Tests for a double-dummy solver

[awm]

Ok, here's a fairly concrete question.

 

Suppose that I am going to open a hand with a spade preempt at some level. We can say this is an okay opening if the following constraints hold:

 

(1) The level of spades that I bid is not beyond the par spot.

(2) The par contract involves our side playing in spades.

 

Hands like AKxx AKx AKx AKx are bad preempts because it's not particularly likely that spades is our best strain.

 

Note that the above constraints indicate that any hand that's "okay" to open 4♠ is also "okay" to open 2♠. This is pretty obvious in terms of safety level, but our goal in general is to preempt to the highest "reasonably safe" level. We'll also need to determine what our preempts should look like; for example if we choose to preempt 2♠ with xxxxx xxx xxx xx, we probably shouldn't also open 2♠ with AKxxxx x Kxx x, even if both are "reasonably safe" 2♠ openings. However, this is perhaps a separate issue.

 

So the question here is, how likely are we to be "okay" if we open various hands with 2♠, 3♠, 4♠ and so forth?

[MickyB]

Yes I like that. Constraint 2 probably needs rephrasing - I think your intention was, at level vul, for 4S making 8 tricks to be ok if they can make 5C? So "We can make at least as many tricks in spades as in any other strain" sounds better. Might be useful if our overbidding/reaching the wrong strain could be measured in IMPs rather than just success or lack of?

[Blofeld]

Adam's suggestion still runs slightly afoul of Tysen's objection, I think:

 

AKQJxxx

AK

AK

x

 

Is a fairly poor preempt, even if though we can open it at the 6-level without going past par, and we're likely to want to play in spades. But possibly combining it with Mike's second constraint could work.

 

---

 

My idea (not really sure about this, but thought it worth throwing into the mêlée) is that preempts will tend to be better on hands where you expect the par contract for NS and the par contract for EW to be relatively close.

[awm]

I don't see a particular problem with opening 6♠ on:

 

AKQJxxxx

AK

AK

x

 

Provided that your agreement is that a 6♠ opening shows 12 top tricks. This will prevent the opponents from finding a good sacrifice.

 

There are a lot of questions about agreements here. For example, a 1♠ preempt on:

 

AKxx

xx

xxxx

xxx

 

might be right for various reasons. But most of us don't have a 1♠ preempt available. The goal here is to figure out which preempts are "too reckless" to be reasonable, not to determine what your agreements should be. I think it's reasonable to discount partner's potential evaluation problems here; it seems clear we need some sort of agreement about what sorts of hands open 2♠ for example. Likely the goal is to devise this agreement such that we include as many as possible of the "reasonable" preempts without giving partner a problem he won't solve.

[Blofeld]

I suppose that the problem is the (all too high) chance of missing a grand slam. Can partner raise with any ace?

 

What about:

AKQJxxxxx

AKQ

-

x

[tysen2k]

Can we come up with another measure besides just par? I dislike it because it just considers our offensive strength and doesn't measure our lack of defense. These two hands shouldn't be rated the same:

 

AKxxxx

xxx

xx

xx

 

xxxxxx

Axx

Ax

xx

 

What else can we use if you don't like "chance of a sacrifice?"

[MickyB]

They won't be rated the same - par will tend to be lower for the 2nd hand because the opps are likely to be able to make less.

[helene_t]

HCP and shape are correlated

Maybe I'm pedantic (we probably all mean the same) but strictly they are not correlated. For any shape, the average HCP is 10.

 

But they are dependend. More precisely, there is a heteroscedastic relationship between the two. Freak shapes have lower* HCP variance.

[sorry I wrote higher HCP variance. Tx, Blofeld]

 

This means that the average number of tricks may be easier to interpret than the chance of making game, since the confounding from high cards is probably less severe (this statement is not trivial but I guess it's true).

 

Therefore, when I analyzed this dataset (the GIB data, also used by Tyssen) to find the optimal HCP count scale, I first thought it should be a linear model like

AvgTricks ~ aA + bK ...... (A being number of aces, K number of kings etc).

 

The problem with this is that since the relationship in fact not linear, the estimated coefficients may reflect some compromize between the HCP scale for low-level decisions and the HCP scale for high-level decision. Then I decided to use the game criterion instead, using logistic regression:

P(weMake3NT) ~ logit(aA + bK ....)

