tysen2k Posted June 9, 2004 Report Share Posted June 9, 2004 Okay, prepare yourself for another massive data dump. The table below takes the 13,094 hands posted on my yahoo group, looks at how many Zar points there are and how many tricks are actually taken. I’ve also been able to isolate the situations where the points predict slam, but the partnership is missing some top tricks so that Blackwood, etc. would enable you to stay out of slam. ACTUAL TRICKS TAKEN Zar+fit 9 10 11 12 13 Score Ave 40 1 0 0 0 0 140 140 41 6 0 0 0 0 840 140 42 10 0 0 0 0 1400 140 43 22 0 0 0 0 3080 140 44 36 1 0 0 0 5210 141 45 87 5 0 0 0 13030 142 46 146 12 0 0 0 22480 142 47 195 29 1 0 0 32430 144 48 285 44 5 0 0 48380 145 49 387 91 7 0 0 71050 146 50 473 134 12 0 0 91400 148 51 478 232 34 2 0 113620 152 52 495 297 43 5 0 121740 145 53 526 341 89 5 1 159880 166 54 419 409 111 18 0 209420 219 55 381 404 140 14 0 220350 235 56 286 418 192 22 1 258730 282 57 235 408 255 45 1 296470 314 58 149 312 255 66 2 271040 346 59 105 275 282 86 6 281490 373 60 67 190 256 83 3 233020 389 61 48 199 226 107 15 241890 407 62+ no cntl 58 208 335 1* 0 123430 205 62 4 46 116 111 11 98290 341 63 2 23 99 98 11 94300 405 64 5 19 65 96 18 100810 497 65 0 8 64 97 18 107640 576 66 0 9 28 79 16 89480 678 67+ no cntl 8 20 118 159 0 139920 459 67 0 2 1 28 26 36560 641 68 0 1 5 10 17 23170 702 69 0 1 3 6 26 37560 1043 70 0 0 2 12 14 19940 712 71 0 0 2 6 11 15710 827 72 0 0 1 3 13 19180 1128 73 0 0 1 1 6 8710 1089 74 0 0 0 1 3 4480 1120 75 0 0 0 0 6 9060 1510 76+ 0 0 0 2 4 5940 990 Total 3631270 275 These hands only include those which can take at least 9 tricks. There would be many more hands that take 8 or fewer if this were allowed. The SCORE is the sum total of points that would be won on these hands if bid to the level predicted by the points (not vul). AVE is the average score per hand. So for example, the 52-point Zar hands have 495 going down (-50) + 297*420 + 43*450 + 5*480 = 121740 points or an average of 145 points per hand. Note that if I had allowed hands that can only take 7 or 8 tricks, this score would be even lower as you include hands that go down multiple tricks. I was also generous in allowing the 57-61 point hands to bid only 4M and never 5M. The hands separated out under “62+ no cntl” means that there were 2+ top tricks missing and I assume the pair could stop at 5M. There is one hand in the bunch that has 2 top tricks missing, but the slam still makes since the defense can’t cash them due to blockage. Now let’s look at the numbers for TSP: ACTUAL TRICKS TAKEN TSP+fit 9 10 11 12 13 Score Ave 26 3 0 0 0 0 420 140 27 8 0 0 0 0 1120 140 28 18 0 0 0 0 2520 140 29 39 1 0 0 0 5630 141 30 81 1 0 0 0 11510 140 31 142 8 0 0 0 21240 142 32 226 22 1 0 0 35580 143 33 356 50 2 0 0 58740 144 34 401 82 8 0 0 71680 146 35 502 129 9 0 0 94010 147 36 498 227 22 0 0 112710 151 37 543 308 35 3 0 136070 153 38 523 398 75 5 0 157030 157 39 421 417 104 11 0 206170 216 40 375 460 150 12 0 247710 248 41 240 420 224 19 1 274830 304 42 191 383 229 40 2 274580 325 43 118 340 278 70 1 296110 367 44 81 280 290 70 4 279690 386 45 57 195 275 107 3 255690 401 46 29 142 228 94 5 208460 419 47 22 99 212 105 13 192910 428 48 14 68 168 107 9 159410 436 49+ no cntl 13 51 173 0 0 71400 301 49 0 12 61 110 28 129430 613 50 3 11 50 99 15 104870 589 51 1 10 34 78 23 94470 647 52 2 1 20 71 27 94550 781 53 0 2 10 54 18 70000 833 54+ no cntl 0 5 30 78 0 73440 650 54 2 12 16 22960 765 55 1 7 20 29550 1055 56 1 4 19 28190 1175 57 1 0 2 7 9970 997 58 1 7 10520 1315 59 2 3 4430 886 60 1 0 3 4230 1058 61 2 2 2920 730 62+ 3 4530 1510 Total 3859280 294 TSP scores about 19 points per hand more than Zar. Most of these points come from Zar overbidding on many of the games and slams, even when all the controls are there. I limited the hands (by request) to 9+ tricks because that makes Zar look better. If we lower that requirement Zar looks even worse when it goes down multiple tricks. Tysen Quote Link to comment Share on other sites More sharing options...
