Gib version disclosure
#1
Posted 2017-February-06, 07:16
Does Gib have a version number? Is this public information? Where can it be seen?
#2
Posted 2017-February-06, 14:47
The original author's name is Matthew Ginsberg. He sold the program to BBO quite some years ago. His focus in creating the program was on card play. Unfortunately he had intermediate (at best) knowledge of bidding. Thus the myriad poor descriptions of bids in GIB. BBO slowly makes improvements but there is still an immense distance to go, and little financial incentive for BBO to speed the process.
Issue reporting mechanisms are: 1) This Forum and 2) While playing in a tournament you can click on "Send Robot Report" to tell BBO about GIB issues.
#3
Posted 2017-February-06, 15:19
I think I was asking for a wrong thing. It seemed to me, that GIB improved its play recently (1-2 days ago). I assumed it had to be GIB new version. But I guess the assumption was wrong.
Is it possible that GIB bidding system changed, while GIB is still the same version?
Is the bidding system of GIB also versioned?
#4
Posted 2017-February-06, 15:21
iandayre, on 2017-February-06, 14:47, said:
This is an invalid conclusion. GIB bidding poorly in certain situations has extremely little to do with the programmer's knowledge of bidding. You could have the best bidder in the world do the programming , it would still be extremely difficult to transfer that knowledge into a dumb computer. It's not like there is a Vulcan mind meld going on, and the computer brain absorbs the programmer's bidding knowledge through osmosis, and bids as well as the programmer does. Imagine instead asking the best bidder in the world, to write down a set of say 25000 rules to calculate the meaning of all bids in all possible sequences, and to calculate the best bid with a given hand after any sequence, *to be successfully followed by someone of IQ 80 who has never played bridge for a day in his life and will never figure out general principles, just follow the rule flowchart to the letter*.
That is the difficulty. Not the original programmers lack of bridge knowledge (Rod Ludwig of Meadowlark bridge did most of the bidding engine originally).
#5
Posted 2017-February-06, 15:52
Anyway, presumably the demise of the windows client will cut down on the multitude of versions that until last month you could encounter at the table.
That said, given that a host could place either a basic or advanced robot at your table depending on the depth of his pocket, it would still be nice to know which, I guess. Also in the brief period following a Gib update it would be nice to know when playing in an instant tourney whether the 3 robots at your table are the same as other tables.
Psyche (pron. sahy-kee): The human soul, spirit or mind (derived, personification thereof, beloved of Eros, Greek myth).
Masterminding (pron. mstr-mnding) tr. v. - Any bid made by bridge player with which partner disagrees.
"Gentlemen, when the barrage lifts." 9th battalion, King's own Yorkshire light infantry,
2000 years earlier: "morituri te salutant"
"I will be with you, whatever". Blair to Bush, precursor to invasion of Iraq
#6
Posted 2017-February-06, 19:30
Stephen Tu, on 2017-February-06, 15:21, said:
That is the difficulty. Not the original programmers lack of bridge knowledge (Rod Ludwig of Meadowlark bridge did most of the bidding engine originally).
Stephen, I would like to understand this better. I completely understand that there must be great difficulty in programming GIB to bid accurately. Isn't that a different issue? Let's take a simple example: the many situations where the bidding description calls for 25+ or something close for a particular action. There are many, many others. Doesn't the description come first, then the programmer tries to get GIB to bid according to that description? If the descriptions are not in accordance with sound bridge logic, GIB's bidding will be poor even if programmed perfectly, no?
As for Jack's comment regarding Ginsberg's bridge pedigree, I was going by Fred's message to me, now I think about 15 months ago. I am sure you read it. I know nothing of him other than from that.
#7
Posted 2017-February-06, 21:36
iandayre, on 2017-February-06, 19:30, said:
You might reread and try to understand what I wrote in your earlier thread:
http://www.bridgebas...post__p__874159
and ask questions about what you still don't understand.
I think the main gist of it is that you still don't really understand how the bidding descriptions are generated. You seem to think that someone went through a list of all possible bridge auctions and filled them in with what the bids should mean. And that that someone was a lousy bridge player because sometimes the descriptions spit out for a bid are lousy, that the programmer actually looked at the particular auction you are complaining about and specifically required 25+ pts for this bid. This is not how it works, because it is completely infeasible to do it this way given the number of possible bridge auctions (you'd be waiting for like longer than the age of the universe for the guy to finish the list). Instead the computer is given a (still ginormous, but finite and tractable, not practically infinite) list of rules defining what bids ought to mean, and the auction triggers some subset of those rules in some order, that tell it what a bid means, and what bids it ought to choose from now. The rules also have attached info on how to spit out a description for human consumption if it selects that bid. It's not "human writes description for all the auctions", then "programmer makes computer bid to match those descriptions". The descriptions are being generated by the program, by the ruleset.
The 25+ comes from some very basic rules that tell it that generally you want 25+ to bid at game level. It falls back on some of these basic rules when it has it doesn't have rules that tell it what the auction means, it needs to default to something. It needs more rules added governing competitive bids over preempts and opp game bids where you have to bid a lot more games based on two-way shot (you make or good sac vs. them) and assumption of partner having some 7 count on average. Right now some auctions are covered but many still fall through the cracks.
Generally it's difficulty coming up with a complete enough rule set to cover the insane multitude of possible auctions in a sensible manner.
