Tests for a double-dummy solver

[tysen2k]

For those interested, I have the chance that the opps have a game as well.

 

Shape   Game?    Error Opps?    Error
2-2=5-4  24%     0.3%   35%     0.3%
3-2=4-4  24%     0.2%   32%     0.2%
3-3=4-3  25%     0.2%   31%     0.2%
4-3=3-3  25%     0.2%   29%     0.2%
3-2=5-3  26%     0.2%   33%     0.2%
4-2=4-3  26%     0.1%   30%     0.1%
3-3=5-2  27%     0.2%   32%     0.2%
4-3=4-2  27%     0.1%   28%     0.1%
2-2=6-3  27%     0.4%   38%     0.4%
4-4=3-2  28%     0.2%   26%     0.2%
3-2=6-2  28%     0.3%   36%     0.3%
4-2=5-2  28%     0.2%   32%     0.2%
5-2=3-3  28%     0.2%   31%     0.2%
5-3=3-2  28%     0.2%   29%     0.2%
5-2=4-2  29%     0.2%   31%     0.2%
3-1=5-4  29%     0.3%   33%     0.3%
2-2=7-2  29%  *  0.7%   41%  *  0.8%
2-1=6-4  30%     0.4%   38%     0.5%
2-1=5-5  30%  *  0.5%   35%  *  0.5%
4-1=4-4  31%     0.3%   30%     0.3%
3-1=6-3  31%     0.4%   36%     0.4%
5-4=2-2  31%     0.3%   27%     0.3%
2-1=7-3  32%  *  0.7%   41%  *  0.7%
4-1=5-3  32%     0.3%   32%     0.3%
3-3=6-1  32%  *  0.5%   31%  *  0.5%
6-2=3-2  32%     0.3%   34%     0.3%
6-3=2-2  33%     0.4%   32%     0.4%
4-3=5-1  33%     0.3%   28%     0.3%
5-1=4-3  34%     0.3%   31%     0.3%
2-2=8-1  34%  *  2.2%   46%  *  2.3%
3-1=7-2  34%  *  0.7%   42%  *  0.7%
4-4=4-1  34%     0.3%   25%     0.3%
5-3=4-1  35%     0.3%   27%     0.3%
3-2=7-1  35%  *  0.7%   37%  *  0.7%
4-2=6-1  35%     0.5%   32%     0.4%
4-1=6-2  35%     0.4%   35%     0.4%
5-4=3-1  37%     0.3%   25%     0.2%
2-1=8-2  37%  *  1.6%   47%  *  1.6%
6-1=3-3  37%  *  0.5%   35%  *  0.5%
6-3=3-1  38%     0.4%   29%     0.4%
5-2=5-1  38%     0.4%   30%     0.4%
7-2=2-2  38%  *  0.8%   37%  *  0.8%
1-1=6-5  38%  *  1.2%   40%  *  1.2%
5-1=5-2  39%     0.4%   33%     0.4%
3-0=5-5  39%  *  1.1%   33%  *  1.0%
6-1=4-2  40%     0.5%   35%     0.5%
1-1=7-4  40%  *  1.7%   40%  *  1.7%
6-2=4-1  41%     0.5%   32%     0.4%
3-0=6-4  41%  *  0.9%   36%  *  0.9%
4-0=5-4  42%  *  0.6%   31%  *  0.6%
6-4=2-1  43%     0.5%   28%     0.4%
4-0=6-3  43%  *  0.9%   33%  *  0.8%
3-0=7-3  43%  *  1.4%   38%  *  1.4%
7-2=3-1  43%  *  0.7%   37%  *  0.7%
7-1=3-2  43%  *  0.7%   38%  *  0.7%
5-5=2-1  43%  *  0.6%   26%     0.5%
7-3=2-1  44%  *  0.8%   34%  *  0.7%
