tysen2k

Judgement evolves. Goren may have said 26 back in the 30's, but most people bid them on 25 today. And that 25 includes extra fit points (+2 per extra trump is pretty good, try adding it to your comparison). 24 might be even better, especially since you are only testing vulnerable. If the author says 10 and you find that 12 gives a better score, let the author know. He might thank you.

Okay, I want to make sure I'm counting Zar points the right way... Since there is no bidding in these evaluation methods, how do you determine who is opener and who is responder for the various adjustments? Because there are still a few cases where you get different points for the combined hands depending on who is opener. 1) Ben showed a 5-5 fit example that was solved using Misfit points, but there are still other cases such: AJxxx x xxx xxxx xxxxx xx xxx xxx This only has 2 Misfit points. If South opens, North gets 2 bonus points for honors, plus 4 for the two extra trumps with a singleton. That's 6 bonus points. But if North opens, South only gets 2 bonus points, by either the Misfit method or the supertrumps. So how do you deal with this 4 point difference? 2) You get similar results sometimes when you have two 9-card fits. Who is the opener in that case? 3) Also, why is this pair of hands: KQx xxx xxx xxxx xxxxx xxx xxx xx worth 2 more points than this pair of hands: xxx xxx xxx xxxx KQxxx xxx xxx xx

It says right above it that it is for hands with no fit. Very clear. I agree. I might be able to do this in the next few days, but I really have been spending too much time here when I really should be working... :) But really, I'll see if I can modify my counts so that it uses the right Zar adjustments. Tysen

Use whatever number gives that evaluation method the best score. This should be done for all evaluation methods. Very simple.

I don't remember mentioning HCP anywhere, just points. If I'm wrong, then tell me how you determine if a Goren bidder bids game. Tysen

Yes, if partner had the same number of trumps in both cases. But the real situation is that a 4441 hand opposite a random partner takes about the same number of tricks as a 6331 hand opposite a random partner. The difference is about 0.1 tricks. Zar predicts the difference should be about 0.6 tricks. Poll your data and you'll see. What is the average number of tricks taken when one of the hands is a 4441 hand? Now what is the average number of tricks when one of them is a 6331? We're working with the same DD database from GIB, so our answers should agree. Tysen

The link you provided even gives reference to the 13,000 hands of data that I posted at a yahoo group. That group got canceled since there was no activity for 90 days, but I could create another one. Ben, you looked over those 13,000 hands and said that you felt I had calculated the points correctly (using the +3 points per extra trump method). The reason I used the strict +3 points per trump method was because that was exactly the method Zar was using in his data/calculations at the time. I wanted to use the same method Zar was using in his calculations. He has now repeated his analysis using the more advanced fit calculations. Tysen

I also would like to see a little due dilligence applied to the "accuracy vs. aggression" issue. Zar has done much better in his analysis by looking at IMPs performance for both overbidding and underbidding but I think there is still a major issue. Look at this exerpt from Zar's book on game vs. partscore decisions. http://img202.imageshack.us/img202/324/gamecomparison5vz.png Let's focus on the Goren Points (GP). On the hands game was possible, GP underbid 13,088 times and bid game 2,564 times. On the hands where there was no game, GP bid the partscore 23,000 times and bid game 1,107 times. GP loses 130,000 IMPs on the game hands and only 6,600 IMPs on the partscore hands. This is a huge imbalance (20x more IMPs in underbidding), whereas the Zar is much more evenly distributed. This is because Zar requires the GP bidders to have 26 combined points to bid game. If this were dropped to 25, 24, 23, etc., I bet Goren's performance would improve dramatically. The same applies for BP and WTC which have very different IMP performances in the two categories. The point is, Zar needs to apply due dilligence and give all of the other evaluators the "best" performance possible by selecting the appropriate "point levels" for bidding game and slam for each method. Otherwise you are totally ignoring accuracy and just penalizing based on the levels set by the user. Tysen