tysen2k Posted June 9, 2004 Author Report Share Posted June 9, 2004 I'm just taking a closer look at the data... Looks like others were right when they guessed TSP might be too conservative. Bumping the requirement for game down to 38 TSP gives some more points, bringing the average points up to 21 per hand better than Zar. There aren't enough hands to say what the slam ranges should be; I'll have to look at the larger database. Quote Link to comment Share on other sites More sharing options...
MickyB Posted June 9, 2004 Report Share Posted June 9, 2004 This is great Tysen, just the sort of data I've been waiting to see. With regards to required point counts for game and slam, double dummy analysis cannot be relied on too heavily anyway. Quote Link to comment Share on other sites More sharing options...
inquiry Posted June 10, 2004 Report Share Posted June 10, 2004 Tysen, this is nice data. I wonder if you could edit your data on the yahoo group site to reflect the ZAR fit and TSP fit points you used. I will point out a flaw, however in your data. You assumed all the ocntracts were not vul, of course aggressive bidding pays off most at imps when vul, not vul. If you change the assumption to they were all vul, ZAR points wins. If you change the assumption that half the contracts would be "vul" and the other half not vul, the two system is a virtual tie. To do this, I used for making game, rather than 420, you 520 pts (average of 50% vul, 50% not vul). And for down one, rather than 50 you can use minus 75, etc. If you take the data you posted (which I did), and apply that metric, the difference between TSP + fit and Zar + fit is 1.45 points per board. That is not imps, that is points per board. Now, TSP was still ahead, and maybe deservingly so. But I would like to examine the hands and the way you calculated ZAR and TSP points... and updating your file is the way to allow me us to do that. For example, I would like to look at, say, all the 52 or 53 ZAR point hands to see why your data is in such disagreement with ZAR's. Ben Quote Link to comment Share on other sites More sharing options...
MickyB Posted June 10, 2004 Report Share Posted June 10, 2004 Interesting, Ben. What happens if the target point count for both evaluators is optimised based on this data? Quote Link to comment Share on other sites More sharing options...
inquiry Posted June 10, 2004 Report Share Posted June 10, 2004 I am not sure. The first problem is to see if Tysen applied Zar + fit points correctly. I am willing to make the assumption that he handled his TSP points correctly. So I want to spot check some hands to see how Tysen handled the fit points. If after checking 20 or 40 hands, the fit points are calculated correctly, I am prepared to accept the data as presented, and then try to figure out the ramifications. Oh, and for my recalculation of the data, I used the older 39 TSP for game, not the 38. Ben Quote Link to comment Share on other sites More sharing options...
tysen2k Posted June 10, 2004 Author Report Share Posted June 10, 2004 Tysen, this is nice data. I wonder if you could edit your data on the yahoo group site to reflect the ZAR fit and TSP fit points you used. I will point out a flaw, however in your data. You assumed all the ocntracts were not vul, of course aggressive bidding pays off most at imps when vul, not vul. If you change the assumption to they were all vul, ZAR points wins. If you change the assumption that half the contracts would be "vul" and the other half not vul, the two system is a virtual tie. To do this, I used for making game, rather than 420, you 520 pts (average of 50% vul, 50% not vul). And for down one, rather than 50 you can use minus 75, etc. I just used the simple fit method that Zar uses on his tests. +3 for each extra trump. Maybe extreme, but I wanted to use the same thing that Zar uses. With regard to the vul/not thing. Yes I agree. The best thing to do would actually be to have different point requirements for bidding game depending on vulnerability (say 52 when vul, but 53 when not). However, it also depends on the scoring system since it doesn't pay to be as aggressive at matchpoints. I used total points here since it's something that everyone can actually see. IMPs and MP involve comparisons and can't easily be posted. I pointed out on my original TSP post that the requirements for game, slam, etc. should be modified depending on the vul & scoring system. Quote Link to comment Share on other sites More sharing options...