5-0=4-4  44%  *  0.9%   31%  *  0.8%
4-1=7-1  44%  *  1.2%   36%  *  1.1%
3-3=7-0  44%  *  2.0%   29%  *  1.8%
4-4=5-0  45%  *  0.9%   24%  *  0.8%
3-1=8-1  45%  *  2.1%   44%  *  2.1%
2-0=7-4  45%  *  1.7%   38%  *  1.7%
2-0=6-5  45%  *  1.3%   36%  *  1.2%
4-3=6-0  45%  *  0.9%   28%  *  0.8%
5-0=5-3  46%  *  0.8%   32%  *  0.7%
1-1=8-3  46%  *  2.9%   44%  *  2.9%
3-0=8-2  47%  *  3.1%   46%  *  3.1%
5-3=5-0  48%  *  0.8%   26%  *  0.7%
5-1=6-1  48%  *  0.9%   35%  *  0.8%
5-4=4-0  49%  *  0.7%   23%  *  0.5%
2-0=8-3  50%  *  3.1%   42%  *  3.1%
6-0=4-3  50%  *  0.9%   31%  *  0.8%
6-3=4-0  51%  *  0.9%   26%  *  0.8%
4-2=7-0  51%  *  1.7%   32%  *  1.6%
3-2=8-0  51%  *  3.2%   36%  *  3.1%
6-1=5-1  51%  *  0.9%   35%  *  0.8%
8-1=2-2  52%  *  2.2%   45%  *  2.2%
4-0=7-2  52%  *  1.7%   35%  *  1.6%
5-5=3-0  52%  *  1.1%   22%  *  0.9%
6-4=3-0  52%  *  0.9%   25%  *  0.8%
8-2=2-1  53%  *  1.7%   42%  *  1.7%
5-2=6-0  53%  *  1.3%   31%  *  1.2%
7-0=3-3  54%  *  1.9%   38%  *  1.9%
5-0=6-2  55%  *  1.3%   36%  *  1.2%
1-0=7-5  55%  *  3.0%   41%  *  3.0%
6-2=5-0  56%  *  1.3%   27%  *  1.1%
7-3=3-0  57%  *  1.4%   29%  *  1.3%
6-0=5-2  57%  *  1.3%   35%  *  1.2%
7-1=4-1  57%  *  1.2%   35%  *  1.1%
4-1=8-0  57%  *  4.6%   43%  *  4.6%
1-0=6-6  58%  *  3.7%   44%  *  3.7%
6-5=1-1  58%  *  1.2%   27%  *  1.1%
7-4=1-1  59%  *  1.6%   30%  *  1.5%
4-0=8-1  59%  *  4.7%   36%  *  4.6%
7-0=4-2  60%  *  1.7%   34%  *  1.6%
7-2=4-0  60%  *  1.7%   31%  *  1.6%
8-0=3-2  60%  *  3.1%   43%  *  3.1%
8-1=3-1  61%  *  2.1%   41%  *  2.1%
6-5=2-0  61%  *  1.2%   25%  *  1.1%
1-0=8-4  63%  *  4.7%   48%  *  4.9%
7-4=2-0  63%  *  1.6%   27%  *  1.5%
8-2=3-0  63%  *  3.0%   35%  *  3.0%
5-0=7-1  64%  *  2.8%   42%  *  2.9%
8-3=1-1  65%  *  2.7%   32%  *  2.6%
5-1=7-0  65%  *  3.0%   33%  *  3.0%
8-3=2-0  67%  *  2.9%   36%  *  3.0%
6-1=6-0  70%  *  2.5%   35%  *  2.5%
7-0=5-1  70%  *  2.8%   37%  *  3.0%
6-0=6-1  71%  *  2.4%   38%  *  2.6%
7-5=1-0  72%  *  2.9%   30%  *  3.0%
8-0=4-1  73%  *  4.3%   41%  *  4.8%
7-1=5-0  73%  *  2.8%   35%  *  3.0%
8-1=4-0  74%  *  4.0%   44%  *  4.5%
8-4=1-0  75%  *  4.3%   28%  *  4.4%
6-6=1-0  83%  *  3.1%   24%  *  3.5%

 

There is less of a dependence on overall shape, but still a major factor if we have major cards or not.

