Software Reviews

[pdmunro]

I would like to start a topic, Software Reviews.

 

I have been impressed with some recent software I downloaded.

 

Have a look at the list on http://www.jackbridge.com/ewkprt.htm

[pdmunro]

Autogenerate Teaching Hands

 

It seems Yves Costel's WBridge5 has a unique capability. It can be used to generate hands that conform to a bidding sequence. It can then autoplay those hands.

 

The deal and play are saved as PBN files.

 

Open WBridge5 and click "Donne | Donne Manuelle... | Oui"

to get a screen titled "Saise manuelle de la donne" (Manual entry of deals)

At the top left, choose "Donneur" (Dealer), by clicking, say, "Est" (East).

Now click "Encheres" (Bids) from the top-bar options.

 

Next, click on the bid options, lower-right, to generate, say, a limit raise

- 1H - 3H

- 4H - -

-

 

Click "Save" (floppy disk symbol). You are asked for a filename; how many deals you want; and "Avec Jeau de la carte" (whether you want the cards autoplayed).

 

It can take a while to play the cards.

 

A teacher might want to upload these pbn files to BBO and add commentary.

 

A few points

 

1) To get WBridge5 to bid 1H-3H as a limit raise I had to first uncheck various conventions such as Truscott 2NT. This is done from the main screen. Click "Option | Encheres..." and uncheck "2SA Truscott; 2SA & 3SA fitté; 2SA spécial; and Bergen".

 

2) Importing files to BBO-online-MOVIE raises the dealer and vulnerability issue. The best options seems to be to set up a teaching table because you can specify the number of the deal you are uploading. Looking at the following table re BBO numbering, you can see that a deal where N is the dealer and NS are vulnerable has to be numbered 5 or 21.

 

Brd Brd Dealer Vul

1 17 N Nil

2 18 E NS

3 19 S EW

4 20 W Both

5 21 N NS

6 22 E EW

7 23 S Both

8 24 W Nil

9 25 N EW

10 26 E Both

11 27 S Nil

12 28 W NS

13 29 N Both

14 30 E Nil

15 31 S NS

16 32 W EW

 

3) If you wanted to go in the reverse direction and play BBO files on WBridge5, you would need to use Wayne Burrows' lin-to-pbn converter. http://www.ebridgenz.com/download.html

[pdmunro]

Play SAYC or add your own bidding system

 

I dowloaded the free demonstration copy of Micro Bridge

http://www.threeweb.ad.jp/~mcbridge/dload_e1.htm

 

It is great. An excellent interface. The free version has SAYC, as well as Acol. There is no time limit as far as I can tell.

 

A good aid for beginners is the System Summary. First choose the bidding system, the computer is to use, by clicking "Tools | Bidding System Manager". Next look at the meaning of any bids you are not sure of, by clicking "Bidding | Summary" If you print it out, you get a very complete SAYC system. This may be useful to someone constructing a Full Disclosure file.

 

The version, that you buy, allows you to set up your own bidding system. I think this function is unique to Micro Bridge. Quite an achievement.

[DenisO]

. 

 

The version, that you buy, allows you to set up your own bidding system.  I think this function is unique to Micro Bridge.  Quite an achievement.

I'd be interested in how easy it is to set up your own bidding system, if you've bought the software. My impression, although I could well be wrong, was that it was more like adding gadgets etc to one of the supplied systems. OTOH, Oxford Bridge(in your list above) lets you build your own system from scratch. I have that program but it is a huge task to build your own system. I started building a Precision system for practising but abandoned it when Bridge Baron added a Precision version :lol:

[pdmunro]

I'd be interested in how easy it is to set up your own bidding system, if you've bought the software.

 

I haven't tried that. Not my cup of tea. I think you would be able to tell what's involved from the free demo version.

[Free]

Jack 3.0 has MOSCITO <_< It's a very good program imo, lots of conventions, but his defense sometimes completely messes up.

 

I'm having a look at easybridge and wbridge at the moment, so no comment on these yet.

[DenisO]

Autogenerate Teaching Hands

 

It seems Yves Costel's WBridge5 has a unique capability. It can be used to generate hands that conform to a bidding sequence.  It can then autoplay those hands. 

