[straube]

Only 5 hands so far but an early lead for king parity as the first question. Here's the totals of extra steps (low is good) so far for the four methods I've run...

short to long, K parity last.........12
short to long, K parity first........2
long to short, K parity last.........13
long to short, K parity first........5

If anyone wants to comment in a general way on these threads, it would be nice to do it on this one (why I'm starting it really). We can use the others to bid hands and correct the math, etc

[straube]

Hand #6 was bad for king parity first as it couldn't resolve the cards (it came down to two choices). So I'm wondering if it's a one-off or not, but with only 6 hands bid so far it's concerning.

1. King parity first seems to give the faster picture of the hand than king parity after...but they are actually two different questions and I wonder if combining them would make sense. Quick picture, better resolution, at a cost of another question.

2. I've been skipping doubletons when even one way (short to long) and stopping the other (long to short) which introduces some unnecessary variance. Do folks have an opinion which is better?

3. I'm still liking the idea of scanning singletons first and probably the whole short to long idea, but it's very early.

Glad to see foobar and yunling are testing ideas.

[foobar]

 straube, on 2017-October-18, 22:53, said:

Hand #6 was bad for king parity first as it couldn't resolve the cards (it came down to two choices). So I'm wondering if it's a one-off or not, but with only 6 hands bid so far it's concerning.

1. King parity first seems to give the faster picture of the hand than king parity after...but they are actually two different questions and I wonder if combining them would make sense. Quick picture, better resolution, at a cost of another question.

2. I've been skipping doubletons when even one way (short to long) and stopping the other (long to short) which introduces some unnecessary variance. Do folks have an opinion which is better?

3. I'm still liking the idea of scanning singletons first and probably the whole short to long idea, but it's very early.

Agree with all three above. Also think that awm's idea of always using K=2 makes logical sense if you consider that it reduces the number of ambiguous combinations, especially in conjunction with early K-parity (but with no exceptions for the stiff). One tweak to consider for the short suit scanning is to skip with nothing in both singletons and doubletons and then revert to normal scanning order for 4+ card suits.

On a side note, when scanning the strong hand, my conjecture is that A-parity first might make more sense, but it's just speculation at this point.

[sacto123]

Can you please explain what PCB, DCB, and QP stand for? I have never heard of these methods.

[Zelandakh]

 sacto123, on 2017-October-19, 00:46, said:

Can you please explain what PCB, DCB, and QP stand for? I have never heard of these methods.

Parity Cue Bids, Denial Cue Bids and Queen Points.

PCBs are a method where instead of answering a Yes/No question in each suit (control/no control) you answer whether you have an odd or an even number of cards there. Advocates say that this provides a better picture of the hand more quickly in combination with the information for amount of controls and the cards held in the strong hand itself.

DCBs are the traditional approach in relay bidding. For each suit, you skip a step if a control is held there and make the corresponding bid if no control is held. This tends to be good on hands where one control is held in each suit but bad on hands with the critical cards concentrated in a single suit or on hands where several queens need to be located.

QPs are an alternative to the more traditional Control Points (A = 2, K = 1), with A = 3, K = 2, Q = 1. They are equivalent to the popular 4.5/3/1.5/x hcp methods but with jacks ignored, which makes sense for slam auctions.

[straube]

Atul, if you are counting A=3, K=2, Q=1, do you think you ought to start a step higher as your QP total will be higher? I'm wondering if the method you are testing will be able to disambiguate board 7 sooner.

On another note, I'm thinking it would be nice for...

king parity question
short to long, skipping small singletons and even doubletons
asking whether a Q is held in the longer/higher of two or more evens or the longer/higher of three or more odds

What do folks think of this last question? The goal would be to sometimes disambiguate KQ or AQ from two evens or sometimes differentiate a K in one suit from a Q in another.

[sacto123]

 Zelandakh, on 2017-October-19, 03:41, said:

Parity Cue Bids, Denial Cue Bids and Queen Points.

PCBs are a method where instead of answering a Yes/No question in each suit (control/no control) you answer whether you have an odd or an even number of cards there. Advocates say that this provides a better picture of the hand more quickly in combination with the information for amount of controls and the cards held in the strong hand itself.

