[thepossum]

Hi

Something I keep wondering about is the use of DD as the standard or par for assessing a result or play. Surely it's more accurate to compare with an incomplete information model

How would best declarer and best defence operate with incomplete information. Or is that how it works. They often seem to show totally unrealistic outcomes and plays

Regards P

[smerriman]

It's not meant to be used as showing what the correct result is. You're meant to use it as a starting point along with human analysis and adjustments.

Single dummy analysis is simply too computationally difficult (if it wasn't, GIB would be unbeatable), so it's all you have.

[thepossum]

smerriman, on 2020-November-17, 01:14, said:

It's not meant to be used as showing what the correct result is. You're meant to use it as a starting point along with human analysis and adjustments.

Single dummy analysis is simply too computationally difficult (if it wasn't, GIB would be unbeatable), so it's all you have.

I appreciate that, but it is used extensively, and often in situations where I would challenge the appropriateness of its use. I wasnt talking so much about GiB but the broader aspect of how DD is used to compare and judge hands. Surely in many cases single dummy simulation (using DD perhaps) combined with recommended/understood/widely used behaviours (especially leads, signals etc) would be more suitable. And of course, most important of all, what any sane or practical bidding system can reach

I appreciate your point about computational requirements. As I said, I'm very new to thinking about this kind of thing, but just wondering about getting greater reality into simulations. Of course I understand in larger scale operations that DD is all that is feasible (ATM)

I had a recent NT hand in a tourney. The scores ranged from 3NT+3 to anything less than that. DD showed the contract as a 1NT hand etc

[smerriman]

thepossum, on 2020-November-17, 01:43, said:

I appreciate that, but it is used extensively, and often in situations where I would challenge the appropriateness of its use.

Do you have any examples of where they're used inappropriately?

For example, tournaments provide the double dummy 'par' scores in a printout, but I don't know of anyone who interprets them as 'this is what you should have made'; they're well known to be courtesy only. Like I said, you're meant to analyse them yourself - if you see that double dummy required dropping a singleton king, then you adjust.

If people you know do use them like that, then the problem is their interpretation - not the fact they're provided. There simply isn't anything else that can be easily provided.

[AL78]

I use them as an initial guide to what the theoretical maximum number of tricks in any contract is, compare to what I did, then try to see how that number of tricks can be made. If it means having to take unlikely plays like dropping stiff king or doubleton queen offside in a seven card fit, or it requires guessing a layout correctly with insufficient information, I'm not going to beat myself up for not finding it. It is better to look at what the other players sitting your way did on a board. If the majority made more tricks than you, that is more likely to suggest you made a mistake, even then you can get bottoms through no fault of your own because you have no influence over your opponents at duplicate.

[pescetom]

smerriman, on 2020-November-17, 02:28, said:

For example, tournaments provide the double dummy 'par' scores in a printout, but I don't know of anyone who interprets them as 'this is what you should have made'; they're well known to be courtesy only.

Does anyone know of a study comparing DD tricks with actual tricks in expert play, or with expert analysis of the likely outcome in play?
There certainly is a difference, but my gut feeling is that the majority of results would coincide, and that differences of more than one trick would be concentrated in a small percentage of hands.
Differences of PAR score are a different matter and much more extreme of course, because the assumption is that players always and only call game/slam which makes, double opponents going down, make useful sacrifices u.s.w.

[paulg]

pescetom, on 2020-November-17, 04:05, said:

Does anyone know of a study comparing DD tricks with actual tricks in expert play, or with expert analysis of the likely outcome in play?
There certainly is a difference, but my gut feeling is that the majority of results would coincide, and that differences of more than one trick would be concentrated in a small percentage of hands.
Differences of PAR score are a different matter and much more extreme of course, because the assumption is that players always and only call game/slam which makes, double opponents going down, make useful sacrifices u.s.w.

From my results on BBO during lockdown:

Month      Hands 	Played to DD/We play worse than DD    They played worse than DD
April 2020 476   	261 			102                      113
May 2020   392   	204 			88   					98
June 2020  406   	218 			70   					111
July 2020  360   	194 			68   					96
Aug 2020   374   	193 			67   					110
Sep 2020   168        79 			38   					50
Oct 2020   176        85 			34   					56

These results are almost all from IMP games, so the focus is often on making (or beating) the contract rather than overtricks, mainly against expert or advanced opponents.

Results from Double Dummy Solver (http://bridgecaptain...downloadDD.html).

[paulg]

I use PAR and double dummy a lot when I am analyzing a pair's performance and I don't have the time to go through every hand (like the European Championships).