 

The deal and play are saved as PBN files. 

 

This looks a really interesting feature. However when I tried it (8 boards), I only get the play to the first trick in the PBN file. Any suggestions on what I'm doing wrong?

[pdmunro]

Like all computer things, I would start small. Below is what I get with 2 deals. It auto generated all the deals and play. I simply set the bidding as 1H-3H-4H. Interestingly, when it rebid Board 2 the program did not bid on to 4H.

 

[Event "#"]

[board "1"]

[Dealer "E"]

[Date "2006.01.11"]

[North "WBridge5"]

[East "WBridge5"]

[south "Joueur"]

[West "WBridge5"]

[Vulnerable "Both"]

[Deal "N:.J985.AT753.QJT3 AK975.T7.J86.954 632.AKQ64.K9.A62 QJT84.32.Q42.K87"]

{

Nord

S:

H: J 9 8 5

D: A10 7 5 3

C: Q J10 3

Ouest Est

S: Q J10 8 4 S: A K 9 7 5

H: 3 2 H:10 7

D: Q 4 2 D: J 8 6

C: K 8 7 C: 9 5 4

Sud

S: 6 3 2

H: A K Q 6 4

D: K 9

C: A 6 2

}

[Declarer "S"]

[Contract "4H"]

[Result "12"]

[score "NS 680"]

[Auction "E"]

PASS 1H PASS 3H

PASS 4H PASS PASS

PASS

[Play "W"]

H2 H5 H7 HA

H3 H8 HT HK

D2 D3 D6 DK

C7 C3 C4 CA

D4 DA D8 D9

DQ DT DJ H6

S4 H9 S5 S3

S8 D7 C5 C2

ST D5 C9 C6

C8 CQ S7 H4

SJ CT S9 S6

CK CJ SK HQ

SQ HJ SA S2

 

[Event "#"]

[board "2"]

[Dealer "E"]

[Date "2006.01.11"]

[North "WBridge5"]

[East "WBridge5"]

[south "Joueur"]

[West "WBridge5"]

[Vulnerable "None"]

[Deal "N:KQJ3.T942.4.Q875 9762.83.A732.A32 54.AKJ76.QJ5.K64 AT8.Q5.KT986.JT9"]

{

Nord

S: K Q J 3

H:10 9 4 2

D: 4

C: Q 8 7 5

Ouest Est

S: A10 8 S: 9 7 6 2

H: Q 5 H: 8 3

D: K10 9 8 6 D: A 7 3 2

C: J10 9 C: A 3 2

Sud

S: 5 4

H: A K J 7 6

D: Q J 5

C: K 6 4

}

[Declarer "S"]

[Contract "3H"]

[Result "9"]

[score "NS 140"]

[Auction "E"]

PASS 1H PASS 3H

PASS PASS PASS

[Play "W"]

CJ C5 C2 CK

H5 H2 H8 HA

DK D4 D7 DQ

CT C7 C3 C6

C9 CQ CA C4

SA S3 S7 S5

DT H4 DA D5

HQ HT H3 HK

D6 C8 D2 DJ

S8 SJ S2 S4

ST SK S6 H7

D8 H9 D3 H6

D9 SQ S9 HJ

 

As I wrote previously, the analysis is fairly slow. This is a function of the simulations it attempts. I think it does about 10 per sec at any one turn.

 

There is a bit of a pause after the autoplay of deal 1. Maybe you needed to wait a bit.

 

Use Notepad to see what the pbn or lin files look like. They are just text. Excuse me if I am just telling you things you already know.

 

Perhaps just play a few games of random deals yourself. There may be something I did that is a normal part of the program that I am not telling you because I do it without thinking.

 

There are some Help files in English on the main web site. http://perso.chello.fr/users/y/yvescostel/index.htm

I also use the translation tool that Google provides in order to translate their web pages.

[Al_U_Card]

I'd be interested in how easy it is to set up your own bidding system, if you've bought the software.

 

I haven't tried that. Not my cup of tea. I think you would be able to tell what's involved from the free demo version.

I bought Positronic Bridge about 15 years ago. It had programming engrams that would , in "teaching" mode, build files that grew with practice to be able to play any consistent system.