This is exactly the place where I was looking at the Misfit points. If you look at my original post, I stated that I took a set of hands that have no 8+ fit. And I did use the entire M4 for these hands. And if you look at the graph I posted, it shows pretty much no correlation between M4 and how many points you should adjust by. Tysen

I've been thinking about this in more detail and I think there are a couple problems with it. It ignores preemption and any ability to make it tough for the opponents. It only looks at the constructive elements. It ignores the level of the opening. If I define 7NT = 10-15, 6+ spades, it would score well. It ignores the fact that finding a fit in a major is more useful than a minor fit. It ignores negative inferences. I like using a 11-15 NT that denies a 4cM. The fact that responder knows for certain that there is no fit is useful information. Some "either or" bids would be unfairly penalized. An artificial 1♦ opening that promises a 4cM is an example. Responder couldn't guarantee a fit unless he has support for both. However, further bidding will reveal a lot more information very efficiently. A multi 2♦ would have the same problem to a lesser extent. Maybe it could work with a couple tweaks. I'd still like to see what you came up with if you can ever find it. Tysen

One of the things I've been thinking about on the back-burner for a while is polishing up my "Shape System" which prioritizes showing shape early. One of the features was the Pass = 0-16 balanced that I discussed here in the forum several months ago. That's a great way to show shape with every bid, including your most common, Pass! My current scheme has: Pass = balanced, 0-16 1♣ = any 4441, any 5440, or any 6+ single suited hand 1♦/1♥/1♠ = natural but always a two-suiter 1NT = balanced, 17-21 2x = EHAA style preempts The 1♣ bid seems a little weird but if you think about it, 1-suiters and 3-suiters should be very easy to separate out in later bidding, even if the opponents compete. In fact, the opponents bidding may clear things up for responder even earlier than without competition. Plus you can now use all NT rebids artificially. The other 1 bids are always 2-suiters so that means you can use both NT rebids as well as same suit rebids artificially. There is absolutely no opening bid that shows just strength but always have some shape component to it. Tysen

I was wondering about this as well since it didn't seem intuitive, but it's actually right (sort of). I looked into it and I forget the exact numbers but it was something like: 9% of hands with 4-4 in spades score the highest in NT 13% of hands with 5-3 in spades score the highest in NT So, it would probably be more accurate to say that 5-3 is more likely to play well in NT, but both belong in suit (usually). Tysen

I am relieved to hear it was a joke too. Sometimes it's hard to tell. But seriously, I would like to see what tests you ran that show how Misfit points help your performance. All I'm saying is that when I ran my numbers it was worse with the misfit points than without. If I see how you did it, then it's quite possible that I can see some sort of mistake I've made. Tysen

So are you saying that you've never actually tested misfit points to see if they work?

That's a tougher question to answer. I think you'd have to use something like GIB's simulation capability to answer it. I was trying to do just that, but I think there is a rule in GIB where if the book bid is 1NT then always bid it without simulating. So it won't simulate to consider other bids. Sorry. Tysen

My simulations show that weak NT is pretty much just as good as strong NT (perhaps this is the reason there is always an argument for which is better?). You have to consider not only the times you open, but the times where you don't open 1N as well. Those negative inferences can be important. What my simulations did show that the weaker your NT gets, the more important it is not to include a 5cM in it. With 15-17 you should include the 5cM, with 12-14 it's about a wash if you include it or not, weaker than that and you're better off not having the 5cM. The weaker you are, the more likely you'll want to be in 2M rather than 1NT. However what's more important than all of these is how it fits into the rest of your system. All of the above is small compared to what you could lose if the rest of the system doesn't mesh with it. I personally use a 11-15 NT with no 4cM. Tysen