DCBs are the traditional approach in relay bidding. For each suit, you skip a step if a control is held there and make the corresponding bid if no control is held. This tends to be good on hands where one control is held in each suit but bad on hands with the critical cards concentrated in a single suit or on hands where several queens need to be located.

QPs are an alternative to the more traditional Control Points (A = 2, K = 1), with A = 3, K = 2, Q = 1. They are equivalent to the popular 4.5/3/1.5/x hcp methods but with jacks ignored, which makes sense for slam auctions.

I appreciate your detailed answer. Thank you.

[foobar]

 straube, on 2017-October-19, 20:36, said:

Atul, if you are counting A=3, K=2, Q=1, do you think you ought to start a step higher as your QP total will be higher? I'm wondering if the method you are testing will be able to disambiguate board 7 sooner.

On another note, I'm thinking it would be nice for...

king parity question
short to long, skipping small singletons and even doubletons
asking whether a Q is held in the longer/higher of two or more evens or the longer/higher of three or more odds

What do folks think of this last question? The goal would be to sometimes disambiguate KQ or AQ from two evens or sometimes differentiate a K in one suit from a Q in another.

I was thinking along similar lines...basically, for doubletons and singletons use inverse-classic DCB (scanning for A/K only, but stopping if held). Obviously this can introduce some ambiguity in case of Qx, KQ, AQ, but perhaps it can be offset by the relay captain's holding and the early K-parity?

[straube]

 awm, on 2017-October-20, 07:14, said:

Here's what's in our current notes:

1. If there is a pair of suits where one has KQ and the other has none, the first such pairing (length order, followed by high to low) is picked. Stop if first suit has KQ, skip if second suit has KQ.
2. Otherwise if there is a pair of suits where one has K and the other has Q, the first such pairing (length order, followed by high to low) is picked. Stop if first suit has K, skip if second suit has K.
3. Otherwise this step does not exist.

This appeared slightly better than king parity in my (long ago, not going to be repeated) simulation.

Moving this post here so I can find it again. It looks like awm solved this a long time ago. I want to run through the hands with it. One has a marriage problem and another has a K and Q in separate suits problem.

[sieong]

FWIW, if the only goal is to resolve honor structure as quickly as possible, the following (untested) structure is more space efficient than classic PCB, assuming the parities of the shorter suits are equally likely to be odd or even (to be validated; I have not got a chance to run a sim. If not true, there are ways to optimize this further). It gains on average 3/8 of a step, and can be made never worse than PCB on any honor parities assuming we always want full resolution.

NOTE: I am not suggesting that one would play this, since it is unclear if it allows as easy a way to get out opposite the wrong honors as classic PCB. Full resolution is not the only goal for honor structure relay.

Resolution rules:
Longest to shortest, tire broken by rank, highest to lowest
All suits are scanned
Step 1 pretty for first suit
Next question is an ask across all the remaining suits at the same time, answers in parities for these three suits are
... ooo, eoo, oee
... oeo, eoe
... eee
... ooe
... eeo
Final question: same as IMprecision

This may look a little abstract so I will bid the hands as examples. I will refer to this as EPCB in the posts.

[straube]

 sieong, on 2017-October-21, 22:55, said:

FWIW, if the only goal is to resolve honor structure as quickly as possible, the following (untested) structure is more space efficient than classic PCB, assuming the parities of the shorter suits are equally likely to be odd or even (to be validated; I have not got a chance to run a sim. If not true, there are ways to optimize this further). It gains on average 3/8 of a step, and can be made never worse than PCB on any honor parities assuming we always want full resolution.

NOTE: I am not suggesting that one would play this, since it is unclear if it allows as easy a way to get out opposite the wrong honors as classic PCB. Full resolution is not the only goal for honor structure relay.

Resolution rules:
Longest to shortest, tire broken by rank, highest to lowest
All suits are scanned
Step 1 pretty for first suit
Next question is an ask across all the remaining suits at the same time, answers in parities for these three suits are
... ooo, eoo, oee
... oeo, eoe
... eee
... ooe
... eeo
Final question: same as IMprecision

This may look a little abstract so I will bid the hands as examples. I will refer to this as EPCB in the posts.