 

I "taught" it my 10-12 weak NT system (pretty much home-made with full 2/1 major bidding sequences) After about 1000 practice hands it could do a credible job. The play of the cards left a lot to be desired so I never really practiced with it but still have all the files.

[pdmunro]

Autogenerate Teaching Hands

 

It seems Yves Costel's WBridge5 has a unique capability. It can be used to generate hands that conform to a bidding sequence.  It can then autoplay those hands. 

 

The deal and play are saved as PBN files. 

 

This looks a really interesting feature. However when I tried it (8 boards), I only get the play to the first trick in the PBN file. Any suggestions on what I'm doing wrong?

 

Ah sleep, again it works its magic. I have just woken. I think you had the stop button (bottom right) depressed. This pauses the program at the end of every trick. Give it a click to change this mode. And try everything again.

[pdmunro]

Minibridge software

 

There is free MiniBridge software available. You play against the computer. Might be a good way to learn card play.

[chicken]

as a converter, generator, and analyzer still dealmaster pro with implemented deep finesse is unbeatable. only 4 the presentation of hands u need a program with a nice easy interface , here i use jackbridge.

[pigpenz]

I think for analysis nothing beats Bridge Base III.....Hint Fred!

I would still buy it if it were available but its only a DOS program.

 

I remember seeing it in action it will give you suit analysis as each card is played and will show you the %'s as each card is played etc. Unfortuanagtelly I think Fred said it was a C complited program, too bad it couldnt be redone for Windows.

[Sigi_BC84]

I bought Positronic Bridge about 15 years ago.  It had programming engrams that would , in "teaching" mode, build files that grew with practice to be able to play any consistent system. 

 

I "taught" it my 10-12 weak NT system (pretty much home-made with full 2/1 major bidding sequences)  After about 1000 practice hands it could do a credible job.  The play of the cards left a lot to be desired so I never really practiced with it but still have all the files.

Wow, this is cool. Combine that with GIB's playing abilities and you'll get a nice computer contestant. Unfortunately probably not many people will go so far as to practice 1000 hands with the computer (of which the first few hundred will probably be quite frustrating). Maybe one could efficiently seed the training database with information about the system and then improve on that.

 

I'm wondering why Matt Ginsberg doesn't mention this approach in his paper about GIB (maybe I've missed it, however).

 

--Sigi

[CarlRitner]

I'm wondering why Matt Ginsberg doesn't mention this approach in his paper about GIB (maybe I've missed it, however).

 

--Sigi

I should preface all of this by saying the following is my impression.

 

Matt's basic approach is to let GIB "solve" bridge by search and simulation, without domain-specific knowledge (or at least minimising this). That precludes the use of "learning" techniques.

 

One can eventually teach Positronic to lead 4th best by repeatedly leading 4th best, but then it takes nearly a lifetime to teach it all of the exceptions (plus you can never really teach it *why*). Alternately, one could build a large opening lead table, but then you are transferring domain-specific (and likely person-specific) information to the program, and if that's wrong, or changes over time, you need to revise the program.

 

GIB relies more on general rules and strategy non-specific to bridge. Now obviously, some specific knowledge such as the scoring tables and laws of play must be built into the structure. And the bidding database must be told what each call means and what requirements needs to be present to fire the rules. But wherever possible, Matt programmed GIB to solve card play by applying risk/reward algorithms and simulations (double dummy) which is essentially brute-force search enhanced by some advancements in implementation.

 

This is part of the reason why GIB's mistakes tend to look goofy to us humans. We avoid the obvious silly play, whereas GIB explores the possibilities and takes the most promising path it can find in limited time, and it has no idea of what silly or goofy means, and couldn't care less. It may not make the same mistake later, because the simulation set is different, but it doesn't "remember" the problem or the answer, and it will re-solve it every time the same problem occurs.

 

There are other approaches to computer bridge, such as data mining and rule-based expert systems, but Positronic and GIB are probably at polar opposites of the spectrum.

 

Hopefully Matt will be available someday to clarify and correct me.

 

Cheers,

 

Carl

www.carlritner.com <-- bridge books, cordless!

