BBO Discussion Forums: Another nutty auction - BBO Discussion Forums

Jump to content

Page 1 of 1
  • You cannot start a new topic
  • You cannot reply to this topic

Another nutty auction

#1 User is offline   wbartley 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 289
  • Joined: 2010-July-23

Posted 2017-August-31, 22:15

Just play bridge robot.

GIB holds as North and deals:

Axxx
KJ
AKxxx
Kx

1=1=1=6 (explained correctly as 20+ pts with 4-4 in the majors)=Pass!!!!!

Last I checked 18+20=38.

The hand makes a grand slam in spades if the AK of diamonds are the deuce and trey.
0

#2 User is offline   The_Badger 

  • PipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 1,125
  • Joined: 2013-January-25
  • Gender:Male
  • Location:England
  • Interests:Bridge, Chess, Film, Literature, Herbal Medicine, Nutrition

Posted 2017-August-31, 23:48

No consolation for you but at least the Robot knows the Principle of Fast Arrival - not!
0

#3 User is offline   jdonn 

  • - - T98765432 AQT8
  • PipPipPipPipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 15,085
  • Joined: 2005-June-23
  • Gender:Male
  • Location:Las Vegas, NV

Posted 2017-September-01, 16:46

Its better to post a link to the entire hand, instead of just typing out one of the hands. But the likely reason for this is that GIB wasn't able to simulate any hands that meet the conditions. I remember instances from several years ago where someone would open 2NT and responder-GIB had 21 high card points, which is "impossible", thus it couldn't simulate hands and automatically passed. That particular problem may have been fixed (I'm not sure) but I think this is a similar phenomenon.
Please let me know about any questions or interest or bug reports about GIB.
0

#4 User is offline   wbartley 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 289
  • Joined: 2010-July-23

Posted 2017-September-03, 09:42

View Postjdonn, on 2017-September-01, 16:46, said:

Its better to post a link to the entire hand, instead of just typing out one of the hands. But the likely reason for this is that GIB wasn't able to simulate any hands that meet the conditions. I remember instances from several years ago where someone would open 2NT and responder-GIB had 21 high card points, which is "impossible", thus it couldn't simulate hands and automatically passed. That particular problem may have been fixed (I'm not sure) but I think this is a similar phenomenon.


I held

KQJx
AQxxx
-
AQxx

But you don't need the other hand to analyse the situation. All you need is to create a layout where East and West have hands with which they would pass. If the cause is what you speculate, it should be easy to step through and prove it.

That is, assuming you think there's anything wrong at all. Perhaps it IS impossible for South to have a hand that justifies the auction.

It seems like, at the very least, the restrictions going into the hand generator would be 20+ "points", less than 5 clubs, at least 4 hearts and at least four spades. The hand can't be specifically 4540 (or maybe even 4531) and can't have 6+ solid hearts because an initial bid of 2H would have been correct. Unless there are some restrictions that I'm not aware of, the hand I held meets all these requirements.

I won't quarrel with an algorithm that attempts to end an auction sensibly when it determines that it's partner's bids are "impossible" although that isn't exactly what you said. What you said was that maybe it automatically passes when partner makes a bid inconsistent with its previous bids. That would be a very bad algorithm. But, depending on how the hand generator generates hands, it may be a good idea to control the number of attempts to find a matching hand using the sum of the point count of the two hands considering that last bid at face value. When that sum is high, it will usually take more attempts to generate a matching hand. Also, since GIB counts "points" and not HCP, it could very well be giving itself more than 20 points for its own hand here and if one of the criteria for impossibility is that the sum of its "points" and partner's "points" can't be more than 40, that would be an outright bug because there are more than 40 "points" in the deck.
0

#5 User is offline   iandayre 

  • PipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 1,114
  • Joined: 2013-December-23
  • Gender:Male

Posted 2017-September-03, 12:29

The only problem here is your failure to bid on to the grand slam yourself. You had shown nothing more than a minimum opening bid when the robot jumped to slam.
0

#6 User is online   smerriman 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 3,723
  • Joined: 2014-March-15
  • Gender:Male

Posted 2017-September-03, 16:29

View Postiandayre, on 2017-September-03, 12:29, said:

The only problem here is your failure to bid on to the grand slam yourself. You had shown nothing more than a minimum opening bid when the robot jumped to slam.

It was the human that jumped to slam.

If this was Just Play Bridge, it's using basic robots, which if I'm not mistaken, don't simulate at all on the bidding. So it's not that it passes because it can't simulate a matching hand, just that it passes because it doesn't have a rule that tells it what to do in that sequence.

Interestingly, if GIB bids with itself, it does reach the grand via 1 - 1 - 1 - 2 - 3nt - 7.