inquiry Posted June 10, 2004 Report Share Posted June 10, 2004 Tysen, this is nice data. I wonder if you could edit your data on the yahoo group site to reflect the ZAR fit and TSP fit points you used. I will point out a flaw, however in your data. You assumed all the ocntracts were not vul, of course aggressive bidding pays off most at imps when vul, not vul. If you change the assumption to they were all vul, ZAR points wins. If you change the assumption that half the contracts would be "vul" and the other half not vul, the two system is a virtual tie. To do this, I used for making game, rather than 420, you 520 pts (average of 50% vul, 50% not vul). And for down one, rather than 50 you can use minus 75, etc. I just used the simple fit method that Zar uses on his tests. +3 for each extra trump. Maybe extreme, but I wanted to use the same thing that Zar uses. With regard to the vul/not thing. Yes I agree. The best thing to do would actually be to have different point requirements for bidding game depending on vulnerability (say 52 when vul, but 53 when not). However, it also depends on the scoring system since it doesn't pay to be as aggressive at matchpoints. I used total points here since it's something that everyone can actually see. IMPs and MP involve comparisons and can't easily be posted. I pointed out on my original TSP post that the requirements for game, slam, etc. should be modified depending on the vul & scoring system. Well, your Zar Fit points are wrong then. Zar only counts plus three points for fit if hte hand that is doing the counting also has a VOID. You get plus 2 points if the hand doing the counting has a singleton, and plus 1 point if it has a doubleton. This is explained clearly in ZAR's document. So for example if your partner opens 1S and you hold the following hand patterns.. xxxx xxx xxx xxx You get no distributional fit points for the "ninth" spade xxxx xx xxxx xxx You get one point xxxxx xx xxx xxx You get two points (one for the fourth and one for the fifth spade, because you have a doubleton). xxxx x xxxx xxxx You get two zar fit points. The way it sounds like you did it, was give 3 points for example 1 (instead of 0). 3 for example 2 (instead of 1)), 6 for example 3 (instead of 2), and 3 for the last one (instead of 2). If so, you are overestimating a fairly large number of ZAR hands. Essentially you are only RIGHT when the hand has a void. Ben Quote Link to comment Share on other sites More sharing options...
tysen2k Posted June 10, 2004 Author Report Share Posted June 10, 2004 Well, your Zar Fit points are wrong then. Zar only counts plus three points for fit if hte hand that is doing the counting also has a VOID. You get plus 2 points if the hand doing the counting has a singleton, and plus 1 point if it has a doubleton. This is explained clearly in ZAR's document. So for example if your partner opens 1S and you hold the following hand patterns..I know this is the way that Zar fit point are usually counted. I have read the articles. But in all of Zar's computerized tests he uses the "simplified" Zar fit count of a straight +3 per trump. So I'm just doing it the same way he does so we can compare. There is also no bonus for honor's in partner's suit, etc. Quote Link to comment Share on other sites More sharing options...
mikestar Posted June 10, 2004 Report Share Posted June 10, 2004 Tysen is correct. In all of Zar's articles "Zar+Fit" is based on +3 per extra trump. The presumably more accurate count where the addition varies with the length of the short suit is consistently called "Zar Ruffing Power" and he has not used it in his computer studies. Quote Link to comment Share on other sites More sharing options...
inquiry Posted June 11, 2004 Report Share Posted June 11, 2004 Well Tysen maybe right, but that matters not a wit for me. When I do Zar points, I count fit honors, and fit points as he suggested in his article and misfit points the way Eric and he agreed here. I add to that the two quick trick check (absolutely necessary with Zar points that can sore to unbelievable heights). This is the way I check it, and I do the same with TSP. All these comparisions are valid if you are goint to test x versus y only if you use the system being advocated. What good is it to simulate something that is not what is being advocated to use? This is directed to both Tysen and Zar. Tysen has done a lot of good work here with eliminating the hands with two quick losers for slam and one quick loser for grand slam. This is more than ZAR did I think. Just a little more work, and maybe he can have a test bed that really gets somewhere. I can tell a quick screening of the hands with 9 tricks and 52 and 53 Zar points counted his way (3 pts for each extra trump), many of these come in very less than 52. Also, a lot of these are misfit hands, were no fit exist but high ZAR totals show up. But if you start subtracting for less than adequate trump fit, the totals drop just as fast. I guess my liking ZAR is because I use real world hands, and I apply the count the way I think it is advocated. Testing something else with Zar or Tsp, seems not a realistic approach. But as I have said many times, I am not a programmer...so maybe it is just too hard to do it right. Ben Quote Link to comment Share on other sites More sharing options...