[Sigi_BC84]

I'm wondering why Matt Ginsberg doesn't mention this approach in his paper about GIB (maybe I've missed it, however).

 

--Sigi

Matt's basic approach is to let GIB "solve" bridge by search and simulation, without domain-specific knowledge (or at least minimising this). That precludes the use of "learning" techniques.

[...]

There are other approaches to computer bridge, such as data mining and rule-based expert systems, but Positronic and GIB are probably at polar opposites of the spectrum.

I was simply wondering why Matt's paper from 2000 didn't mention this learning approach. He makes it very clear that he thinks brute force and simulation is superior to applying expert knowledge during play. However he also admits that GIB is much weaker during the auction. This still holds true with the current incarnation of the program (the paper is from several years ago).

 

I'm only saying that it looks promising to me at first sight to try to improve GIB's bidding by implementing some kind of learning/mining/expert system. Probably different approaches are needed concerning bidding and play. I also see major problems in communicating the meaning of opposing bids to the computer. To play "real" bridge, the computer has to be able to understand simple explanations, as there is no time to formulate complicated expressions describing a call during play.

 

--Sigi

[CarlRitner]

I'm only saying that it looks promising to me at first sight to try to improve GIB's bidding by implementing some kind of learning/mining/expert system. Probably different approaches are needed concerning bidding and play. I also see major problems in communicating the meaning of opposing bids to the computer. To play "real" bridge, the computer has to be able to understand simple explanations, as there is no time to formulate complicated expressions describing a call during play.

 

--Sigi

I agree completely. Computer bridge bidding needs a different approach than currently implemented. The disclosure issues -- getting useful and accurate info from your opponents in a transferrable format, as well as telling the opps everything that is a partnership agreement (after all, they deserve to know) -- this is a daunting challenge.

 

I have looked at most of the past approaches, and I don't see the silver bullet yet. If you reduce the problem to all players being PC-based and playing the exact same system, you still have the enormous number of plausible auction sequences to deal with. Matt has made inroads here with general rules, but it seems today that almost every bid has some conventional component, such that the natural inferences of a call need to be (almost) uniquely adjusted or tweaked.

 

Perhaps that is why progress on GIB's bidding database is dormant; why shovel grains of sand when the beach is essentially infinite?

 

Cheers,

Carl

[hrothgar]

I agree completely. Computer bridge bidding needs a different approach than currently implemented. The disclosure issues -- getting useful and accurate info from your opponents in a transferrable format, as well as telling the opps everything that is a partnership agreement (after all, they deserve to know) -- this is a daunting challenge.

I've spent a fair amount of time thinking about disclosure issues:

 

I think that there is a fairly simple way to handle disclosure (at least so far as computers are concerned). Disclosure systems are intended to provide players with information regarding the set of hands consistent with bid XYZ. Why not simply provide a set of hands as the disclosure mechanism. These hands could be drawn from past history using an application like BridgeBrowser. If there were not sufficient hands to provide a statistically significant sample a dealer program could be used to bulk things out.

 

This approach would would work extremely well for programs which use simulations in bidding/play. Systems that used a rules based approach would require a mechanism to collapse the set of hands into an expression that the computer was capable of understanding.

[EarlPurple]

The problem with Jack is that it thinks the same amount of time on every bid/play so it gets tiresome to wait if you set it to a more expert level.

 

I would prefer it to use its time when it has something to think about and otherwise to play at a much faster tempo.

 

In addition, a downside of Jack is that you cannot play your own system on it (i.e. a custom system) and if you try playing one against it you cannot explain to Jack what your bids really mean so it is clueless as to how to defend.

 

My only other issue is that it asks me to insert disks rather far too often. It would also be nice to be able to load BBO hands directly in or even get it to plug into BBO when I'm logged in there so I don't have to type the hands in (when I'm kibbing, of course). Its strongest point is its double-dummy analysis which tells you exactly how many tricks can be made at any stage in the play (including before it).

 

It would also be nice if it would save my settings (i.e. all players, etc) instead of reverting to me being South only.

[ArcLight]

Where is the GIB paper?

 

Does the way GIB play on defense work like this?

