed in the classroom. I did not read any books to come up with it. There are better methods though, I agree, although I would say that my idea is better than the tally method for adding/subtracting because you can move individual fingers rather than having to convert the thumb mid-calculation. For computer freaks and mathematicians, using the fingers for binary is cool (right thumb = 1; right fingers = 2/4/8/16; left fingers = 32/64/128/256; left thumb = 512). That allows you to add up to 1023. I doubt such a method is practical for most though. If you were really anal and combined that with my 3 state per finger idea you could use the fingers and thumbs as base 3 - that would get you close to 60000!

Statto, on 2012-June-17, 22:03, said:

Agree. The idea that QJT9 and AK32 both equal two tricks is fairly hopeless as a basis of hand evaluation. Tempo matters.

han, on 2012-June-21, 01:47, said:

I am not taking it personally, Han. Why should I? It has nothing to do with me. I doubt whether Statto has investigated it very thoroughly; perhaps he has, so I am wrong, but I doubt it.. However this sort of comment often appears. Perhaps a better analogy would have been someone attacking Dawkin's views without ever having read any of his bbooks. (And this does happen frequently).

the hog, on 2012-June-21, 04:14, said:

What was wrong with han's analogy which seems to be much more persuasive. if we take Dawkins' views as our anaology instead, the equivalent would be that scientific evidence proved to a high level of statistical significancy that, for example, creationism was true and that this result was confirmed by mulriple scientific studies and generally accepted by the scientific community. In this case one might be willing to dismiss Dawkins' views without having read his books, at least until such time as further evidence came along to suggest that this position was incorrect.

Since there is no such scientific consensus, the comparison between rejecting Banzai points and standing against modern evolutionary theory is at the very least highly misleading, and probably closer to trolling than anything that should be taken seriously.

There is a guy emailing me (and countless other mathematicians) short proofs of two very famous open problem every couple of months. He proofs both of them in about 2.5 pages. I confess I dismiss them without reading them. However, I am doing so on much flimsier grounds than dismissing Banzai count.

I am actually a fan of using statistical methods to disprove conventional wisdom. The evidence that NFL coaches are way too conservative in their play-calling (especially on the decision whether to punt on 4th down) is extremely strong. It seems to have taken longer than necessary in the NBA until teams systematically started trying to play for more three-point shots. I loved everything Tysen wrote (one thing I will always remember is that he convinced me that conventional preempting wisdom doesn't sufficiently take shortness in the majors into account). But there is also a lot of rubbish justified by rubbish statistics - whether it's measuring individual defense in the NBA via statistics or analyzing pass-versus-run calls in the NBA (as these don't seem to be able to distinguish between original playcalls and reactions to defense positioning).

It's pretty easy to tell which side Banzai points belong to, either by looking at their justification or by looking at the end result.

helene_t, on 2012-June-21, 02:05, said:

Very good summary of the main issue.

han, on 2012-June-21, 01:47, said:

Show me where all these "serious statistical investigations" are. Yes, I've read Thomas Andrews' and Tysen Streib's articles on r.g.b., but you imply there is much more out there. Where is it? Where, for that matter, are the articles (other than those I've just mentioned) proving that the method presented in this book (not the original paper on which it's based) — published all of two years ago, btw — is so terrible?

Somebody upthread wondered why a man with Ron Klinger's reputation would co-write a book on such a flawed method. One might also ask why Eric Kokish would write an introduction praising it. Has anyone asked them why they did these things?

If Banzai points are that bad, fine, I'll burn the damn book and mail the ashes to Klinger. But I'm not going to do that based on some vague "I don't have to read the book to know it's wrong". Show me the evidence. And if you tell me "screw you, go find it yourself," well, so be it, but don't be surprised if I take your unsupported opinion with a pound or three of salt.

I wouldn't burn books or confronting authors with the notion that their books are bad. What I would recommend is something far less dramatic: remember that aces are worth a little more than 4, not less.

(Note that the 5-4-3-2-1 count is equivalent to 3 1/3 for aces, 2 2/3 for kings, 2 for queens, 1 1/3 for jacks and 2/3 for tens.)