Self-learning FD

[sfbp]

Fred,

 

I have to be honest, I gave up on FD when I saw that I would have to write my own program to generate the sequences for it (let's not debate that conclusion for now, we discussed what I think is the critical thing I would need to restart efforts)

 

This is an attempt to be constructive: most people still don't use FD and many find that the FD cards that are there are irritating - either because they are filled out wrong, or ignored.

 

What about a self-learning switch for FD?

 

This would mean that you could describe your auction as you play, and the default would be for FD to remember the meanings of your bids. Of course you might have to take the step of selecting "your system" so that the bids you make in precision wouldn't start to overwrite the ones when you play 2/1. But that's not a big deal for most people who play only one system.

 

Establishing a bid usage history of sorts might have other advantages too ;)

 

I suggest that there might be available to the player a simple way to alter the description of a bid that pops up from previous sessions.

 

The default would be that explanations and alerts would be copied in AT LEAST when card was blank in those spots, and an option when card wasnt blank.

 

This also means that the common sequences will get covered, and the system freaks will now have enough time to spend on the uncommon sequences.

 

Sound like a plan?

 

Stephen

[csdenmark]

Fred,

 

I have to be honest, I gave up on FD when I saw that I would have to write my own program to generate the sequences for it (let's not debate that conclusion for now, we discussed what I think is the critical thing I would need to restart efforts)

 

This is an attempt to be constructive: most people still don't use FD and many find that the FD cards that are there are irritating - either because they are filled out wrong, or ignored.

 

What about a self-learning switch for FD?

 

This would mean that you could describe your auction as you play, and the default would be for FD to remember the meanings of your bids. Of course you might have to take the step of selecting "your system" so that the bids you make in precision wouldn't start to overwrite the ones when you play 2/1. But that's not a big deal for most people who play only one system.

 

Establishing a bid usage history of sorts might have other advantages too ;)

 

I suggest that there might be available to the player a simple way to alter the description of a bid that pops up from previous sessions.

 

The default would be that explanations and alerts would be copied in AT LEAST when card was blank in those spots, and an option when card wasnt blank.

 

This also means that the common sequences will get covered, and the system freaks will now have enough time to spend on the uncommon sequences.

 

Sound like a plan?

 

Stephen

Looks interesting Stephen. I understand you are a programmer - how about offering your service?

 

As I see there will need to be a mechanism to secure and double-check modifications so they will only take effect if carefully considered by the owner and not by some kind of mistake, human mistake or simple misbid.

 

If you go to ZONE.com you will see their convention cards, a popping up format in plain text, generally to be used by approx. 40-50%. They all play standard systems(SAYC/2o1), the general standard level is much the same. I wonder the reason for the more general usage of convention cards, especially because they need so less there than by BBO. Any explanation of that phenomen Stephen?

 

As approx. 95% of the players need nothing else than a default card and for the rest they will mostly be able to produce a decent convention card in 2 hours. It looks to me that these constant commando raids against a strong tool working very well is not well in place. Everything can be better of course but the lack of usage first of all reflects lack of seriousity.

[A2003]

suggestions:

Max length column may be deleted.

HCP column may be replaced in that space.

Disposition: "other" may be added to include user defined disposition(eg.. splinter).

"Qualify" block what is that mean? when something like Michaels written in that column, it is not clear what is happening during the bidding (when opponent bids 2H over 1H).

 

Currently the options are Vul (9 Choices), dealer (7 choices)

If this format is changed to the below option will it save many combination?

Board Number (1-36), player position: 4 choices (E, W, N, S).

Because Vul, Non vul and dealer are all fixed for these boards.

Board number is specified example 1 thru 36, player position east or west or north or south

May be it is easier to fill some details for some positions as play goes on.

[OrShoham]

Sounds like a novel idea to me, but I don't think a system could be expected to read the BBO bidding and fill your card based on that. It would be more practical, I think, to have a program that reads a directory of LIN files. Using simple-enough algorithms, you could probably extrapolate a significant number of bids from a smallish set. Conventions, in particular, would be easy to locate and extrapolate to all relevant situations (i.e. Drury for all 1H and 1S openers in 3rd/4th seat, any vul). The results could then be exported to a FS-readable format.

 

If nobody else takes it up, I may make an effort at this when I have time - sadly, this will not be before the school year ends, so we're looking at a goodish 6 months at best.

[Gerben42]

I think for this you do not need BBO at all if you can just fish out the bids when they are sent to your computer. But a great idea indeed I hope it is realizable.

[DrTodd13]

When FD was introduced, I personally thought that this is one of the primary ways it would be done. When you make a bid that is unknown then the box pops up and asks you for the explanation and thereafter the bid is in the FD file. I still think it makes sense to do this for uncontested auctions. Many of us here are programmers and have offered to help but understandably Fred and Uday have not allowed anybody else to see the source code.