Though oddly, if for some reason I happen to bid 5nt showing 19-21 in that sequence instead of 7, it will sign off in 6nt described as 15-16 HCP, rather than 7nt described as 17+ HCP. I guess it can't count its own points.
0

#7 User is offline   jdonn 

  • - - T98765432 AQT8
  • PipPipPipPipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 15,085
  • Joined: 2005-June-23
  • Gender:Male
  • Location:Las Vegas, NV

Posted 2017-September-04, 10:50

It's not helpful to dispute what I say and tell us what you think we need. I am telling you the way to present the hand that gives the programmer the most ability to analyze exactly what happened and fix it, as well as maintain it for reference. Linking to the actual hand does that, as well as ensuring all the facts are correct.
Please let me know about any questions or interest or bug reports about GIB.
0

#8 User is offline   wbartley 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 289
  • Joined: 2010-July-23

Posted 2017-September-05, 10:21

View Postjdonn, on 2017-September-04, 10:50, said:

It's not helpful to dispute what I say and tell us what you think we need. I am telling you the way to present the hand that gives the programmer the most ability to analyze exactly what happened and fix it, as well as maintain it for reference. Linking to the actual hand does that, as well as ensuring all the facts are correct.


I'll try not to second guess your programmers in the future. You can look back at my previous posts. I try very hard not to post things that wouldn't be of interest to the programmers since I am myself a professional programmer and have been for the last 35 years.
0

#9 User is offline   Stephen Tu 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 4,070
  • Joined: 2003-May-14

Posted 2017-September-05, 11:49

I don't think it is reasonable to ask them to program robots to bid seven based on point count after human wild unscientific leap to 6. Most of the time this is based on wild distribution and can't really know which hcp actually function or not. In my human partnerships if I jump to 6 it's basically a commandment for Pete's sake don't guess when to bid 7, maybe we'll work out the scientific way later. Computer is way worse than human at constructing hands for me, so I really want it to not be speculating on 7, or correcting to 6nt for that matter. I don't mind forced auto pass in general in such sequences.

This hand holding hkj and sa human supposed raise seven, since how the hell can south bid missing those cards, but I don't see how computer supposed to be able to calculate that.

No good player just raises 1 to 6 like ever, why should a robot know what that means? Take it slow, in theory it is best if North takes captaincy, finds 3 keys, trump q, hq, then can bid 7nt with reasonable confidence.

Not that bot can do it now, it needs some fixes in 6 level asks after keycard, both in finding a second king(like south asking for hk after North shows ck), or in 3rd round control/q asking, if North asks and bypasses king ask or south shows sQ and denies side king with 5nt. This science stuff is more important than dealing with guessing after jumps to 6.
0

#10 User is offline   wbartley 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 289
  • Joined: 2010-July-23

Posted 2017-September-07, 10:15

View PostStephen Tu, on 2017-September-05, 11:49, said:

I don't think it is reasonable to ask them to program robots to bid seven based on point count after human wild unscientific leap to 6. Most of the time this is based on wild distribution and can't really know which hcp actually function or not. In my human partnerships if I jump to 6 it's basically a commandment for Pete's sake don't guess when to bid 7, maybe we'll work out the scientific way later. Computer is way worse than human at constructing hands for me, so I really want it to not be speculating on 7, or correcting to 6nt for that matter. I don't mind forced auto pass in general in such sequences.

This hand holding hkj and sa human supposed raise seven, since how the hell can south bid missing those cards, but I don't see how computer supposed to be able to calculate that.

No good player just raises 1 to 6 like ever, why should a robot know what that means? Take it slow, in theory it is best if North takes captaincy, finds 3 keys, trump q, hq, then can bid 7nt with reasonable confidence.

Not that bot can do it now, it needs some fixes in 6 level asks after keycard, both in finding a second king(like south asking for hk after North shows ck), or in 3rd round control/q asking, if North asks and bypasses king ask or south shows sQ and denies side king with 5nt. This science stuff is more important than dealing with guessing after jumps to 6.


Just because a bid doesn't have some artificial meaning doesn't mean it's unscientific. There's nothing unscientific about 1N=6N. If the bot assumes I'm playing its system, which it should, and not making some wild bid out of the blue then it should realize it has a WAY better hand than can reasonably be expected. If it were to bid 7S after the given auction and go down due to anything other than a freak lie of the cards, then the 6S bid was clearly irrational.
0

#11 User is offline   Stephen Tu 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 4,070
  • Joined: 2003-May-14

Posted 2017-September-07, 13:30

1n 6n is a systemic bid, 1n is tightly limited and responder is captain, passing is mandated, it's reasonable to bid this way with two balanced hands just on total hcp estimate.

1m 1h 1s 6s is totally unscientific and is in nobody's system. It grabs the definition out of some default rules assuming certain amount of points per level.

Now on the actual holding, a human can realize the hkj and spade ace are the nuts, and raise. (And still think you are a crazy nut because he could hold none of those cards, xxx hearts and you end up down 2). But then you are ascribing human like reasoning abilities to a bot. They don't work that way. How would you formalize it in a more simplistic way that says raise to 7 with this, but not some other 18 like xxxx kj akqjx kj?

