[wbartley]

Just Play Bridge robots. Human South.

---

[DozyDom]

Is that really a problem as problems go? It supported you when it mattered, and got you to the best contract or thereabouts. There are enough problematic problems to worry about with GIB before you go worrying about it showing extraneous shape.

[wbartley]

DozyDom, on 2018-June-07, 11:17, said:

Is that really a problem as problems go? It supported you when it mattered, and got you to the best contract or thereabouts. There are enough problematic problems to worry about with GIB before you go worrying about it showing extraneous shape.

It's bizarre and might be more important than it seems. GIB isn't a human. A mistake in a database or some code might manifest as a 3♣ bid in this situation and a jump to 7N in another. I'm not posting here to complain about or malign GIB. I'm trying to help by pointing out what I believe to be errors. The people maintaining it can ignore my comments at their leisure.

Maybe you would rather I made up another acronym for GIB? Isn't it funny how people who couldn't in a million years conceive of how to implement a bridge playing computer program mock so effortlessly and with so little imagination those who have. Ginsberg's Unintelligent Bridge player. I'm sure Matt Ginsberg is laughing his head off at that one.

[Stephen Tu]

Guess this one fell through the cracks since the usual system bid with South is 2c.

[DozyDom]

wbartley, on 2018-June-07, 14:34, said:

It's bizarre and might be more important than it seems. GIB isn't a human. A mistake in a database or some code might manifest as a 3♣ bid in this situation and a jump to 7N in another. I'm not posting here to complain about or malign GIB. I'm trying to help by pointing out what I believe to be errors. The people maintaining it can ignore my comments at their leisure.

Maybe you would rather I made up another acronym for GIB? Isn't it funny how people who couldn't in a million years conceive of how to implement a bridge playing computer program mock so effortlessly and with so little imagination those who have. Ginsberg's Unintelligent Bridge player. I'm sure Matt Ginsberg is laughing his head off at that one.

I... don't quite get what you're saying with that last part. I'm not the one mocking anyone here, nor do I think I'm showing a lack of imagination. Really, just where is this comment coming from?

[virgosrock]

Stephen Tu, on 2018-June-07, 18:51, said:

Guess this one fell through the cracks since the usual system bid with South is 2c.

Indeed Mr. Tu. nmf
The way the hand is posted I cant see the blurb for 2S. I have a feeling it is 4+H and 4+S.
GUBBO has already denied 4+ spades so now it is just bidding out it's shape.
The correct bid per system is 2C new minor forcing.

[wbartley]

DozyDom, on 2018-June-08, 04:20, said:

I... don't quite get what you're saying with that last part. I'm not the one mocking anyone here, nor do I think I'm showing a lack of imagination. Really, just where is this comment coming from?

I wasn't talking about you and it probably says something good about you that you didn't get it

[wbartley]

I'm pretty sure the explanation for 2♠ was 4+ spades; 5+ hearts, blah blah blah.

If it weren't, I wouldn't have posted this. But, it's possible I made an error. If 2♠ doesn't systemically show a strong hand with at least 45 in the majors and more hearts than spades, what does it show? What is the systemic bid with 5-6-2-0? 2♣?

[Stephen Tu]

2S shows a GF hand with reverse shape. I agree GIB should just support hearts with fit here and not bid minor randomly. I just mentioned that the system bid is 2c, because GIB doesn't bid 2S with this shape, which is why they probably went so long without fixing this, it won't show up in automated testing.

The reason 2S isn't typically bid here is that GIB is an up-the-line bidder, and can't have 4 cd spades. So as responder, you usually checkback with NMF for 3 cd heart fit, but there's really no great reason to reveal your 4 cd spades, since you won't have a fit there. You might as well conceal your shape from the opponents which in theory should make them lead less accurately and defend less accurately early in the hand if you end up in hearts. 2S showing spades is more utilized when playing styles where balanced 4 cd spades (4333, for some people 4324/4234 also) is required to rebid 1nt instead.

With 5-6 majors strong, maybe 3S should show that?

[johnu]

Stephen Tu, on 2018-June-08, 11:07, said:

I just mentioned that the system bid is 2c, because GIB doesn't bid 2S with this shape, which is why they probably went so long without fixing this, it won't show up in automated testing.

I think that logic is way too subtle for GIB programming from 25 years ago. And what automated testing? Are you saying BBO does automated regression testing of their release updates?

[Stephen Tu]

They have stated they run long matches between older and newer versions before release, to test for improvement, which ought to catch some percentage of possible regressions. I assume they must be automating at least this part of the testing as this would be infeasible to do otherwise.

[johnu]

Stephen Tu, on 2018-June-08, 11:58, said:

They have stated they run long matches between older and newer versions before release, to test for improvement, which ought to catch some percentage of possible regressions. I assume they must be automating at least this part of the testing as this would be infeasible to do otherwise.

IMHO, they would have to have a large set of hands that are run every release, and bids and plays that are different need to be logged and analyzed to see why there are changes. Most of these shouldn't be random hands, but hands specifically chosen to expose problems encountered in the past.

I think that if the same random number seeds are used, that the random carding algorithms and the simulations used for bidding and play should be the same, so differences should be due to coding changes (???).

[virgosrock]

johnu, on 2018-June-08, 13:12, said:

IMHO, they would have to have a large set of hands that are run every release, and bids and plays that are different need to be logged and analyzed to see why there are changes. Most of these shouldn't be random hands, but hands specifically chosen to expose problems encountered in the past.

I think that if the same random number seeds are used, that the random carding algorithms and the simulations used for bidding and play should be the same, so differences should be due to coding changes (???).

define "large" in "large set of hands". I can imagine the human labor involved in checking the results of the regression. If the human labor part is automated then the automation metrics come into question. I probably don't know that much about the regression process. easier to deal with integrated circuits