 

At this point, I feel the major limitation of FD is dealing with competitive situations. My responses depend on the meaning of the competitive bid. There is no way to specify different possible meanings for the same bid by opps and then to react differently. Moreover, even if you could specify that you couldn't automate the process because you'd have to match what the opp _intended_ to show by his bid with the various options the FD users have specified. None of the options might match or multiple ones might match. Plus, you can't know what the opp intended to show so if he psyched or made a mistake then if the system tried to do something automatic it might reveal the mistake or the psyche by matching it with a rule that the opp didn't intend for it to fall under. The only answer would be to ask the opp what was meant by the bid but you'd need a much better way of describing a hand than what is present now. A textual description is not going to allow anything to be automated.

 

I spent quite a bit of time looking at how to make something like FD that would automatically learn and that would be precise and deal with the different meanings that opps may give to bids. I could not convince myself that this was even possible but I did convince myself that if it were possible it would be so complicated that nobody would ever use it.

[inquiry]

One artificial way to do this is to use BRIDGEBROWSER when you sit down for a long match against a specific pair.

 

1. Search for them as a partnership

2. Turn on "bid analysis"

3. When they bid, you can look at the type of hands they have for specific calls.

 

You can quickly find all the hands, you can quickly sort them by auction (first 7 bids only, but all subsequent auctions are there. Looking at the table through, say the 14 hands that began 1H-P-2C-P-2H- to find out what type of hands responder has for a jump to 4C would require you to to that auction then look at all 14 hands for which ones had a 4C jump (if any).

 

So at the table, search a search is hardly useful (I use this follow FANTUNES auctions when they play in BBO, so as a kibitzer it works welll, and they have played hundreds and hundreds of hands, and the list is growing all the time).

 

I suspect it might be possible to harnest the data from bridebrowser which gives you a lot of statistics (average hcp, st dev, suit lenght, suit quality) to help with an automated FD development type of thing -- as it will sort the hands by auction and partnerships for you quite easily, but as a tool for the 100's of thousands of players on BBO, this will not work.

[csdenmark]

When FD was introduced, I personally thought that this is one of the primary ways it would be done.  When you make a bid that is unknown then the box pops up and asks you for the explanation and thereafter the bid is in the FD file.  I still think it makes sense to do this for uncontested auctions.  Many of us here are programmers and have offered to help but understandably Fred and Uday have not allowed anybody else to see the source code.

Exactly - therefore I often wonder why it is needed to discuss unendlessly about what looks like fairly simple to do for any solid computer-specialist. I think not everything need to be scheduled by Fred but of course the correct specifications, the correct intensions and the needed loyalty must be there. Ought not to be a problem.

 

Unfortunately I am not a programmer - I am a banker but working with re-engineering for many years. So I have good knowledge but this kind of specific tasks I will not be able to handle.

 

Bidedit looks like a small straightforward application(128KB + 3KB for a configure file). I think it will be fairly simple to construct such a stand-alone application in Assembler or C++. It is not needed that Fred will use it - as long it produces bss-files in correct format according to configuration file.

[DrTodd13]

The problem isn't the bss file format. Any reasonable programmer could write a bss file editor. The problem is that you need the GUI to interface with the bss editor in a way that it currently isn't doing. To make GUI changes you Fred to do it.

[hrothgar]

When FD was introduced, I personally thought that this is one of the primary ways it would be done.  When you make a bid that is unknown then the box pops up and asks you for the explanation and thereafter the bid is in the FD file.  I still think it makes sense to do this for uncontested auctions.  Many of us here are programmers and have offered to help but understandably Fred and Uday have not allowed anybody else to see the source code.

Exactly - therefore I often wonder why it is needed to discuss unendlessly about what looks like fairly simple to do for any solid computer-specialist. I think not everything need to be scheduled by Fred but of course the correct specifications, the correct intensions and the needed loyalty must be there. Ought not to be a problem.

 

Unfortunately I am not a programmer - I am a banker but working with re-engineering for many years. So I have good knowledge but this kind of specific tasks I will not be able to handle.

 

Bidedit looks like a small straightforward application(128KB + 3KB for a configure file). I think it will be fairly simple to construct such a stand-alone application in Assembler or C++. It is not needed that Fred will use it - as long it produces bss-files in correct format according to configuration file.

I think that you are missing the point:

 

I'm not sure how difficult it would be to design an FD system that featured some kind of adaptive learning mode, however, I think that it would be substantially more difficult than you let on. You can't design this type of system in isolation. By definition it needs to get integrated with the rest of the BBO application. This could (potentially) make the whole development integration process enormously more difficult. Even if you successfully build this type intelligent FD application you still need some kind of editor. You need some mechanism to go in and correct spelling errors, typos, mis-information, and the like. An intelligent FD system will certainly ease some apects of the design process, but its no panacea.