It looks like you ordered these specifically for some reason. I would think

...ooo, ooe, oeo
...eee, eeo
...oee
...eoe
...eoo

because then you know odd or even for two suits early.

How about amending your rule to be...look at marriages first, the k or q, then AQ problems?

3/8 step is huge savings.

[Zelandakh]

If I were doing this sort of structure I would probably go for the first step being one even (ooe/oeo/eoo), the second step all the same (eee/ooo)and the third-fifth steps one odd (oee>eoe>eeo) (ie the last step being odd in the shortest suit. This strikes me as much easier on the old grey cells than something completely abstract. In summary:-

1. oxxx
... - 3. eooe/eoeo/eeoo
... - ... - 5. eooe
... - ... - 6. eoeo
... - ... - 7. eeoo
... - 4. eeee/eooo
... - ... - 6. eeee
... - ... - 7. eooo
... - 5. eoee
... - 6. eeoe
... - 7. eooe
2. eooe/eoeo/eeoo
... - 4. eooe
... - 5. eoeo
... - 6. eeoo
3. eeee/eooo
... - 5. eeee
... - 6. eooo
4. eoee
5. eeoe
6. eooe

It also strikes me that this concept, using Fibonnaci combinations, offers several other possible logical divisions. There are 16 combinations so we need 7 steps, with only 3 combinations using the final step. An example:-

1. 2 even suits (eeoo eoeo eooe oeeo oeoe ooee)
... - 3. oxxx
... - ... - 5. ooee
... - ... - 6. oeoe
... - ... - 7. oeeo
... - 4. eooe
... - 5. eoeo
... - 6. eeoo
2. 3 even suits (eeeo eeoe eoee oeee)
... - 4. oeee
... - 5. eoee
... - 6. eeoe
... - 7. eeeo
3. 0 or 4 veen suits (eeee/oooo)
... - 5. eeee
... - 6. oooo
4. oooe (4-7 = 1 even suit)
5. ooeo
6. oeoo
7. eooo

It would need some testing whether such a parity-parity approach was competitive with the suit1-parity and suits methods already provided.

[awm]

I think one thing that is missed, is that ~7 steps will usually be too many. There is some advantage to stopping more often than Fibbonacci would give, because you can sometimes determine to sign off after a cheap stop whereas zooming to 6-7 steps often is too high.

[sieong]

 awm, on 2017-October-22, 11:34, said:

I think one thing that is missed, is that ~7 steps will usually be too many. There is some advantage to stopping more often than Fibbonacci would give, because you can sometimes determine to sign off after a cheap stop whereas zooming to 6-7 steps often is too high.

Yeah. The benefit of being able to resolve complete parity lower must be weighed against the cost of not being able to sign off opposite some combinations of honors as low as classic PCB. This can likely be answered by a simulation. I suspect there may be gains when combined QP is greater than 19, since typically the 5 level is safe, but may be a loss when combined QP is less.

To answer why this particular ordering (I know it looks like madness, but I promise there is some method to the madness, and for sure there is plenty of madness in the method): the packing is meant to be xor. First step is all odds, or longest suit has different parity; second step is middle suit has different parity, etc.

As to why... When opener has a suit with all three honors (not in responder's longest suit), the complete parity will be known right away for the first two steps. Also, in cases where responder has a short suit, if one assume the short suit had even parity (usually a good assumption, so this is good for "guessing scenarios"), complete parity is also "known".

One trade-off here is whether knowing a bit about all three suits, or knowing one suit fully is more useful. I do not know the answer. Again it may be combined strength dependent. On the example hands in the forum, it appeared that knowing one suit fully is more useful., but sample size is too small. Fwiw, an alternative I considered look like what David proposed.

As to Zel's point on Fibonacci series: there are indeed several ways to pack the combinations, which opened up opportunities to optimize for secondary objectives, for example, stopping as much and as often as possible. Fwiw, first step can have at most have 8 possibilities, any more and we will spill over in steps. There are almost for sure better ways to pack this. Also, the assumption that all parities are equally likely is almost surely incorrect, so if one wants to optimize more, there are lots of room.

