[cherdano]

The frustrating part from my perspective is that, even though quite a lot of work has gone into, there are still so many basic issues that we have all known for years, and that aren't getting fixed. I don't want to make too strong of a claim here - clearly (IMO) GIB has improved. But also this progress sometimes seems like tuning the windshield wipers when we do need a new engine.

Example 1: No, jumping to game (especially in competition) doesn't show the show the world.
http://tinyurl.com/yal7597g
(East's double is insane but it's probably caused by the ill definition of West's jump to 4♥. And I still encounter auctions where just competing to game is deemed to be worth 20+ total points, thus risking GIB partner to double any bid by the opposition now matter how weak he is.)

Example 2: No, if I open suit A and then bid suit B at the high-level in competition, that doesn't mean my suit B is much longer than my suit A.
http://tinyurl.com/ybuoqkp3
(North seems to play me for 4=6+ in the minors.)

Example 3: GIB can't handle double-and-bid.
http://tinyurl.com/y94ymkh2
(Four trumps and an ace is certainly worth a raise!)

Issue 4: Rebidding 2m after opening 1m doesn't just show a "biddable suit"
http://tinyurl.com/y7l5yfyp

Basically, I'd feel that posting specific auctions would just distract GIB programmers from trying to solve such issues more systematically. Does it really need 5+ years to identify all auctions where GIB thinks competing to game shows 20+ points? In a well set up architecture, it surely should be possible to systematically identify such auctions, and then so systematically fix them?
If not, wouldn't putting all the effort that is going into GIB into a bridge program with a better architecture result in *much* faster progress?

Quite some time ago I spend quite a lot of time on improving an open source go program (GNU Go), so I do know that these things always look easier from the outside. But stil...

[helene_t]

Yeah, I have the same feeling. If the problem is serious it is already known, if it isn't serious why post it, there are more serious things for the programmers to deal with.

I suppose I should stop posting here also.

[cherdano]

Btw, all these examples were from just two 55% rebate tourneys, i.e. 24 hands.

[hrothgar]

Agreed.

Can't help but believe that it would be much more useful to be working on a a standard interface to allow folks to use Jack or WinBridge or whatever they prefer...

[eagles123]

good post Cherdano, I agree. I understand there are freak hands that are hard to deal with, but most GIB mistakes just seem to be the same old errors day after day after day. after a while it seems futile reporting them.

[pescetom]

hrothgar, on 2018-August-20, 23:58, said:

Can't help but believe that it would be much more useful to be working on a a standard interface to allow folks to use Jack or WinBridge or whatever they prefer...

I think this is the crux.
There is no good reason why the owner/developer of the game platform and the owner/developer of a robot should coincide.
Especially if the in-house robot happens to be significantly inferior to the platform and to user's expectations.
Just think about how Microsoft lost credibility by doing everything possible to impose it's feeble IE browser instead of cutting it's losses and improving Windows instead.
BBO and competitors like Funbridge have a common interest in defining and making available such a standard interface rather than wasting energy on further in-house development.
Independent developers of Robots have even more to gain.
It's time they all got together to design that interface and decide how to govern it's development and deployment.

[Vampyr]

hrothgar, on 2018-August-20, 23:58, said:

Agreed.

Can't help but believe that it would be much more useful to be working on a a standard interface to allow folks to use Jack or WinBridge or whatever they prefer...

Yes. Otherwise someone else will build a good site around these robots and people will play there instead.

[zhasbeen]

cherdano, on 2018-August-20, 16:37, said:

The frustrating part from my perspective is that, even though quite a lot of work has gone into, there are still so many basic issues that we have all known for years, and that aren't getting fixed. I don't want to make too strong of a claim here - clearly (IMO) GIB has improved. But also this progress sometimes seems like tuning the windshield wipers when we do need a new engine.

Example 1: No, jumping to game (especially in competition) doesn't show the show the world.
http://tinyurl.com/yal7597g
(East's double is insane but it's probably caused by the ill definition of West's jump to 4♥. And I still encounter auctions where just competing to game is deemed to be worth 20+ total points, thus risking GIB partner to double any bid by the opposition now matter how weak he is.)

Example 2: No, if I open suit A and then bid suit B at the high-level in competition, that doesn't mean my suit B is much longer than my suit A.
http://tinyurl.com/ybuoqkp3
(North seems to play me for 4=6+ in the minors.)

Example 3: GIB can't handle double-and-bid.
http://tinyurl.com/y94ymkh2
(Four trumps and an ace is certainly worth a raise!)

Issue 4: Rebidding 2m after opening 1m doesn't just show a "biddable suit"
http://tinyurl.com/y7l5yfyp

Basically, I'd feel that posting specific auctions would just distract GIB programmers from trying to solve such issues more systematically. Does it really need 5+ years to identify all auctions where GIB thinks competing to game shows 20+ points? In a well set up architecture, it surely should be possible to systematically identify such auctions, and then so systematically fix them?
If not, wouldn't putting all the effort that is going into GIB into a bridge program with a better architecture result in *much* faster progress?

Quite some time ago I spend quite a lot of time on improving an open source go program (GNU Go), so I do know that these things always look easier from the outside. But stil...

Now you are talking.

“But also this progress sometimes seems like tuning the windshield wipers when we do need a new engine. “

Yes!!

[zhasbeen]

Vampyr, on 2018-August-21, 09:09, said:

Yes. Otherwise someone else will build a good site around these robots and people will play there instead.

"Can't help but believe that it would be much more useful to be working on a a standard interface to allow folks to use Jack or WinBridge or whatever they prefer.."

Seems like there was talk about something like that last year but it wasn't feasible.

Whatever the case, something needs to be done.

[helene_t]

Vampyr, on 2018-August-21, 09:09, said:

Yes. Otherwise someone else will build a good site around these robots and people will play there instead.

There's already at least two decent sites build around Jack. BBO is more than robots, though. I don't see any of them becoming serious competitors to BBO any time soon.

[Vampyr]

helene_t, on 2018-August-21, 15:18, said:

There's already at least two decent sites build around Jack. BBO is more than robots, though. I don't see any of them becoming serious competitors to BBO any time soon.

Well, you could ply with Jack when you want to lay. Withrobots, and BBO when you wNt to play with people.

[helene_t]

Vampyr, on 2018-August-21, 18:11, said:

Well, you could ply with Jack when you want to lay. Withrobots, and BBO when you wNt to play with people.

I speak Dutch and have many friends in the Dutch bridge society, so it would be convenient for me to subscribe to one of the Dutch Jack sites.

But I just play against robots to kill time and then I like the fact that my some of my bbo friends are logged in so I can chat with them and set up a game against robots with a human partner.

If I seriously wanted to use robots to practice my bridge skills, I would use Jack offline.

Anyway, back to the topic. Jack and WB5 need their opps to play one of their systems so that they can make inference from the bidding. It would take a lot work to make them work in an environment where different robots play at the same table.

Maybe better just for BBO to license Jack and use it instead of GIB. But I think Jack uses more CPU time than GIB does.