Okay, everyone knows that I will have to eventually chime in here... Let me just start off by saying that I've spent a couple hours reading all of Zar's stuff. He has put a lot of effort into this. First let me comment on what Ben and Richard were saying about my comparisons of Zar vs. other systems. The one Richard posted here in this thread was my initial evaluation where it simply compares initial evaluations. Zar doesn't have fit points but neither does any other system. I later posted another one that compares all of them with fit points. Zar did much better, but was still behind BUMRAP+531+2 per extra trump (and behind TSP). R2 Ave Error Score HCP 0.65 1.21 -0.33 Bergen 0.71 1.12 -0.03 HCP + 321 0.71 1.11 0.00 Zar 0.72 1.10 0.04 HCP + fit 0.73 1.07 0.14 BUM RAP + 321 0.73 1.07 0.14 TSP 0.74 1.06 0.20 Zar + fit 0.74 1.05 0.22 BUM RAP + fit 0.75 1.03 0.32 Binky 0.75 1.02 0.33 TSP + fit 0.76 0.99 0.44 Evolved Binky 0.78 0.97 0.54 I also did another study that went into a lot more detail and also took care of the "missing 2 aces" problem. That study was here. Okay, on to Zar's new stuff. I might comment on his bidding system later, but for now I'll just talk about the points since they are seperable. Zar's MISFIT points seemed like an interesting idea, but I did notice something was missing from Zar's work. Zar, where is your data that shows how Zar+Fit/Misfit performs better than simply Zar+Fit? The thing is, when I did a quick study of it's performance, the Zar+Fit/Misfit was actually worse than the Zar+Fit. So I decided to look into the Misfit points. I looked at a bunch of hands that have no 8-card fit. Then I compared M4 to (Unadjusted Zar) - (Tricks * 5) - 2. This number is the amount of points we "should" adjust down for the misfit. For example if the unadjusted Zar total was 60, but we can only take 10 tricks, then we should adjust down by 8 points (60-10*5-2) to get down to 52 points. Zar's Misfit theory says that these two numbers should be about the same since he wants you to subtract the Misfit points if you have no fit. Here is a graph of the two plotted against each other: http://img108.imageshack.us/img108/6449/zar2xp.png It doesn't take a PhD in statistics to know what an R2 of 0.105 means. Misfit points as they are calculated don't have anything to do with how many points you need to downgrade by. Tysen

Just had this hand in a lunchtime game and was unsure about the best play. [hv=d=s&v=n&n=sakqt4haq5dj974ck&s=sj75hkt3da86ca652]133|200|Scoring: Rubber You reach 6♠ with no opposition bidding. Opening lead is a heart.[/hv]

That sounds very interesting. I'd love to hear more if you can dig it up. Tysen

Super-light openings will do a lot better in MPs. The ability to compete and find a fit early on the partscore hands is a tremendous advantage.

Yeah, this was a little contest I ran about a year ago in an effort to actually quantify how good one system is over another. It was a lot of fun and a lot of the participants really enjoyed the experience. Tysen

When I was teaching my class of beginning students, I put a bunch of bridge "rules" on the board. The first one was: 1. Every bridge rule has its exception, including this one.

The given hand is worth 2.57 tricks under Binky. The average 17HCP balanced hand is worth 6.38 tricks. However, based on the studies I've done (Evolved Binky), having 4+ card support for responder gives opener a bonus depending on his shape: 4333 +0.51 4432 +0.55 5332 +0.64 (average = 0.55) So Evolved Binky predicts an average of 2.57+6.38+0.55 = 9.50 tricks. I did a quick simulation and found these actual results: 8 tricks = 16% 9 tricks = 34% 10 tricks = 36% 11 tricks = 12% Which matches pretty much exactly. I couldn't have asked for a better example! Tysen

One thing I've heard of that might be superior to showing a doubleton is the "anti-splinter." Opener bids the suit where he has a concentration of values that would be wasted opposite shortness. This helps partner revalute his holdings a lot more accurately as it allows partner to upgrade with fitting honors and downgrade with shortness. Tysen

1♠. This auction is going to be competitive. It's always to your advantage to describe your hand as quickly as possible in competitive auctions. First-mover advantage. Pass does not describe your hand well. Tysen