 

Is this a case against some assumed-fit preempts that show both majors? The opps are much less likely to have a game that we need to sacrifice against.

[han]

Very nice, once again. It also deals with Fluffy's conjecture that 4333's make it more likely that the oponents have game. I celebrate this as a personal victory, WINE!

 

However, it doesn't look like my prediction [more distributional -> more likely opponents have game] is true either, look for instance at 6-6-1-0, just 24%. However, it does show that the more cards we have in the minors, the more likely it is that they have game. Is this an argument for opening light with cards in the minors, or at least for preempting a lot with cards in the minors? I think so.

 

Perhaps Jlall is right and bridge really is more complicated, sigh.

[Echognome]

Yes Tysen. Very nice work indeed.

 

To test the assumed fit theory, it would be great to have a common range in as well (since we are only preempting on weak hands). Thus limiting our table to hands that are 4-4=3-2, 5-4=3-1, 5-4=2-2 in the 5-9 range. (4-4=4-1, 5-5=2-1, and 5-5=3-0 are also commonly included) Maybe one with only the first hands and one with all of them?

 

Along those lines, it would be really interesting to see the "is a weak 2♥ really preemptive?" debate have some light shed on it. Having some 2♠ and 2♥ openings done with limited ranges would also be interesting.

 

Would be nice to know how 'effective' preempts are and how much uncertainty we are putting our opponents under. The next step would be to vary the ranges of the preempts until we made their game prospects low enough that they had difficulty. For example, I used to play an intermediate 2♥/2♠ and really felt they were IMP winners when they came up, but dropped them due to frequency.

[tysen2k]

This line of study has also made me to start thinking about what hands make the best preempts. Here's something I was thinking about and let me know if you think it's the right track.

 

If you look at the par contract for a DD deal, there is some chance that the par contract is a sacrifice for us. I'll try if I can find some characteristics that would predict the chance that par is going to be a sacrifice for us. Would that be a good indication of whether it is likely to be a good preempting hand? I think it would combine all the needed elements:

• Enough trick taking ability for a sac to be profitable
  
• Our opponents having a good enough hand on their own that we need to sac
  
• Not having a majority of the strength since otherwise they would have the sac against us
  

Is it clear what I'm proposing?

 

Note that it would entail not only the estimated offensive and deffensive strengths of our hand, but the exact hand pattern as well, not just majors vs. minors. So it would get pretty complicated. I don't know if anything meaningful would come out of it. What do you think?

[Guest Jlall]

I consider 2 things, honor location and spots in your suits, to be of EXTREME importance in deciding whether to preempt. Much more than high cards even. Make sure you try to factor those in. My .02

[tysen2k]

I consider 2 things, honor location and spots in your suits, to be of EXTREME importance in deciding whether to preempt. Much more than high cards even. Make sure you try to factor those in. My .02

Naturally they are very important in determining both the offensive and defensive potential of our hand.

[tysen2k]

Okay...

 

I've plugged and chugged and got some results that I think are somewhat reasonable. The only problem is that it's way too complicated to explain in detail. So I'll give you all some take-away thoughts that I gathered from looking over the results. Remember, what I was trying to do was come up with a way that by looking at our hand we estimate the chance that the par contract is going to be us sacrificing. So if you buy that this is a good indication of whether you should preempt, here is the computer's advice about which hands make the best preempts.

• Diamond preempts rock. :D Be more prone to preempt with diamonds even if the suit quality isn't there or if you've got some outside strength
  
• 55+ hands make very good preempts unless it's both majors. Both minors are the best and 54 minor hands can be good too.
  
