[Stephen Tu]

The point is that GIB has no preprogrammed knowledge of suit combinations. It is essentially figuring out how to play suits on the fly, every card. If it is given enough time, it will deal out enough hands to see that playing the K first is better on the original combo. Not given enough time, it can do some silly things.

The basic gist of the algorithm is:
1. deal bunch of hands that fit the auction, constrained by amount of time it has to analyze deals.
2. on those hands, calculate the double dummy score for playing each card.
3. assign score for each card weighted depending on how it does over all the hands.
4. pick randomly among cards that tie for highest score.

So the discussion is kind of pointless. GIB plays how it plays. If you give it more time it will play better, to a certain extent. Yes, it screwed up on this hand. SO WHAT??? There is no simple "bug fix" the programmers can do. It's not like it has a table of suit combos and the entry for this combo is wrong. That's not how GIB thinks. To alter how it thinks is a very difficult problem, other developers have been working on their programs over *decades*.

And by the way on your last combo aqxxx vs. xxx it matters if you have the 9xx or not. having the 9, if the ace drops jack or ten, you should lead low but then DUCK if the next person plays low.

[dkham]

Interesting to hear how GIB works. I like the simplicity of its algorithm.

Stephen was saying it can't easily be improved, but perhaps in the final stage:

4. pick randomly among cards that tie for highest score.

it could try and guess better, by using some logic. It wouldn't matter if this logic wasn't great, as after all it this is only a tie-break among cards which are supposed equally good.

For the OP I presume at trick two all of the trumps tied for the highest score (and possibly a club too), so a rule of thumb such as "play honour from shortest hand first" might have worked here to help GIB identify the ♦K as a better card among "equals".

Note this isn't a practical suggestion that I'm proposing anyone actually do, just an observation. I'm a computer programmer too and know how messy these things get!

[pigpenz]

maybe then since GIB is so hopeless
we should just close this whole forum on GIB

seems most of the times we notice the idiosyncacies of GIB
but inreality there isnt much that can be done about it

[Stephen Tu]

pigpenz, on 2012-May-04, 09:13, said:

seems most of the times we notice the idiosyncacies of GIB
but inreality there isnt much that can be done about it

A lot can be done about the *bidding* idiosyncrasies. This is much more easily alterable and within the skill set of the BBO developers.

The *play and defense*, not so much.