They are two metrics that I use to help identify hands that might be of interest for examination.

[pescetom]

paulg, on 2020-November-17, 06:54, said:

These results are almost all from IMP games, so the focus is often on making (or beating) the contract rather than overtricks, mainly against expert or advanced opponents.

Results from Double Dummy Solver (http://bridgecaptain...downloadDD.html).

Thanks.
That looks like 52% played to DD on average: a bit less than I expected, but still more than half.

[barmar]

One of the biggest advantages of DD is that it's an objective standard. Anything else introduces unpredictable factors like choice of system, expertise of opponents, and psychology.

[Tramticket]

smerriman, on 2020-November-17, 02:28, said:

Do you have any examples of where they're used inappropriately?

I have often sat at a table where declarer has (say) failed to make 4♠ because they didn't drop my singleton king of trumps. Dummy stares at the Bridgemate and chastises partner: "You should have made it".

I think that there is a lot of misunderstanding about what a double dummy solver is and no, it should not be used as the "par" for assessing an individual hand. But aggregate data can be interesting.

If you know and understand how to use a utility like Bridgesolver, it can be an incredibly powerful tool for walking through a hand.

[pescetom]

Tramticket, on 2020-November-19, 10:06, said:

I have often sat at a table where declarer has (say) failed to make 4♠ because they didn't drop my singleton king of trumps. Dummy stares at the Bridgemate and chastises partner: "You should have made it".

I had one this afternoon (a hand like this, fortunately not a partner like that) where on defence against 3NT after leading AK♦ I should have cashed A♥ as my only chance to avoid overtricks... interesting, but not good bridge, even at MP.

Tramticket, on 2020-November-19, 10:06, said:

If you know and understand how to use a utility like Bridgesolver, it can be an incredibly powerful tool for walking through a hand.

Italians say "l'appetito vien mangiando"... it's much clearer than the BBO equivalent, but I would love to be able to turn on and off the technocolor assistance where it is all too easy to fool oneself that a play was obvious.

[helene_t]

To produce a duplicate score without actual duplication, IMP or MP (or Paton or whatever) against PAR is not bad, and surely better than total points or IMP against passout. And as Barry says, it is objective.

Matchpoints against DD Par creates a bit of perverse incentives (for example, if you notice that you could have dropped a sec king offside you know that you are behind DD Par so you have to change your strategy to the assumption that you are NOT in the par contract) so I think I would prefer IMP against DD par.

On BBO, I prefer to score against a match from the Vugraph archieve, though. Scoring against top-level play seems better than DD, but if you want to assure fairness you should probably swap direction halfway within each set so that you don't have a bias in case the EW Vugraph pairs are stronger (or weaker) then the NS Vugraph pairs. And comparing your bidding and play to that of human experts is more interesting than comparing to computer sims.

Using Jack, you can compare against a Monte-Carlo simulated MP field.

[thepossum]

Tramticket, on 2020-November-19, 10:06, said:

If you know and understand how to use a utility like Bridgesolver, it can be an incredibly powerful tool for walking through a hand.

I use Bridgesolver to analyse most of my hands (and sometimes others') to see where I (or often Gib) went wrong

Or maybe more correctly with GiB to see where a simulation differs from DD

I also download into Double Dummy Solver

[thepossum]

smerriman, on 2020-November-17, 02:28, said:

Do you have any examples of where they're used inappropriately?

For example, tournaments provide the double dummy 'par' scores in a printout, but I don't know of anyone who interprets them as 'this is what you should have made'; they're well known to be courtesy only. Like I said, you're meant to analyse them yourself - if you see that double dummy required dropping a singleton king, then you adjust.

If people you know do use them like that, then the problem is their interpretation - not the fact they're provided. There simply isn't anything else that can be easily provided.

They are often used as a comparator in terms of how people have played. For example, I read an article a while ago (by Matthew Kidd) relating to analysis of slams bid and made and he used DD potential slams as comparison. And its highly educational the percentages of slams (from potential DD slams) made by different level players
I believe DD is the most widely used comparator when I read analyses

Oh, nearly forgot. I also often like to play with different hands that people have posted (especially leads or sometimes bidding) and see how different packages simulate (I presume most of them are variants on DD simulations) and how leads differ/compare to expert leads and plays etc
ATM I have bddeal/bcalc/blead (Beling), but also have looked at WBridge, QPlus, and afew other champion computer bridge packages.