1. Generate a number of hands (H), constrained by what is known from the bididng and play

2. Solve them Double Dummy and see what solution (S) is most common across all the hands generated.

3. Select that card

 

repeat this process after each card 13 times.

 

 

 

>Perhaps that is why progress on GIB's bidding database is dormant; why shovel grains of sand when the beach is essentially infinite?

 

I was told a while back that Mr. Ginsburg hasn't worked on GIB for a while. I think its because he is working on a job that pays real money and he doesn't have much time.

[Sigi_BC84]

I think that there is a fairly simple way to handle disclosure (at least so far as computers are concerned). Disclosure systems are intended to provide players with information regarding the set of hands consistent with bid XYZ. Why not simply provide a set of hands as the disclosure mechanism. These hands could be drawn from past history using an application like BridgeBrowser. If there were not sufficient hands to provide a statistically significant sample a dealer program could be used to bulk things out.

Very nice idea, but incompatible with "real" bridge. Ideally the computer accepts explanations as they are given in f2f bridge across the table (or typed into an alert window on BBO). To correctly interpret such information, some natural language processing would be needed as well as an expert system that would, for example, know what "capp" means.

 

To implement your "disclosure by example" scheme, the pair opposing the computer would have to carry around a sufficiently large example database (or have enough records in BridgeBrowser or BBO or whatever you'd be using). This is often not the case. Furthermore you'll be ending up without enough information too often. Take denial cuebids after a relay sequence as an example. The denial cues are very easily formalized in the context of the preceding relays. However, finding a sample for a given denial cue sequence will be hard (since it is rather unlikely that a sufficient number of identical sequences has occured so far).

 

--Sigi

[inquiry]

It would also be nice to be able to load BBO hands directly in or even get it to plug into BBO when I'm logged in there so I don't have to type the hands in (when I'm kibbing, of course). Its strongest point is its double-dummy analysis which tells you exactly how many tricks can be made at any stage in the play (including before it).

You can load BBO hands into jack.

 

Try using either cascade's wonderful linconverter or Richard Pavelik's conversion program to go from lin to pbn and load the pbn files. Or, if you play in Homebase and/or purchase rec software's Bridgebrowser program you can export any hands you like to PBN format for use in JACK.

 

If I remember correctly, if you use cascade's lin converter, you may have to open the file and delete a non-printing character from the end of the file to get jack to read it. Files coverted with Pavelik's dos utility and those exported from Bridgebrowser work fine.

[Sigi_BC84]

Does the way GIB play on defense work like this?

1. Generate a number of hands (H), constrained by what is known from the bididng and play

2. Solve them Double Dummy and see what solution (S) is most common across all the hands generated.

3. Select that card

repeat this process after each card  13 times.

This is how it basically works on both declaring and defense. AFAIK MG has improved on this idea to make decisions during later stages of play when GIB has collected enough information -- the computer then tries to select the best line of play rather than just solving the problem locally.

 

The paper can be found here: http://www.jair.org/abstracts/ginsberg01a.html

 

--Sigi

[pigpenz]

Where is the GIB paper?

 

Does the way GIB play on defense work like this?

1. Generate a number of hands (H), constrained by what is known from the bididng and play

2. Solve them Double Dummy and see what solution (S) is most common across all the hands generated.

3. Select that card

 

repeat this process after each card 13 times.

 

 

 

>Perhaps that is why progress on GIB's bidding database is dormant; why shovel grains of sand when the beach is essentially infinite?

 

I was told a while back that Mr. Ginsburg hasn't worked on GIB for a while. I think its because he is working on a job that pays real money and he doesn't have much time.

Well the commercial version of GIB has some other helpful tools very bridge analysis. And I am not sure as to how this works with the playing GIB...but there is a file in GIB called bridge.exe which allows you input hands at matchpoints or imps include an auction and it will break down which bid it thinks is correct or lead. I think this is probably what happens when you ask gib for help but is more extensive in its search for the correct or % wise answer.

[pdmunro]

Two interesting links re simulation are "How Computers Play Bridge" by two Bridge Baron programmers, and an extended interview with Yves Costel the programmer who wrote WBridge5, the winner of the 2005 Computer Bridge Championship. I downloaded the zip file and then used Google translator to get the gist of his ideas.