• Heart preempts are okay
  
• Spade preempts without heart shortness or minor voids might not be as good as you think. KQxxxx Qxx xx xx is not that good. :D
  
• Also beware of club preempts with diamond shortness. They might not have a game as often as you think.
  
• Lack of outside strength is more important than playing tricks/suit quality. xxx - Qxxxxxx xxx is great.
  
• Any void makes it likely that you will sacrifice as long as your total strength is not too high. If you have more than ~6-7HCP with a void you are probably too strong to preempt. An exception seems to be 5440 hands with honors in all 3 suits. Even pretty strong hands like Axxx - Kxxxx Qxxx or - KQxx Qxxx Qxxxx were deemed pretty good.
  

Let the discussion begin. :D

[Echognome]

Again Tysen, thanks for all your work. This definitely gives me something to think about on the other topic of designing a preempt structure.

 

I have one additional question along these lines. How do the results you gave vary (if at all) if the par spot is sacrificing in a part-score? This I believe is where a lot of our preempting lies, when the opponents might have a difficult decision whether to come in and when they do we can smack them.

 

I like the "summary style" you used for presenting these last results. Very useful for understanding the results and implications.

[tysen2k]

I have one additional question along these lines.  How do the results you gave vary (if at all) if the par spot is sacrificing in a part-score?  This I believe is where a lot of our preempting lies, when the opponents might have a difficult decision whether to come in and when they do we can smack them.

Partscore sacs are considered just as much as game or higher, all based on frequency. Basically I just considered the question, "What is the chance that the best spot is a sacrifice?"

 

I'm sure it varies by vulnerability, but for favorable (which is the condition I tested) when we have a sacrifice it is against a:

 

Partscore 40%

Game 49%

Slam 9%

Grand 2%

[Blofeld]

Can I just check: your guidelines for the suitability of preempting are based on the probability that our par score is a sacrifice?

 

If so, I imagine that 5440 hands rank highly because it's likely that there are big fits all over the place. But your preemptive methods need to be able to locate the right fit. Unless you're checking the probability that we have a sacrifice in our longest suit?

 

It's interesting data, but if you have enough time, I'd really like to know precisely what it represents. :D

[tysen2k]

Can I just check: your guidelines for the suitability of preempting are based on the probability that our par score is a sacrifice?

 

If so, I imagine that 5440 hands rank highly because it's likely that there are big fits all over the place. But your preemptive methods need to be able to locate the right fit. Unless you're checking the probability that we have a sacrifice in our longest suit?

 

It's interesting data, but if you have enough time, I'd really like to know precisely what it represents.  :D

You are right that the data has no indication of how easy it is to find our best fit. This is just to give an indication of which hands are the kinds that are likely to produce sacrifices. But I don't think finding a sacrifice in our longest suit is really the right way to go. After all if you specify a preemptive opening that is specifically a 5440 hand, you are much more likely to find a fit very easily. So I guess it's data that suggests how you should design your preempt structure.

[Blofeld]

Yes, but it would be also be interesting to know the effect of having a side suit when making a standard pre-empt.

 

e.g. Opening 2♠ with 4♥s on the side. Assume that we're never to going to find the heart fit. How often will this cost? How often will it gain?

 

I suppose that ideally I'd like to see both sets of data. Figures for the ideal case, and figures that may tell us when it's relatively safe to distort.

[tysen2k]

e.g. Opening 2♠ with 4♥s on the side. Assume that we're never to going to find the heart fit. How often will this cost? How often will it gain?

I'm not going to re-run everything again, but I can give you this quick simulation:

 

If we have 6 spades and 4 hearts, our best sacrifice is:

 

spades 69%

hearts 25%

minor 6%

 

If we have 4 spades and 6 hearts, then our best is:

 

hearts 64%

spades 29%

minor 7%

 

Note that for this test, if we can take the same number of tricks in both majors sacrificing against a minor, then spades is said to be the best (about 7% of the cases).