 

Equally significant; Assume for the moment that it is relatively simple to build an intelligent FD application. Do you think that this feature is more or less important than any one of 50 other enhancements that Fred and Uday could be working on? I'm a big fan of the FD application, but I think that this is much less important than any number of other enhancements that I'd like to see

 

1. Overhaul the chat system

2. Permanent floating Indy

3. Bottom up matching system for finding team games

4. yada

5. yada

6. yada

 

BTW, if I were enhancing the FD system, I'd focus on trying to build more modular conventions and link these into a graphical convention card. Unfortunately, I haven't really figured out how to chop a bidding system into a set of separate modules. I tried to raise this issue a couple tims on the forums, but no one seemed that interested.

[csdenmark]

No matter how simple it might be to create such a tool it will be nothing than waste of time and good spirit.

 

95% of the players have problems to know their fairly simple basic systems adding a few conventions. None of these are relevant users - they need default card.

 

Several of the rest, myself inclusive, have already entered the relevant system(s) and we will neither be relevant.

 

Well for practicing programming skills the task might be useful - but for nothing else.

 

I agree with Richard that important tasks ought to have priority. I will also be able to name a list like Richard even it will be different.

 

What is GUI?

[csdenmark]

The problem isn't the bss file format.  Any reasonable programmer could write a bss file editor.  The problem is that you need the GUI to interface with the bss editor in a way that it currently isn't doing.  To make GUI changes you Fred to do it.

I think and hope you know that it is NOT your downloaded Bidedit version you watch on BBO. You upload the BSS-file and watch it on Bidedit version residing on BBO server which might be with different configure settings than your private version.

[1eyedjack]

Coming at this from the viewpoint of a NON-programmer, who has no idea how difficult it is, one problem that I have with the heuristic approach:

 

Say the auction goes

1C-(1D)-1H

and you wish to enter that 1H shows (say) 5+ Hearts, forcing, and a smattering of points.

Would FD then extrapolate this so that when you get the auction

1C-(1H)-1S

it then introduces that rule? Would you want to trust FD to get this extrapolation right all the time?

Then, what if the opponents alert the 1D overcall and say it shows Diamonds and Hearts, and your 1H was, in light of that, an artificial bid? So every time the opponents intervene (and they will be playing a wide variety of methods) you will have to enter their bids' meanings before FD can hope to compute the meaning of your bids.

 

All sounds horrendously difficult. Glad someone else is trying this, not me.

[hotShot]

What could a self learning FD do?

 

It could create a database entry for every bid stating which partner and how many HCP, ZAR-Points and what kind of distribution he was holding.

 

It could than disclose a message like that:

 

player XYZ and ABC had the bidding sequence

1[cl]

5 times before.
[space] [space] [space] [space] [space] [space] [space] [space] [space] [space] min [space] [space] [space]max [space] [space] [space] [space]avg
HCP [space] [space] [space] [space] [space] [space] [space] [space]12 [space] [space] [space] [space] 17 [space] [space] [space] [space]12.5
length [cl] [space] [space] [space] 3 [space] [space] [space] [space] [space] 5 [space] [space] [space] [space] [space]4.2 
length [di] [space] [space] [space] 2 [space] [space] [space] [space] [space] 4 [space] [space] [space] [space] [space]3.9
length [he] [space] [space] [space] 1 [space] [space] [space] [space] [space] 4 [space] [space] [space] [space] [space]3.5
length [sp] [space] [space] [space] 2 [space] [space] [space] [space] [space] 4 [space] [space] [space] [space] [space]3.4 

[hrothgar]

What could a self learning FD do?

 

It could create a database entry for every bid stating which partner and how many HCP, ZAR-Points and what kind of distribution he was holding.

 

It could than disclose a message like that:

 

player XYZ and ABC had the bidding sequence

1[cl]

5 times before.
[space] [space] [space] [space] [space] [space] [space] [space] [space] [space] min [space] [space] [space]max [space] [space] [space] [space]avg
HCP [space] [space] [space] [space] [space] [space] [space] [space]12 [space] [space] [space] [space] 17 [space] [space] [space] [space]12.5
length [cl] [space] [space] [space] 3 [space] [space] [space] [space] [space] 5 [space] [space] [space] [space] [space]4.2 
length [di] [space] [space] [space] 2 [space] [space] [space] [space] [space] 4 [space] [space] [space] [space] [space]3.9
length [he] [space] [space] [space] 1 [space] [space] [space] [space] [space] 4 [space] [space] [space] [space] [space]3.5
length [sp] [space] [space] [space] 2 [space] [space] [space] [space] [space] 4 [space] [space] [space] [space] [space]3.4 

No need to reinvent the wheel...

 

Stephen Pickett's Bridge Browser application can already do this...

If folks really want this type of fuctionality, it should be posisble to provide it.

 

Some enhancements are obviously necessary. You really need the ability to tag bids with an identifier explaining that this 1♦ opening is MOSCITO mark 8 rather than Precision version 3A. However, we alreayd have the basic tools necessary