 

It turned out that the traditional 4321 scale is very close to accurate for 3NT decisions except that tens should have a weight of appr. 0.4. Interestingly, queens got a weight of 0.8 rather than the traditional 1.0. I wonder if this is really so. It could be argued that DD-simulation understimates the values of queens because declarer often has to guess how to catch the queen: Kings can only be finesed in one way (except for sec Kings, throw-ins etc).

Edited October 4, 2005 by helene_t

[Blofeld]

Freak shapes have lower HCP variance.

Fixified!

 

Good points, though.

[tysen2k]

They won't be rated the same - par will tend to be lower for the 2nd hand because the opps are likely to be able to make less.

You're right. Maybe that was a poor example. The problem I have with the "high par level" definition of a preempt is that you will need an arbitrary limit for the maximum strength or else you will get mostly strong hands that have a high par level. But that strength level has to be set somewhere. What will you use to measure it? HCP? Plus what level do you set it? 10 HCP? Who says that's the right cut-off point? Are all 10 HCP hands the same? You can see all of the issues.

 

I was thinking about the "our par and opponent's par should be close" definition. That seems interesting. It might even be the right definition for matchpoints since we will both want to be around the same level. The problem is you will get hands like this:

 

xxxx

KQJ

QTx

JTx

 

Where is is very likely that both sides can make a 2-level contract. Is that my perfect preempt?

[Rebound]

I think it is a pretty clearly understood principle that more distributional hands require fewer HCP to make at a given bidding level. Isn't that what this discussion is intended to quantify? I.e. when does a distributional hand have too many values to be used to preempt, since we generally don't want to preempt when it's our hand unless the attempt is being made to keep the opponents from their fit.

 

This would seem to argue that it is less neccessary to preempt in the majors, i.e. fewer HCP are required. Taken with Tysen's point that major 2-suiters are more likely to produce game, this suggests it is better to go slower with hands in the upper range for a normal preempt (i.e. open at the one level) and preempt in the minors.

[awm]

I'll try to address the various complaints about measuring par:

 

(1) People give examples of hands which are relatively flat (i.e. Tysen's 4333 9-count). Yes, it is likely that some two level contract can be made when you hold these hands -- it's pretty much always likely that some two-level contract can be made. This is part of the rationale behind weak notrump openings (especially the 9-11 and 10-12 varieties). So it's not necessarily clear that this hand is a "good" or "bad" preempt and will probably depend on the vulnerability (i.e. if we can't make a two-level contract and we get doubled, how expensive does it get). On the other hand, one of my criteria for a good preempt (criterion two, the one that isn't about par) is that we'll do pretty well playing in the strain opened. This is an attempt to measure some of the effect of "preempting our own side" in that it can be hard to reach a minor suit contract after I open 2♠ (for example). Obviously preempts that show two suits can get you to either suit, but in general it is still hard to get to a contract in a suit that opener hasn't shown. Hands like Tysen's 4333 shouldn't be opened with 2♦ weak (for example) because while we can usually do okay in two of something, there's no reason to think we can do well in diamonds at any level.

 

(2) Doesn't suit quality matter? Par will measure suit quality. If I open 2♠ with six weak spades, it becomes much more likely that we go down a lot. Especially if I have side defensive tricks (making it less likely that the opponents actually have a high-scoring contract available if I pass throughout) there will be a tendency for 2♠X to pay the opponents more than the par result. On the other hand, if I hold KQJTxx of spades and out, it will be very rare that I go down more than a few in 2♠, and when I can be set three tricks doubled, the opponents almost surely have a slam that would pay them better (2♠X is not likely to be a better-than-par result for opponents).