As an aside, it is pretty straight forward to extend this to answer more questions. For example, the 5th question could be the IMprecision resolution ask. However, since that question is not always present, I think on average is better not to pack it together. Another possibility is to use the 5th question to clarify interior quality (JT9) of the longest suit.

As another aside, opposite three suiters, it may be better to go straight to the three long suits first, then answer parity on the short suit last (and invert the step for extra optimization).

Anyhow, the post was mostly meant as a thought experiment as to whether it is possible to rearrange PCB to gain steps. Plenty of evaluation, especially on cost-benefit of not being able to stop as often, remains to be done.

[DinDIP]

Sam's proposal for looking at all four suits at once reminds me of two methods developed in Sweden by Johan Ebenius and his group (Kaj Kokko, Daniel Auby, Olle Wademark, Fredrik Nyström and Bo Wiik) in the mid-late 90s and early 2000s.

They developed these methods -- which my testing showed to be effective but (IMO) to require considerable CPU cycles -- as an alternative to their version of DCB, which itself has lots of interesting ideas that are worth testing and perhaps modifying.

First, after asker knows the exact number of QP he relays to find out if teller has an odd or even number of aces. One step shows an even number, while 2 steps+ shows an odd number. After this it will be much easier for asker to figure out teller’s QP-structure. Asker can immediately deduce if there is an even or odd no of Q’s by looking at the no of QP and aces (odd or even). Even number of aces and even number of QP gives even number of Q’s (so does an odd number of aces and an odd number of QP). An even number of aces and an odd number of QP gives an odd number of queens (so does an odd number of aces and an even number of QP).

When teller has at least 2QP outside the aces the next step after AOE is KOE. Teller shows if he has an odd (2+ steps) or even (1step) no of Kings. This makes it easier for asker to visualise the QP-structure.

Second, is the scanning which they called JVCB, joint variable cue bidding. Here the rules are very different from standard DCB or parity, taking advantage that asker either knows the QP structure or has only a small number of alternatives that are possible.

The JVCB-responses work by denying specific high cards in specific suits. We start w K-values in 4+suits. First the longest (lower/lowest if equal length) suit, then the next longest/lowest and so on. Thus a one step bid may deny the HK. The 2-step response would show the HK and deny, say the SK.

The first thing teller does is to determine what level of JVCB applies. The levels are:
1/ K –values in 4+suits 2/ A and Q-values in 5+suits
3/ K-values in 2-3*suits 4/ A and Q-values in 1-4*suits
5/ J-values 6/ SING J

Note: The value of a SING K or Q is one step below the normal value of the card. E.g a singleton Q counts as a J-value.

JVCB starts at level 1. It advances to level 2 when teller has shown all K-values in 4+suits. With both the ace and Q in a suit we stop on that step as we would have done with none. When teller has no or all remaining cards at a level he just proceeds to the next level. When teller has either all aces or all queens in the remaining suits on level 2 and 4, that kind of QP-value ceases to exist for JVCB purposes. When there are no unknown cards in a suit it also ceases to exist for JVCB purposes (e.g SING H that has been shown or a KQ that has been shown in a 2-card suit or a void). The teller always assumes that asker knows his QP-structure, even though that may not be the case. We will never explicitly show QP-values (level 1-4) in the last suit where we have unknown cards. E.g teller has shown 5QP, 1633, even no of aces, even no of kings. He has HK, HQ and DK. He starts by bidding three steps to show the HK (level 1), the HQ (level 2) and deny the CK (level 3). The next relay will be for jacks since a K-value in a SING suit is not possible (the last K must be in D) and there are no further QP-values to be shown. When on level 4 there are two suits left and there is one A-value and one Q-value left to show and there must be one QP-card in each suit (i.e there’s only one unknown card in each suit), we skip the rest of level 4.

The JVCB-level may advance w/o the captain making a bid (which is what we call a zoom). For example, if teller has shown 3433, 3QP, even number of aces and odd number of kings and bids 4 steps to the relay, he shows K,Q and J of H and denies the CJ.