 

Tysen

[tysen2k]

Okay, this is an attempt to show you specifically how important the location of your shortness is when deciding to preempt. I have used my study that I did earlier on predicting the chance of a sacrifice bid and tried to map that to a sort of “preempt index.” On this index, a 2 represents what I would consider to be the bare minimum for a "normal" weak-2 bid. Similarly a 3 is the bare minimum for a 3-level preempt and a 4 is for a 4-level preempt. [Edited definition] Obviously some people may like weaker or stronger definitions for their preempts.

 

For this experiment, I looked at the 24 different suit permutations of the following hand:

 

KQJxxxx

xxx

xx

x

 

So the trick-taking ability of all 24 hands are the same, the only difference is which suits are long or short. So how much difference does it make?

 

Pattern   Preempt Index
2=1=7=3   3.82
1=2=7=3   3.81
3=1=7=2   3.80
1=3=7=2   3.76
3=2=7=1   3.74
2=1=3=7   3.73
2=3=7=1   3.72
1=2=3=7   3.72
3=1=2=7   3.67
7=1=3=2   3.64
1=3=2=7   3.63
7=1=2=3   3.63
1=7=3=2   3.52
1=7=2=3   3.52
7=2=3=1   3.50
7=2=1=3   3.48
3=2=1=7   3.48
2=7=3=1   3.48
2=7=1=3   3.47
2=3=1=7   3.46
3=7=2=1   3.44
3=7=1=2   3.44
7=3=2=1   3.36
7=3=1=2   3.36

[Edited values. Order is preserved, only absolute values have changed]

 

There is almost “half a preempt level” in difference between the top and bottom of the list. Notice that the diamond preempts dominate. Now look at the club preempts. They are pretty good except when they have diamond shortness which really drops their level. Spade preempts with heart shortness are good, but when they have 3 heart cards they fall to the bottom of the list. Heart preempts depend on their spade length, but to a lesser extent.

 

Naturally you could repeat this for other shapes like 7330, but I'm sure you'd find the same trends.

 

Tysen

[tysen2k]

Okay, here's a little contest/game. What is the best preempt you can make that has 2173 shape? This hand:

 

xx

x

KQJxxxx

xxx

 

was given for illustration and has a PI of 3.82. But you can get to 4.01 with different honors. Who can guess it?

[Guest Jlall]

xx x QJT9xxx xxx

[tysen2k]

Well, Justin, you won. Sort of. Okay, I messed up. I had a typo somewhere and with that error, yes indeed

 

xx x QJTxxxx xxx

 

was at the top of the list (there isn't enough data to judge the value of a 9). However, I found the typo and after fixing it, it turns out that

 

xx x KQJxxxx xxx

 

is on top. That's my original example, so it makes this contest quite lame. :rolleyes:

 

These are the values my formula produces (assuming all side suits are empty):

 

KQJxxxx	3.82
KQJTxxx	3.80
QJTxxxx	3.77
JTxxxxx	3.74
KJTxxxx	3.71
Jxxxxxx	3.69
QJxxxxx	3.69
QTxxxxx	3.66
KJxxxxx	3.66
KQxxxxx	3.65
KQTxxxx	3.65
Qxxxxxx	3.60
KTxxxxx	3.57
Kxxxxxx	3.55
Txxxxxx	3.46
xxxxxxx	3.42
AKQJxxx	3.22
AQJTxxx	3.21
AKQJTxx	3.21
AKJTxxx	3.21
AKQxxxx	3.19
AQJxxxx	3.18
AKQTxxx	3.18
AKJxxxx	3.16
AQTxxxx	3.16
AJTxxxx	3.14
AJxxxxx	3.13
AQxxxxx	3.13
AKTxxxx	3.12
AKxxxxx	3.09
ATxxxxx	3.08
Axxxxxx	3.07

 

QJTxxxx is still way up there. Notice that every single preempt with an Ace is at the bottom of the list. It's just to valuable on defense and makes a sacrifice less likely. And there's a big jump between the best hand with an Ace and the worst one without it. Naturally side Aces are even worse than trump Aces. Also note that if you usually count playing tricks to determine your preempts that you will count Axxxxxx and KQxxxxx the same. Some food for thought.