 

(3) What about opening "preempts" with very strong hands? A lot of this is a matter of agreements. Obviously I shouldn't open 2♠ with a 15-count even if I have enough spades that it's likely the best strain (and my spades are good enough that 2♠ is safely par or below).... but this is only true if I play weak two bids. Actually there are some very interesting questions about opening two-bids with decent hands. For example, take Ritong 2♣ players, who will open 2M with a five-card major, four-card clubs, and minimum opening values. Partner will know to look for game on balanced twelves that would pass a weak two, so it's not like the fact that opener has a good hand preempts partner out of the auction. The danger is that you may already be too high (if it's a misfit hand where your side has around half the strength, you may have turned a board where par for you is positive into a negative result). The secondary danger is that the best contract is not in either opener's major or clubs, and that it may be difficult to reach 3♥ (for example) when it's the best contract after a Ritong 2♠ open. I'm willing to assume that our openings are sufficiently well defined that we can trust partner's judgement to reach the right level if our opening names the strain; obviously one can come up with examples where this is not true but that's more a question of "what are our agreements" rather than "what is reasonable to open on."

 

My issue with "likelihood of a sacrifice" is hands like:

 

KQJT9x x Axx xxx

 

KQJT9x x xxx xxx

 

Since the first hand is stronger, the odds that "our best contract is a sacrifice" will be substantially less. But is the first hand really a "worse" preempt than the second? Surely we are less likely to go for a number in 2♠ on the first hand. It is just as likely that spades is a good strain (in both cases it seems almost certain that ♠ is a good strain). On the second hand you can argue that "if 2♠ is three down, then opponents had a slam and two down opponents have a game" but on the first hand you can similarly argue that "2♠ will not go three down, and if it's two down opponents had a game." I'd rate these hands as equally good preempts, basically because an ace has roughly equal value on offense and defense (it's always a trick). Of course, if you open 2♠ with too wide a range of hands partner might misjudge, but it's certainly feasible to agree that a 2♠ opening could show either of these hands (i.e. between 5 and 6 tricks in spades) and partner has room to use ogust or whatever to clarify.

[Rebound]

See my post (edited) above - the evidence seems to indicate the first hand ought to be opened 1♠.

[awm]

Well take KQJT9x x Axx xxx as an example. If we open 2♠:

 

(1) Partner (hopefully) has a good description of our hand right away. He will place the contract in an appropriate number of spades.

 

(2) If the par contract involves the opponents playing in some suit (for example hearts) it will be harder for them to find it. This is true regardless of whether par is opponents making 5♥ (they will be hard pressed after an auction like 2♠-X-4♠ and may well decide to defend) or par is opponents sacrificing in 5♥ over our making 4♠.

 

What are the downsides of opening 2♠?

 

(1) If for some reason we really need to play in a minor suit, it's going to be tough to get there. However, these spades are so good that it's not that likely our best contract will be in a minor, is it?

 

(2) It's vaguely possible that the opponents will double 2♠ and set it a trick or two, when they didn't have a game. But this is pretty unlikely given opener has basically 6 tricks in hand.

 

(3) In my opinion, the only serious downside is that partner might misjudge. In other words, this hand might be "too good " for a 2♠ opening and partner will place the contract at the wrong level. However, that's only true if you'd also open 2♠ on some really ratty hands. Playing a "sound" preempt style, this hand is a perfectly good 2♠ bid.

 

One of the goals here (in my opinion) is to determine how aggressively it is right to preempt. If your weak twos are very sound (i.e. the given hand is a normal max) then you will get very good results when you preempt. However, you may not be preempting enough. On the other hand, if you open weak twos with junky hands then your results when you preempt will be worse, but you will get to preempt a lot more. At some point you will start getting diminishing returns (i.e. your preempts are so bad, that opponents can often get big numbers out of doubling you, and the contract for your side often plays better in a suit other than yours). The question is where this point actually lies. My suspicion is that many people preempt too aggressively at unfavorable colors, and not aggressively enough at favorable.. but we can certainly use double dummy data to get better answers on this than my suspicions.

[tysen2k]

One of the goals here (in my opinion) is to determine how aggressively it is right to preempt.

Naturally, that's the holy grain of preempting structure design. There are two issues:

• The ordering of all hands in a list from best preempt to worst preempt
  
• Where to set the cut-off bar on the list (how often/how high to preempt)
  

I hope that we can do the first without specifying the second. Choosing to use your 2-bids for preempts, strong hands, constructive, etc. should be a seperate issue.

 

If we use a par measurement, we still need some way of deciding when we are "too strong" to preempt. How are we going to decide that?