It didn't arise from "programmer really thought you need 25+ to bid game in competition and you are supposed to assume partner yarb all the time, and purposely programmed GIB to bid this way".
#8
Posted 2017-February-06, 21:44
- It would be simpler to improve Gib's bidding and its bidding-system.
- Gib could provide you with bidding practice for any system, specified in an FD file.
- It would also level the playing-field for Gib, when its opponents employ a different bidding-system.
- FD files might not cover every eventuality. To cope with such lacunae, Gib would still need to fall back on its heuristics.
IMO, Kungsgeten's brilliant bidding-system language would be even better because it allows more generic bidding specifications. He has generously put the source on Github.
Kungsgeten's BML
I wish BBO would adopt it.
#9
Posted 2017-February-07, 14:34
Stephen Tu, on 2017-February-06, 21:36, said:
http://www.bridgebas...post__p__874159
and ask questions about what you still don't understand.
I think the main gist of it is that you still don't really understand how the bidding descriptions are generated. You seem to think that someone went through a list of all possible bridge auctions and filled them in with what the bids should mean. And that that someone was a lousy bridge player because sometimes the descriptions spit out for a bid are lousy, that the programmer actually looked at the particular auction you are complaining about and specifically required 25+ pts for this bid. This is not how it works, because it is completely infeasible to do it this way given the number of possible bridge auctions (you'd be waiting for like longer than the age of the universe for the guy to finish the list). Instead the computer is given a (still ginormous, but finite and tractable, not practically infinite) list of rules defining what bids ought to mean, and the auction triggers some subset of those rules in some order, that tell it what a bid means, and what bids it ought to choose from now. The rules also have attached info on how to spit out a description for human consumption if it selects that bid. It's not "human writes description for all the auctions", then "programmer makes computer bid to match those descriptions". The descriptions are being generated by the program, by the ruleset.
The 25+ comes from some very basic rules that tell it that generally you want 25+ to bid at game level. It falls back on some of these basic rules when it has it doesn't have rules that tell it what the auction means, it needs to default to something. It needs more rules added governing competitive bids over preempts and opp game bids where you have to bid a lot more games based on two-way shot (you make or good sac vs. them) and assumption of partner having some 7 count on average. Right now some auctions are covered but many still fall through the cracks.
Generally it's difficulty coming up with a complete enough rule set to cover the insane multitude of possible auctions in a sensible manner.
It didn't arise from "programmer really thought you need 25+ to bid game in competition and you are supposed to assume partner yarb all the time, and purposely programmed GIB to bid this way".
OK, the descriptions don't come first. Basically you are saying that the programmers do the best they can, and when they are done and can see the descriptions of various calls, they see how close they came. About right? Let me ask this. GIB does pretty well at No Trump bidding. It consistently opens with 15-17 HCP and relatively balanced distributions. Let's say they wanted to change this to adopt a 12-14 NT. By no means am I suggesting this should happen. But in general terms, how would this be approached?
#10
Posted 2017-February-07, 16:00
iandayre, on 2017-February-07, 14:34, said:
They come up with some thousands of rules to cover as many bidding sequences as they can conceive of. Rules to open the bidding. Rules to respond. Rules to double/overcall. Rules to make competitive bids after the overcall. Rules for the second round. And so forth. But inevitably some bidding sequences, especially competitive sequences, fall into some not well defined areas, or not well debugged areas. So sometimes the bid the computer comes up with has a reasonable meaning, sometimes it doesn't. When it doesn't, someone reports the bug, and they have to figure out whether to tweak existing rule(s), or add additional rules to handle the sequence as a special case.
Quote
GIB in PC version already handled different NT ranges long time ago. It supports weak NT, could sort of play K/S, Acol, strong club to some extent. But long ago it was decided to concentrate on "std w/ gadgets", basically 2/1 base, to be the focus, since it's hard enough to get one system solid, let alone 3 or 4.
Different Nt ranges is pretty easy, for auctions starting with 1nt. The rules probably have something that says like opening NT range = X to Y. So deciding to open is easy. And responding should be fairly easy too, you just decide whether hand is partial/game invite/game force/slam invite/slam force, and the ranges for the various continuations are just adjusted to account for stronger/weaker opener.
The much harder part is dealing with the auctions that start with 1x, rather than 1nt, especially in competition. Now a lot of things would have to be tweaked. Like 1x-(1y)-p-p-1nt is 18-19 easy in standard. But playing weak NT you probably want to balance there with a lot of 17s. I suppose 16s can pass. Plus you have to figure out how to handle strong NT hands after 4th hand interference. And you want to figure out how many M you want to bid with 1m-1M-2M/3M on say 16 balanced, whether you want K-S style with strong single raises or not.
#11
Posted 2017-February-08, 08:51
Stephen Tu, on 2017-February-06, 21:36, said:
Generally it's difficulty coming up with a complete enough rule set to cover the insane multitude of possible auctions in a sensible manner.
It didn't arise from "programmer really thought you need 25+ to bid game in competition and you are supposed to assume partner yarb all the time, and purposely programmed GIB to bid this way".
It was lack of testing and probably a lack of planning in the programming that allows these 25+ descriptions to show up all over the place. When the 1st example of a 25+ point description came up, the initial programmer should have realized that this was a priority mistake to fix and gone through the programming to prevent this sort of mistake from happening. This may have included a design change in the bidding code, but fixing auctions one by one was not a viable choice.