The problem is "points" includes distributional points. You can't just say 20+18 = 38 so I bid grand. You have to be able to reason out working points vs duplication of values, which is very hard to teach the computer to calculate. It would be far more useful and easier and more frequently useful to give it better k asking and q asking science after rkc, than to get it to work out when to guess raise 7 after wild undefined leaps to 6.

Personally I'd quit playing with a human who thought 1s 6s is a normal way to bid.
0

#12 User is offline   wbartley 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 289
  • Joined: 2010-July-23

Posted 2017-September-11, 16:30

View PostStephen Tu, on 2017-September-07, 13:30, said:

1n 6n is a systemic bid, 1n is tightly limited and responder is captain, passing is mandated, it's reasonable to bid this way with two balanced hands just on total hcp estimate.

1m 1h 1s 6s is totally unscientific and is in nobody's system. It grabs the definition out of some default rules assuming certain amount of points per level.

Now on the actual holding, a human can realize the hkj and spade ace are the nuts, and raise. (And still think you are a crazy nut because he could hold none of those cards, xxx hearts and you end up down 2). But then you are ascribing human like reasoning abilities to a bot. They don't work that way. How would you formalize it in a more simplistic way that says raise to 7 with this, but not some other 18 like xxxx kj akqjx kj?

The problem is "points" includes distributional points. You can't just say 20+18 = 38 so I bid grand. You have to be able to reason out working points vs duplication of values, which is very hard to teach the computer to calculate. It would be far more useful and easier and more frequently useful to give it better k asking and q asking science after rkc, than to get it to work out when to guess raise 7 after wild undefined leaps to 6.

Personally I'd quit playing with a human who thought 1s 6s is a normal way to bid.


On the hand with which I bid 6S, if opener has neither the K of hearts nor the A of spades - about the worst hand it can hold - 6S is at worst on a finesse, not down 2, and that's only if the opponents lead a heart. If opener's hand is T9xx xx AKxxx Kx, 6S is well above 50-50, even on a heart lead.

There are hands that opener could hold that make seven on minimum values, Axxx KJ xxxxx Kx, for example, which may not be an opening bid for the bot. What sort of scientific approach do you suggest for me to find out about such hands, given that I'm playing with a machine that frequently passes cue bids?

Tell me a hand that the bot counts as 20 "points" that I can have that doesn't produce 7S opposite its hand? It was the bot that said the meaning of my 6S was 4+H, 4+S and 20+ "points". That's what I have. There's nothing inconsistent in my bidding prior to the 6S in the robot's system. I can't bid 2H initially because that shows either a self-sufficient suit or support for opener's suit and I have neither. I could bid fourth suit forcing over the 1S bid but what am I bidding over 2NT? Anything other than 6S at that point is just asking for trouble, so I'm really no better off than just bidding 6S. And look at the bot only auction below for how silly the "scientific" approach is. The bot bids 7S over responder's 3N. That could as easily be based on Axxx xxx AKQx Kx as what it held. Its jump to 7S is no more scientific than my jump to 6S. Less so, in fact, because it's the bot that knows the value of its KJ of hearts and K of clubs. I suppose my 6S could have been based on KQJxx AQxxxx Q Q but no sane point counting system would give me 8 points for those two stiff queens.

You say that it's not normal to bid 1S=6S. Well, it's true that opportunities to do so don't come up often, but, given the constraints of the system we're forced to play with the robots, 6S isn't completely nuts. No more nuts than 1D=1H=1S=2C=3N=7S is.
0

#13 User is offline   Stephen Tu 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 4,070
  • Joined: 2003-May-14

Posted 2017-September-11, 18:56

The robots won't pass fourth suit forcing followed by a spade raise. It's not necessary to be so paranoid that you think you will be dropped in a cue if you go slow that you should just jump to six. It's really rare that a bot passes a cue, actually, that tends to be more in rare weird competitive auctions where it gets confused what should be a cue or what isn't, somebody cues multiple times in a row, etc. And if it does pass such a sequence, that is just bug to be fixed.

The 3n 7s, more info has been exchanged, it is less silly IMO. Though I'm not fond of that auction either, thinking it is too strong for jumping to 3nt. After 3nt, options are limited since no good way to set spades as trump and be forcing in std methods.

All I'm saying is that generally I would prefer the computer not try to guess what I have for a jump to 6, and think it is much more productive for the programmers to work on scientific auctions to grand, and fix the lower sequences so humans aren't worried about being passed in a forcing bid or cue. Rather than work on guessing why a human skipped 5 levels of bidding that they would never do opposite a real human partner, and constructing some complicated rule to figure out what are working points and what is worth a level bump. Opener could easily hold xxx hearts. KJ are huge. Why wouldn't you want it to raise holding just those and not the spade ace? You can't have akqxx aqxxxx - a? It is just hard to formulate rules for what is enough in such auctions, and I'd rather not have programmers spend their limited time optimizing an auction that in my mind should never be perpetrated in a human partnership.
0

Page 1 of 1
  • You cannot start a new topic
  • You cannot reply to this topic

1 User(s) are reading this topic
0 members, 1 guests, 0 anonymous users