[MickyB]

Wow, that aces thing is far more pronounced than I would have predicted. I've got my excuse ready for next time I preempt on xxxxx xxx xxx xx, "At least I didn't have an ace partner!"

 

Think KQJTxxx would come out ahead of KQJxxxx in a larger sample - not that I'm asking for one :rolleyes:

Edited September 30, 2005 by MickyB

[awm]

Some questions about this:

 

(1) How is the "preempt level" determined?

 

(2) Why is "our best contract is a sacrifice?" a good measure? It seems perfectly fine to open 3♠ when we are making 4♠...

 

It seems to me that the bad times to preempt are:

 

(1) Times when the preempt is already past the par contract. For example, I open 3♠ vulnerable and opponents can double me for 800, and they don't have a slam. Obviously sometimes even these bids work out if the opponents judge wrong, but in principle my preempt gives them the chance to obtain a number they never could've gotten if I passed.

 

(2) Times when the par contract is us playing in a suit other than the suit or suits described by my preempt. For example, I open 3♠ and our only making game is 5♣. It will often be difficult to get there (especially if partner doesn't have magnificent clubs of his own and/or if 4♣ over 3♠ is some kind of artificial asking bid).

 

(3) If my preempts are defined in such a way that an extremely wide range of shapes and/or strengths is possible, it may be hard for partner to put us at the proper level. However, this isn't an argument that "we shouldn't preempt on certain hands" so much as that we have to pick and choose which hands our system will permit preempts.

 

I think I'd mostly look for cases (1) and (2) to try to decide which hands not to preempt. With pronounced single-suited hands it's probably good enough to measure the par contract. With two-suited hands when only single-suited preempts are available (for example, I hold 6-1-1-5 with no two-suited weak bid available, how many spades should I open?) the second constraint becomes more critical.

[Blofeld]

Nice data!

 

Am interested:

Naturally side Aces are even worse than trump Aces.

How much worse? In particular, I can see trump aces being bad when they add to your defence by being able to take them and then give partner a ruff.

[tysen2k]

(2) Why is "our best contract is a sacrifice?" a good measure? It seems perfectly fine to open 3♠ when we are making 4♠...

This is one of the things we are debating here. Is this really the right thing to measure? If not, what would be better?

 

You bring up a lot of other valid points, but I'm not sure that all of them answer the question "when is the best time to preempt?" If you open 3♠ and can make 4♠ and (I'm assuming) the opps can't make anything higher, was that really the best time to preempt? Yes we made it to the right spot but if the opps don't have anything higher that we can't bid over did we really need to jump? Every time 4♠ makes doesn't mean we always get there. Partner may not guess right to raise it up. We might reach 4♠ more often if we start with 1♠ than with 3♠.

[tysen2k]

Naturally side Aces are even worse than trump Aces.

How much worse? In particular, I can see trump aces being bad when they add to your defence by being able to take them and then give partner a ruff.

It looks like changing a club spot card to an ace lowers the rating by about 25% more than changing a diamond spot to an ace.

[MickyB]

I reckon the best measure of a preempt is what level we can bid to without going beyond par.

[tysen2k]

I reckon the best measure of a preempt is what level we can bid to without going beyond par.

That won't work because

 

AKxx

AKx

AKx

AKx

 

Is a horrible preempt but has a very high expected par level.

[MickyB]

It is a horrible preempt because you need the room for your own investigation, which isn't really part of the scope of this thread so far, and would be very difficult to incorporate. I think putting a limit on the maximum strength of the preempting hand would make the 'par level method' work.