[MrAce]

I recently saw a msg in BBO that says something like " Upgrated GIB available for older version now" And the next login my update was downloaded.

Yayy, i was excited that the GIB wont do retarded things like it does. I even asked to UDAY if they will make it available so i can set them as teammates or pd in teammatches. And rented it for 30 days this time.



I sit South, rest is GIB, check out the lead from "Upgrated GIB" !

---

[barmar]

This is an old problem, nothing to do with the upgrade. The upgrade is mainly to the bidding DB, its play and defense logic hasn't been touched significantly.

When GIB is doing its simulations, it doesn't see many hands where its top tricks will go away, so it thinks it's safe to lead something else.

It could have been worse. We've seen it double grands and not lead its ace.

[cloa513]

barmar, on 2012-March-16, 02:29, said:

This is an old problem, nothing to do with the upgrade. The upgrade is mainly to the bidding DB, its play and defense logic hasn't been touched significantly.

When GIB is doing its simulations, it doesn't see many hands where its top tricks will go away, so it thinks it's safe to lead something else.

It could have been worse. We've seen it double grands and not lead its ace.

Absolutely everything to do with the "upgrate". Nothing important ever fixed.

[MrAce]

barmar, on 2012-March-16, 02:29, said:

This is an old problem, nothing to do with the upgrade. The upgrade is mainly to the bidding DB, its play and defense logic hasn't been touched significantly.

When GIB is doing its simulations, it doesn't see many hands where its top tricks will go away, so it thinks it's safe to lead something else.

It could have been worse. We've seen it double grands and not lead its ace.

So u are telling us this is an old problem and known problem...wow.

I see the pinned posts that says Gib version 18 released, 19 released, 20 released, 21 released....and this problem is old and known? ...One would think it is a major malfunction for a bot which is incapable of defeating 6 NT with AK in its hand and it is the one on lead and needs to be fixed as prior to what kind of stayman or balancing NT ranges or whatever it uses in bidding that has been changed.

[ahydra]

Right, the coding of GiB's play needs to be revamped to make 100% plays before considering anything else. This should prevent it throwing kings under aces for a start, or solving declarer's guesses for him by playing Q from Qxx when declarer leads towards KJ in dummy.

ahydra

[barmar]

MrAce, on 2012-March-16, 06:31, said:

So u are telling us this is an old problem and known problem...wow.

I see the pinned posts that says Gib version 18 released, 19 released, 20 released, 21 released....and this problem is old and known?

We agonize over this on a daily basis. We're fixing things that we know how to fix, and that's mainly the bidding rules. We don't understand the play engine well, and we're not sure how to fix it, so these bugs persist. It's about 95% the same code we got from Ginsberg years ago (and most of those changes are not to the bridge logic, but just to allow the BBO servers to interact with it).

[Scarabin]

barmar, on 2012-March-16, 12:40, said:

We agonize over this on a daily basis. We're fixing things that we know how to fix, and that's mainly the bidding rules. We don't understand the play engine well, and we're not sure how to fix it, so these bugs persist. It's about 95% the same code we got from Ginsberg years ago (and most of those changes are not to the bridge logic, but just to allow the BBO servers to interact with it).

There seems to be a flood of complaints about GIB setting up the setting tricks but not cashing them. Fixing this may require serious reprogramming of GIB's play engine but don't you think the time has come when someone has to bite the bullet and make a start on this?

After all, superimposing a pragmatic interface in which GIB recognises the need to cash-out, examines what could go wrong and manages the trump suit, would eliminate most of these problems. Of course there is the butterfly in Brazil syndrome: a change at one end of a program may cause apparently totally unrelated crashes.

I fear if nothing is done GIB will fall behind other robots and,eventually, fall into disuse. Maybe its value as a partner/tutor who never complains will ensure its continued use.

[pigpenz]

Scarabin, on 2012-March-16, 18:14, said:

There seems to be a flood of complaints about GIB setting up the setting tricks but not cashing them. Fixing this may require serious reprogramming of GIB's play engine but don't you think the time has come when someone has to bite the bullet and make a start on this?

After all, superimposing a pragmatic interface in which GIB recognises the need to cash-out, examines what could go wrong and manages the trump suit, would eliminate most of these problems. Of course there is the butterfly in Brazil syndrome: a change at one end of a program may cause apparently totally unrelated crashes.