[tysen2k]

My issue with "likelihood of a sacrifice" is hands like:

 

KQJT9x x Axx xxx

 

KQJT9x x xxx xxx

 

...I'd rate these hands as equally good preempts, basically because an ace has roughly equal value on offense and defense (it's always a trick).

I suspect you'd be in the minority here. If the addition of a side ace makes no difference in your preempt, I think something is wrong. Most experts agree that the presence of side aces is a huge detriment to preempting.

[Guest Jlall]

Most experts agree that the presence of side aces is a huge detriment to preempting.

I don't know how true this is anymore. I see it in all of these textbooks, and must say I really never believed, and don't believe, that a side ace is a flaw for a preempt. I much prefer it to a side QJ(tight) for instance. I also see many experts preempting with side aces despite this "flaw."

[junyi_zhu]

Most experts agree that the presence of side aces is a huge detriment to preempting.

I don't know how true this is anymore. I see it in all of these textbooks, and must say I really never believed, and don't believe, that a side ace is a flaw for a preempt. I much prefer it to a side QJ(tight) for instance. I also see many experts preempting with side aces despite this "flaw."

I think the major goal of preempts is not to find the best sacrafice, it's to create a problem for opps. You preempt, and they have to guess. The more you preempt, the more times they may misguess. It's fine that we find a sacrafice spot at high level due to your preempts; it's also fine that you direct a good lead from your preempts; but your major goal is to create a problem for them: they don't know how to bid with 22 HCP and balanced; they don't know how to bid with weak 5 card, a side suit void and no stopper in your suits and 17 HCP; they may miss a game when both of them hold 12-13 HCP, balanced and nobody is strong enough or has the right shape to bid at 3 level. These are the more valid concerns for opps. In that sense, what you hold isn't that important sometimes, especially when white. The basic idea of destructive bidding is not to find your best spot, but to create problems and at the same you don't get burned severely. That's why so many good players just preempt regardless side suit aces, second side suit, the other major, low quality suits, especially when white. Bridge is never a double dummy game, and over preempts, it's even often hard to find the second best contract. So for the preemptive bidding, the basic idea is just the normal military strategy: "destroy enemies (good contracts) and protect yourself (from being hurt severely)". Still, all the researches above are valuable, because they show some new thinking of how to make marginal decisions in preempts and why 2D preempts is still popular.

[tysen2k]

I took 3 different hands (I'm not showing the hands) and generated several par scores for them. How should we go from this data to determining if we should preempt and how high?

 

     Hand A   Hand B   Hand C
7x    3%       0%       0%
6S   10%       7%       3%
6x    3%       3%      11%
5S   16%       7%      11%
5x    3%       0%       3%
4S   23%      50%      14%
4x    0%       3%       9%
3S   26%      13%      26%
3x    0%       3%       9%
2S   16%      13%       9%
2x    0%       0%       3%
1S    0%       0%       0%
1x    0%       0%       3%

 

I'm actually quite surprised about something in the data because I know which hands generated which results. But what can you tell me about each hand using the data alone?

[Blofeld]

Could you explain what, say, 4x and 4S mean here?

[tysen2k]

Could you explain what, say, 4x and 4S mean here?

Yeah, sorry. 4S means our par contract is 4♠. 4x means that our par contract is 4 of something other than spades.

[hrothgar]

I took 3 different hands (I'm not showing the hands) and generated several par scores for them.  How should we go from this data to determining if we should preempt and how high?

 

 [space] [space] Hand A [space] Hand B [space] Hand C
7x [space] [space]3% [space] [space] [space] 0% [space] [space] [space] 0%
6S [space] 10% [space] [space] [space] 7% [space] [space] [space] 3%
6x [space] [space]3% [space] [space] [space] 3% [space] [space] [space]11%
5S [space] 16% [space] [space] [space] 7% [space] [space] [space]11%
5x [space] [space]3% [space] [space] [space] 0% [space] [space] [space] 3%
4S [space] 23% [space] [space] [space]50% [space] [space] [space]14%
4x [space] [space]0% [space] [space] [space] 3% [space] [space] [space] 9%
3S [space] 26% [space] [space] [space]13% [space] [space] [space]26%
3x [space] [space]0% [space] [space] [space] 3% [space] [space] [space] 9%
2S [space] 16% [space] [space] [space]13% [space] [space] [space] 9%
2x [space] [space]0% [space] [space] [space] 0% [space] [space] [space] 3%
1S [space] [space]0% [space] [space] [space] 0% [space] [space] [space] 0%
1x [space] [space]0% [space] [space] [space] 0% [space] [space] [space] 3%

 

I'm actually quite surprised about something in the data because I know which hands generated which results.  But what can you tell me about each hand using the data alone?