tysen2k Posted June 11, 2004 Author Report Share Posted June 11, 2004 Also, a lot of these are misfit hands, were no fit exist but high ZAR totals show up. But if you start subtracting for less than adequate trump fit, the totals drop just as fast.I did subtract 3 points for the Zar hands if there was no 8-card fit. I'm not sure if Zar did this for his studies or not. Quote Link to comment Share on other sites More sharing options...
inquiry Posted June 11, 2004 Report Share Posted June 11, 2004 Also, a lot of these are misfit hands, were no fit exist but high ZAR totals show up. But if you start subtracting for less than adequate trump fit, the totals drop just as fast.I did subtract 3 points for the Zar hands if there was no 8-card fit. I'm not sure if Zar did this for his studies or not. That's another good start. Could you take down the old evaulator comparison zip file (which lacks, for instance TSP), and repost the data with the hands the current Zar+fit and TSP+fit is based upon? Ben Quote Link to comment Share on other sites More sharing options...
Zar Posted June 11, 2004 Report Share Posted June 11, 2004 Sorry I was away for awhile ... I have no idea how we ended up calling Richard Pavlicek's points TSN or TSP or ... Richard adds 1 pt for every suit length above 4 and the 1-3-5 points for shortness, which is exactly what we discuss here (besides the HP + CTRL or 6-4-2-1 for honors). You can download the Richard Pavlicek article from his website www.rpbridge.com and the article has some coded name of “7Z70.pdf”. The problem with the Pavlicek points is that it doesn't reflect the impotence of the flat 4333 pattern. Even Goren subtracts 1 pt for the 4333 shape, making the difference between the 4432 pattern and the 4333 be 2 points, while the difference between 5332 and 4432 is “virtually” non existent, each valued at 1 pt (no pattern actually gets 0 points and everything after that is linear). If you say that the great Goren lived in different times and you happen to rely on the so called trick-taking potential, you'll end up in the same mud-spot: 4333 - 7.76 4432 + 0.31 5332 + 0.06 meaning that the first difference is 0.31 while the second difference is 0.06, highlighting a similar "drop" in value for the 4333. In Zar Points the flat pattern is 2 points below the linear “pack” with 1-point difference which follows after that. Make it a great day: ZAR Quote Link to comment Share on other sites More sharing options...
tysen2k Posted June 11, 2004 Author Report Share Posted June 11, 2004 Richard adds 1 pt for every suit length above 4 and the 1-3-5 points for shortness, which is exactly what we discuss here (besides the HP + CTRL or 6-4-2-1 for honors). You can download the Richard Pavlicek article from his website www.rpbridge.com and the article has some coded name of “7Z70.pdf”. I had no idea this was the case. I derived mine on my own. Zar, do you have a direct link to that article? Pavlicek's website is www.rpbridge.net, (not .com) and I've searched the whole site, but I can't find it. There is a "pavlicek count" mentioned on his site, but that uses shortness points only, not length. Quote Link to comment Share on other sites More sharing options...
tysen2k Posted June 11, 2004 Author Report Share Posted June 11, 2004 That's another good start. Could you take down the old evaulator comparison zip file (which lacks, for instance TSP), and repost the data with the hands the current Zar+fit and TSP+fit is based upon?I will, but not for a few days at least. I just inherited a new project, so I'm pretty busy. Work has to come first sometimes... B) Quote Link to comment Share on other sites More sharing options...
inquiry Posted June 11, 2004 Report Share Posted June 11, 2004 Richard adds 1 pt for every suit length above 4 and the 1-3-5 points for shortness, which is exactly what we discuss here (besides the HP + CTRL or 6-4-2-1 for honors). You can download the Richard Pavlicek article from his website www.rpbridge.com and the article has some coded name of “7Z70.pdf”. I had no idea this was the case. I derived mine on my own. Zar, do you have a direct link to that article? Pavlicek's website is www.rpbridge.net, (not .com) and I've searched the whole site, but I can't find it. There is a "pavlicek count" mentioned on his site, but that uses shortness points only, not length. http://www.rpbridge.net/p/7z70.pdf ... Hey, I am not computer literate, but found it from filename with google, no problems. 3 page document. Ben Quote Link to comment Share on other sites More sharing options...
MickyB Posted June 12, 2004 Report Share Posted June 12, 2004 Richard's count is the same as Tysen's shape count, except hands with two singletons/voids count a point less. Quote Link to comment Share on other sites More sharing options...
tysen2k Posted June 14, 2004 Author Report Share Posted June 14, 2004 http://www.rpbridge.net/p/7z70.pdf ...Ah, I have seen this before. But knowing Richard, this is just what the article says it is: a way to quantify freakness. It's not meant to be a point count. It was never meant to be added to HCP or any derivative of HCP to approximate hand strength. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.