I fear if nothing is done GIB will fall behind other robots and,eventually, fall into disuse. Maybe its value as a partner/tutor who never complains will ensure its continued use.

actually I enjoy playing GIB, can live with some of the idiocy cause it happens in real life as well....but I dont know if the programming is really related to the "theory of Chaos"? You might have to ask Uday that, I believe he knows the person who wrote the book Jim Gleick.

[johnu]

barmar, on 2012-March-16, 12:40, said:

We agonize over this on a daily basis. We're fixing things that we know how to fix, and that's mainly the bidding rules. We don't understand the play engine well, and we're not sure how to fix it, so these bugs persist. It's about 95% the same code we got from Ginsberg years ago (and most of those changes are not to the bridge logic, but just to allow the BBO servers to interact with it).

As a software developer who has to make patches to some very old spaghetti code, I feel your pain . You never know when an innocent looking change may have fatal results in an entirely different section of the code. Each patch seems to take longer to figure out as the code gets more and more tangled, and at some point, you really need to scrap the old program and come up with something better designed and easier to maintain.

It sounds like you may be getting close to that tipping point. I read that GIB was one of the early world computer bridge champs but that was a long time ago and I think most of the leading programs today have undergone a number of rewrites in recent years. How does GIB play in comparison to the leading prgrams in 2012? Has anyone run any comparisons recently? I remember reading other forum members suggest that GIB be replaced by another bridge playing engine. Any news on that front?

[barmar]

Whenever I think about putting in some overriding rules like "Don't lead singleton K of trumps", "Second hand low", or "Cash the setting tricks ASAP", I worry about the unintended consequences. There are times when nonuntuitive plays like these are actually necessary -- Rodwell devotes an entire chapter of his book to when to play second hand high. So could I mess things up if it's discovered a good play, and my rules prevent it from going through with it?

On the other hand, it's probably the case that these clever plays are relatively rare, while its silly logic causes it to screw up routine hands. So it would probably get rid of 10 bad plays for every 1 good play that's lost, for a singificant net gain.

But I still have to figure out where in the code such an override can be put.

[Phil]

One thing I see over and over in GIB is that it does not appear that its decision tree has any structure in many circumstances.

When I am teaching bidding to a novice, it starts with a concept of a hand can have wide range, but with each single call the number of hands gets reduced within the set.

For instance, (and even though I've played thousands of hands against GIBs, I cannot give you a specific case because I have better things to do than report GIB bugs), when the auction starts 1♠ - (pass) - pass....

GIB should know that for every subsequent call that the passer cannot have more than 6 points. Yet, I believe I have seen it where the passer will do something else, and this initial constraint is missed.

[cloa513]

barmar, on 2012-March-17, 10:28, said:

Whenever I think about putting in some overriding rules like "Don't lead singleton K of trumps", "Second hand low", or "Cash the setting tricks ASAP", I worry about the unintended consequences. There are times when nonuntuitive plays like these are actually necessary -- Rodwell devotes an entire chapter of his book to when to play second hand high. So could I mess things up if it's discovered a good play, and my rules prevent it from going through with it?

On the other hand, it's probably the case that these clever plays are relatively rare, while its silly logic causes it to screw up routine hands. So it would probably get rid of 10 bad plays for every 1 good play that's lost, for a singificant net gain.

But I still have to figure out where in the code such an override can be put.

For the 6NT example you don't need special rules, first ensure the bidding is correctly entered into GIB's simulation inputs ( I bet its not given 33 HCP in opponents hands and length in two side suits what other possible play could there be than cash your top two winners), then have GIB run more simulations say 100 (maybe some rule of thumb whether its enough for the given hand) before making the lead ensuring the lead still follows its leads database afterall that's just like humans who think a bit more with 3 hands hidden, if his partner is a GIB then it can run a few more simulations than normal with correct information gained from at least the lead and dummy. All that would reduce the rediculous stuff-ups without changing anything fundamental about GIB.

[Scarabin]

The nice thing about an override is it can be removed if necessary and it should not cause bugs in the existing code.

That said, perhaps it would be possible to offer a choice: GIB with or without override? I know which I would choose.

Good luck Barmar if you decide to give it a go.