I am assuming that 6S refers to the frequency with which 6S is the par contract and 6x refers to the frequency with which some other 6 level contract is the par contract. I'll note in passing that this notion of par is problematic. We're not sure whether 4♠ the par contract because its a good sacrifice over the opponent's 4♥ or whether 4♠ is a making game.... These are VERY different considerations.

 

A few things stand out:

 

1. In each of the three hands, the frequency with which the par contract is a Spade contract is substantially more frequent than any other contract at the same level. In turn, this suggests that each of the three simulations includes one hand with comparatively long Spades.

 

2. The par contract for hands 1 + 2 is a Spade contract, approximately 90% of the time. In contract, the par contract for hand 3 is a Spade contract 63% of the time.

 

3. Hands 2 and 3 feature relatively distinct modes. If we ONLY consider Spade contract, Hand 2 has a 4♠ par 56% of the time. The next most common par is is a tie between 2♠ and 3♠, each of which clocks in a 13%. Hand 3 has a 3♠ par contract 41% of the time. The next most common par is 4♠, which occurs roughly 22% of the time. In contract, hand three mode is 3S which occurs 29% of the time. The next most common par is 4S which occurs 25% of the time. 5S and 2S each are the mode contract 18% of the time.

 

What can we take away from this: If we're ONLY worried about reaching the par contract ASAP, than hand 2 is by far the "best" preempt. Preempting on hand 1 is a crap shoot. The par contract is all over the place. While we can be pretty sure that the hand belongs in Spades, we're badly positioned to judge level. Ideally, we'd like to be able to set Spades as trump and have a contructive auction to establish the right level. In contract, with hand 3 we're not even sure if Spades are the right contract. If Spades are the right contract, then its "safe" to bid up to the 3 level. However, we "only" want to be in game about 44% of the time.

 

With this said and done, its unclear what happens if we consider competitive bidding. I suspect that the same characteristics that make hand 4 such an attractive constructive preempt might hurt it if we consider competitive bidding. The precision of the bid could help the opponents every bit as much as it helps us. In contrast, lets consider hand 3 (the "crap shoot" that might not even belong in Spades). Preempting with hand three is a randomizer. If I open with a 3!S preempt, I'm going to jam our auctions something fierce. However, the opponents are going to be under much the same pressure.

 

I can see an argument that suggests that

 

Hand 1 should be opened 3♠ bid setting trump. Responder will be well positioned to judge level and we don't need to explore alternative strains.

 

Hand 2 should be opened 4♠. Ideally the level of the preempt will compensate for the precision of the hand type (Even if the opps NO that 4♠ is par for us, they'll be poorly positioned to do anything about it)

 

Hand 3 should be opened 2♠ (potentially even 1♠). This opening uses up the least bidding space, and it allows the opponents to introduce their suits before we reach the par contract. Even so, we hold the master suit. I suspect that the additional bidding space will be more useful to us than to them.

[han]

This is what I would guess for the hands:

 

Hand 1: KQxxxx Kxxxx x x.

 

Hand 2: KQJ10xxxx x xxx xx.

 

Hand 3: KQJxx Kxx xxx xx.

[Blofeld]

I largely agree with Hannie's guesses for Hands B & C, but I think hand A is more one-suited, and contains at least one ace; perhaps something more like:

 

AKQJxxx

xx

xx

xx

 

Hand C, though, could well be something more 2-suited. Maybe 6-4 shape.

[tysen2k]

Some good guesses. I'll let more people chime in before showing you the actual hands. But remember that there's another part to this. How are we going to take this data and create some rules/formulas for determining preempts? :lol: