[dwar0123]

Advanced Robot tournament

Exported Hand

Explain trick 6, I challenge you.

[johnu]

You must be a member of the premium Platinum Club. Members are entitled to get 10% random plays from GIB opponents every month.

[pilowsky]

dwar0123, on 2020-September-17, 13:08, said:

Advanced Robot tournament

Exported Hand

Explain trick 6, I challenge you.

I've started a special thread for these hands: https://www.bridgeba...06#entry1007106

Did you know that sometimes Steve Jobs would man the help desk at Apple and some lucky customer would get a free replacement laptop when they called with a problem? Perhaps BBO has the same deal.

[sfi]

dwar0123, on 2020-September-17, 13:08, said:

Explain trick 6, I challenge you.

In the robots' defence, they have form. They made significant errors on tricks 1, 3, 4 and 5. But trick 2 was impeccable.

[johnu]

pilowsky, on 2020-September-17, 18:21, said:

Did you know that sometimes Steve Jobs would man the help desk at Apple and some lucky customer would get a free replacement laptop when they called with a problem? Perhaps BBO has the same deal.

Maybe not as good a deal as Jobs, but if you pony up another entry fee, BBO will let you play another tournament with the same GIBs as the previous tournament.

[sfi]

johnu, on 2020-September-17, 22:06, said:

Maybe not as good a deal as Jobs, but if you pony up another entry fee, BBO will let you play another tournament with the same GIBs as the previous tournament.

Second prize is two tournaments for the price.

[wbartley]

All these play bugs are 100% repeatable issues. A good programmer should be able to identify the problem.

[pilowsky]

wbartley, on 2020-September-27, 11:48, said:

All these play bugs are 100% repeatable issues. A good programmer should be able to identify the problem.

Out of interest, on your scale of 0-100, how good a programmer would they have to be? Would you be able to tell? I'm sure BBO would like to know so that they can work out which of their programmers are up to snuff - according to the 'Bartley' scale.

[wbartley]

Well, you have to know how to use a debugger, so I would say 30 out of 100. It may be in some cases that the programmers do know what's wrong but fixing it is too difficult. This seems like a cut and dried case of complete knowledge of the position being ignored in favor of God only knows what, but I would have to see the code and nobody's going to let me do that.

[mycroft]

So there are IIRC 75000 rules in the bidding database (so debugging is - easy to find, not easy to fix without invoking the Bug Hydra), and play is "double dummy simulate enough times to have a good idea of the best line, and take it." Unfortunately, that leads to three issues, that basically involve, not debugging, but building another bot:

• if there are several cards that lead to the same result, it doesn't matter which one is played;
  
• declarer is going to guess everything perfectly is a part of deciding 'there are several cards that lead to the same result'; and
  
• you are also going to play perfectly, after having seen partner's card.

Those people who think that the play should be another morass of rules can write one. Good luck. After you finish that, write a good definition of "convention" that everyone will agree covers everything.

[wbartley]

This looks like a bug to me and not the result of double dummy simulations with this play versus alternate plays giving the same result. Ducking the spade causes what appears on the surface to be a THREE TRICK differential in the defense's trick taking potential. It seems highly unlikely that this is the result of it "not mattering" which card is played.

I'm well aware that there can be extenuating circumstances that make a particular play seem wrong when it isn't, given the way GIB decides which card to play, but, did you look at the play in question? If I were a programmer working for BBO and I saw this, I would take a very long look at how that play was arrived at. Given that the point range and distribution of the South hand is pretty well bounded, it's hard to imagine that any random distribution of the unknown cards matching what is "known" about declarer's hand would produce the same double dummy result from ducking versus playing the ace.

[smerriman]

wbartley, on 2020-September-29, 16:30, said:

If I were a programmer working for BBO and I saw this, I would take a very long look at how that play was arrived at. Given that the point range and distribution of the South hand is pretty well bounded, it's hard to imagine that any random distribution of the unknown cards matching what is "known" about declarer's hand would produce the same double dummy result from ducking versus playing the ace.

If you want to go one step further on 'hard to imagine' - where you could always conjure up a reason that GIB miraculously simulated the only hands where its play made no difference - here's an example from 3.5 years ago where GIB made a literal 0% play, where any simulation of >= 1 hand would guarantee a better line.

The programmers just don't seem to be capable of looking at or adjusting the play code.

[johnu]

smerriman, on 2020-September-29, 18:47, said:

The programmers just don't seem to be capable of looking at or adjusting the play code.

That programmers don't have the knowledge and/or skill to make major changes in the play code has been acknowledged by BBO many times over the years. It's not a secret. My guess is that there is a bunch of spaghetti code that is condensed and complex that is tied in with lots of other spaghetti code that is more condensed and complex and that a change in one part of the code may adversely decisions in yet another part of the code. So rather than get into an endless cycle of patch, create errors, patch again, create more errors, continue looping, BBO has decided to live with the way things are currently programmed.