I was actually very interested in all the theory and tried to find papers, but havent looked for a few years When my brain was working better I was starting to think about how they could be made better, different approaches. Also was interested in game theory etc etc. And most recently, and most topical have been following discussion on Nic Hammond's work


I had a long life/career/interest in all things mathematical/computational/modeling/AI/statistical/predictive/anything (in very many application areas), and Bridge was a recreation I learned many years ago (and came back to more recently via BBO) so maybe call it quasi-professional, post-professional, semi-retired professional, exhausted professional hobby research and interest

I certainly have no energy to try and do much (in case anyone was worried). I had been considering trying to write articles on anything and everything under the sun but never find time or energy. So prefer just to chat about it instead.

And, also in case anyone wonders or is cocnerned, I spent most of my time out in (I could say the real world) rather than academe so you wont find my name on many things ( a few obscure/specialist papers here and there and buried in some old source code repositories perhaps) But, I have always seen myself as a team/partnership player so hopefully nobody is at all concerned about my interest and contributions, However I will go so far as to say I also regard myself as a seriously cocnerned citizen about many things so may occasionally come across as something of a skeptic or devil's advocate on many issues. But Bridge and games are just a hobby

If anyone wants to contact me (who doesn't know my name) with any concerns about me or my interests they are welcome to do so, but since I hav no Interest at all in Bridge/Bridge software and methods I prefer not to have my name on any posts. That goes for all my non-professional contributions everywhere. They are just conversation. But my identity is hardly a secret and as I said, anyone (who has a trusted name) can contact me if they have any cocnerns about me at all. But my concerns about this world and the power of many intersecting interests. Those concerns are both very personal and also professional/professionally informed. So I hope everyone will understand and respect me keeping my identity to private communication.

There is a saying that knowledge is power. Well these days there seems to be power that trumps knowledge

[hrothgar]

I'm surprised that people haven't brought up one of the most significant ways in which double dummy solvers are currently being used: Trying to detect cheats.

A lot of the metrics that folks like Nicolas Hammond have developed use deviations from double dummy play to assess whether or not a player might be cheating. After the initial assessment, there's a lot more analysis that takes place, but errs compared to DD results are a important initial filter.

[thepossum]

hrothgar, on 2020-November-20, 16:37, said:

I'm surprised that people haven't brought up one of the most significant ways in which double dummy solvers are currently being used: Trying to detect cheats.

A lot of the metrics that folks like Nicolas Hammond have developed use deviations from double dummy play to assess whether or not a player might be cheating. After the initial assessment, there's a lot more analysis that takes place, but errs compared to DD results are a important initial filter.

I did mention I was aware partly of Nic's work and how its being used but do not know enough to discuss it any further. What I am curious about is how that could be done (if its not CIC etc). Is this case is it if play is too similar to DD rather than how a player would genuinely play

Excuse my ignorance of that specific issue but it was the nature of my thread in that DD has both teams knowing all their cards whereas when we play normally we would have say standard leads, based on bidding and contract, signals, working out the lay of the cards as we go etc. But to be honest I don't know enough about how DD is implemented. It just crossed my mind that the best comparator rather than us all looking at DD scores would be (if feasible - which I know its not) a hypothetical play (bidding and card play) between two expert pairs. I know that was mentioned above in the thread

Note, another thing I remember reading (as I said Ive been very tired recently) a few years ago when I first started getting back to Bridge and understanding computer bridge was how DD had actually impacted on the theory of how the game is played, and even the way cards tend to be played now. I cant remember which paper apologies (may have been one of the original Ginsberg GiB papers )

Something I find a bit frustrating, maybe just the nature of where papers are published and the majority of people working in the space, is that they are extremely technical (in terms of detail of the maths and algorithms/techniques used), and very little is written at a more accessible level. My brain isnt really up to that level of maths these days (if it ever was)

I'm actually very interested in any literature about the whole philosophy of, approach to computerised Bridge - if anyone has good refs

Just as an example, because of the way the computerisation is simplified (by necessity) into full information simulations, that is problamtic in intself in analysing how human's actually play the game when it is about communication, building up a picture through the game, simple initial assumptions etc. Apologies for not having delved into the code or methods to know how they are actually implemented, and for trying to think out loud. But there seems something of a philosophical disconnect between the real game and sophisitcated hi-powered analytics and simulations

I'm also trying to read stuff related to the different types of Bridge and how those more game-theoretic type factors could come into play. As I said, all well beyond my brain but interesting in finding dummies' guides

Whenever I search for things like Bridge and Game Theory a certain person's name keeps popping up, and I'm very nervous showing my ignorance in present company

[hrothgar]

thepossum, on 2020-November-20, 16:59, said:

Excuse my ignorance of that specific issue but it was the nature of my thread in that DD has both teams knowing all their cards whereas when we play normally we would have say standard leads, based on bidding and contract, signals, working out the lay of the cards as we go etc. But to be honest I don't know enough about how DD is implemented. It just crossed my mind that the best comparator rather than us all looking at DD scores would be (if feasible - which I know its not) a hypothetical play (bidding and card play) between two expert pairs. I know that was mentioned above in the thread

You don't really need to know how double dummy solvers are implemented to understand their strengths (and their limitations)

The big advantage of double dummy solvers is that they

1. Exist
2. Work
3. Are reasonable inexpensive from a computational perspective

The later is particularly important because it allows folks to use double dummy solvers for a whole bunch of interesting stuff such as creating single dummy solvers that are able to perform at extremely high levels. (By which I mean that a double dummy based play engine was able to score extremely well on a par contest a few years back and outperformed a whole bunch of world class players)

The weakness is that there are certainly types of play positions that they just don't understand well.

My impression is that double dummy techniques have not been proved to be biased. (By which I mean that they are imperfect, but given enough samples, double dummy techniques seem to be relatively good at estimating playing strength of hands and the like)

[thepossum]

hrothgar, on 2020-November-20, 17:48, said:

You don't really need to know how double dummy solvers are implemented to understand their strengths (and their limitations)

The big advantage of double dummy solvers is that they

1. Exist
2. Work
3. Are reasonable inexpensive from a computational perspective

The later is particularly important because it allows folks to use double dummy solvers for a whole bunch of interesting stuff such as creating single dummy solvers that are able to perform at extremely high levels. (By which I mean that a double dummy based play engine was able to score extremely well on a par contest a few years back and outperformed a whole bunch of world class players)

The weakness is that there are certainly types of play positions that they just don't understand well.

My impression is that double dummy techniques have not been proved to be biased. (By which I mean that they are imperfect, but given enough samples, double dummy techniques seem to be relatively good at estimating playing strength of hands and the like)

Thx Richard.

I'm starting to get a greater understanding on how they work with the simulations etc

Regarding the Game Theory stuff, I referred indirectly to your name being someone interested in that area. Has much been written yet
I know you may not want to talk about your own research interests but if you can point me toward anything I would be interested

What crossed my mind in relation to analysing behaviour in tournaments is how hard it would be to incorporate that level of behavioural change into any algorithm for simluating real tournament play and/or detecting changes

Sorry for keeping on editing and adding extra stuff, but I think about things after I've posted. In any tournament the real set of possible scores is essentially what happens with that group of people is it not?

[hrothgar]

thepossum, on 2020-November-20, 18:35, said:

Regarding the Game Theory stuff, I referred indirectly to your name being someone interested in that area. Has much been written yet
I know you may not want to talk about your own research interests but if you can point me toward anything I would be interested

I did my original block of graduate work in mathematical economics, focusing on game theory, however, that's now close to 30 years back. I'm really not up on recent developments in the field.

With this said and done, my impression is that you aren't going to see much serious work on game theory within the confines of bridge.

Let's break the game theorists into two different sets of people:

First, you have game theorists who aren't particularly interested in bridge. My suspicion is that - for the most part - they will prefer to focus on much more simple games. Most of the "classic" problems that you might want to study can be illustrated with games that are much more accessible to the average reader. As such, there are clear incentives to steer clear of bridge where you need to spend much more time describing the rules and idiosyncrasies of the game.

Second, there are game theorists who play bridge and find the game interesting. However, this can actually be an impediment to doing research since many of the areas that one would want to focus on are banned by legislative fiat.

Consider the following simple example: I have long claimed that the notion of a "psych" is an anachronism. It does not actually describe the way in which people play. it would be better if the laws, rules, and regulations replaced this with notions like mixed strategies. This ain't ever going to happen, so what's the point of doing any formal study in these areas? In a similar vein, there are a whole bunch of restrictions around bidding systems. What's the point of investing a bunch of time and effort in this area if the powers that be are going to legislate better systems out of existence?

Note: There are some areas where people have done good game theoretic analysis. Treating restricted choice as a mixed strategy is an obvious example. There are other areas in which game theoretic analysis would be really interesting. Trading off between expected value and variance, both in choice of bidding systems and choice of specific plays being a classic example. Another good one is looking at populations of bidding systems over time (Look at the work that Axelrod did back in the early seventies around populations of programs playing Tic Tac Toe and extent this to bridge).

As far as I am aware of, the only really serious work going on in formally studying bridge these days is the stuff that NUKK-AI is doing over in France, however, they are much more focused on areas around semantics, language, and how to exchanging information between people and computers.