There are special rules for wild hands: When we have shown 10+cards in the two longest suits we get a special extra level that comes directly after level 2. If you have extra length you stop on that step. When we have shown extra length by stopping at that step (denying “no extra length”, a lovely double negation), the next relay is for which suit it was w one step referring to the longest (lowest if equal length) suit. You stop w/o extra length in the denoted suit. Promised 7330 and 7321 may have extra length in three suits, while a two-suiter w high SING can’t include extra length. We can never show a 9*suit or 12* in 2 suits. After the extra length level it’s business as usual. Please remember the rule that we show low SING first w 6511 and 7411, which excludes some possible extra length.

I'll separately post the other two methods after making electronic copies of them.

David

[nullve]

 DinDIP, on 2017-October-31, 04:20, said:

First, after asker knows the exact number of QP he relays to find out if teller has an odd or even number of aces. One step shows an even number, while 2 steps+ shows an odd number. After this it will be much easier for asker to figure out teller’s QP-structure. Asker can immediately deduce if there is an even or odd no of Q’s by looking at the no of QP and aces (odd or even). Even number of aces and even number of QP gives even number of Q’s (so does an odd number of aces and an odd number of QP). An even number of aces and an odd number of QP gives an odd number of queens (so does an odd number of aces and an even number of QP).

When teller has at least 2QP outside the aces the next step after AOE is KOE. Teller shows if he has an odd (2+ steps) or even (1step) no of Kings. This makes it easier for asker to visualise the QP-structure.

I wonder if instead of showing

* # of QPs, then ace parity, then king parity,

it might work just as well to show

* 3-point hcp range, then king parity, then queen parity,

assuming the ace parity can then almost always be inferred.

[DinDIP]

 nullve, on 2017-October-31, 07:48, said:

I wonder if instead of showing

* # of QPs, then ace parity, then king parity,

it might work just as well to show

* 3-point hcp range, then king parity, then queen parity,

assuming the ace parity can then almost always be inferred.

Ulf Nilsson, another Swede, experimented with something very like this. Teller showed HCP in a 3(4)-point band then showed parity of aces then kings and finally queens (always even/odd). Once parity for each type of honour was known the next phase was parity scanning of suits. Passing showed 1 or 3 (A/K/Q or AKQ). Stopping showed 0 or 2 (AK/AQ/KQ). Singletons were ignored. 2nd sweep asked for jacks; never zoom into showing/denying jacks.

David

[straube]

Hi David. You've posted some interesting methods and I sort of get the idea, but would you mind bidding a few of the hands I had posted (separate threads for each)? With explanations please? The example you gave were helpful. Once I get the idea I hope to bid a few of them myself. I'd like to eventually stack these methods side by side and see if one seems to resolve before the others.

Also, I think awm and others have made a good point about using a structure that lets you know early whether a particular slam is a good bet or not, even before placing all of teller's cards. Like identifying a hole in the trump suit. How would you guess JVCB does in this regard?

[nullve]

I've tried to bid all 8 hands using a method (not fully developed) where Teller shows 3-point range first, then king parity, queen parity, king location, queen location and jack location, in that order.

The 3-point range doesn't necessarily have to be shown after shape has been reolved. E.g. using the Ambra-inspired structure I proposed in the QP/strength without scanning thread, the auction in 2 would begin

1♥-2♣ ("10-15, 5+ H, unBAL"; GF relay)
2♦-2♥ (MIN, 3- H; relay)
3♦-3♥ (1-S4C; relay)
3♠-? (1534, hence 10-12 hcp; ?),

which transposes to 0+ symmetric with 3♠ showing 10-12 hcp instead of a number of QPs.

Some rules:

* If Teller's hand contains at least one singleton, then king parity is shown via

step 1 = even # of (non-singleton) Ks, no singleton K
step 2 = singleton K
...step 3
......step 4 = even # of non-singleton Ks
......step 5+ = odd # of non-singleton Ks
step 3+ = odd # of (non-singleton) Ks, no singleton K

and otherwise by

step 1 = even # of Ks
step 2+ = odd # of Ks.

* Queen parity is always shown via

step 1 = even # of non-singleton Qs
step 2+ = odd # of non-singleton Qs.

* Singleton suits are not scanned.

----------

The auction on hand 8 was a predictable fiasco.