A Gib bug

[lycier]

[hv=http://www.bridgebase.com/tools/handviewer.html?sn=lycier&s=SKT9653HTDAJTCA85&wn=机器人&w=S4HKQJ9D864CJT932&nn=机器人&n=SJ7H8732DKQ97532C&en=机器人&e=SAQ82HA654DCKQ764&d=w&v=e&b=&a=P3D(Preempt%20--%207+%20%21D%3B%2010-%20HCP%3B%20%21DQ%3B%206+%20total%20points)D(Takeout%20double%20--%203-5%20%21C%3B%202-%20%21D%3B%203-4%20%21H%3B%203-4%20%21S%3B%2014+%20total%20points)5D(1+%20%21D)PPD(Takeout%20double%20--%203-5%20%21C%3B%202-%20%21D%3B%204%20%21H%3B%204%20%21S%3B%2017+%20total%20points)P6D(11-%20HCP%3B%209-12%20total%20points)PPP&p=DKS2DTD4D5C4DAD6DJD8D9C6HTH9H8HAH6S3HKH7HQH2H4S5HJH3H5S9S4S7SAS6C7CAC3SJSKC2DQS8D7SQC5C9D2CQC8CTD3CKSTCJ]499|350[/hv]

[Bbradley62]

Misclick... he meant 6♣.

[georgi]

Thanks for reporting.

 

Fixed in v31 not to use such high cuebids.

[1eyedjack]

Thanks for reporting.

 

Fixed in v31 not to use such high cuebids.

 

or presumably, when it does, not to pass them as partner? High level cue bids have their place, in rare circumstances (not that this is one of them)

[steve2005]

Yes, need to have a rule never pass cuebids!. My wife uses the play 4 hands for practice so I cant show hands. She's terrified to make a cuebid because Gib passes them!

[Antrax]

What would you have GIB bid when its partner makes a cue bid and it doesn't have a rule how to proceed?

[1eyedjack]

What would you have GIB bid when its partner makes a cue bid and it doesn't have a rule how to proceed?

 

I get that there might be a problem if a human makes a cue bid opposite a bot, with GIB having no instructions on the meaning of the human cue. But when the bot itself makes the cue, which must be based on algorithm, its bot partner is fully aware of cue bidder's hand types and there is no excuse for its having no idea how to proceed.

[Antrax]

To verify there are no gaps in GIB's rule database, you'd have to check every sequence of bids up to a point with every possible hand for responder. That's not realistically possible, so instead they have a default in there, if you can't figure out what to bid, pass. What alternative would you propose?

[1eyedjack]

Not being myself a programmer maybe I am genetically blind to the problem. My proposal would be, in the absence of a rule based approach, to run some sims to identify the likely most profitable spot based on partner's known range of hand types, and then bid it. I would be very surprised if such an exercise produced pass opposite a cue bid as ranking among the viable alternatives. I thought that this is what GIB did anyway, so I have always been a bit confused by repeated comments in the forums about its passing in doubt.

 

As a human, I bid almost entirely according to rules. I don't have the processing power to run sims in my own head although I concede that there are occasional judgemental decisions that involve my attempting to construct some pictures.. Likewise I don't have the mental capacity for a set of separate rules that cover all possible sequences. And yet, I know a cue bid when I see it, even if the sequence has never cropped up before in my life, and I know not to pass. The OP hand is a good example where in real life I would not have encountered the sequence and yet as East I would have bid 6H. Not a successful result, but not as bad as passing, and the 6D bid in this case is largely to blame for the final result. I have a block trying to understand why that principle cannot be imported into a computer.

 

Passing by default, in the absence of any better plan, may be an almost playable temporary fix until something better is found, when the contract currently rests with the opponents and partner still has another bid. It might also be workable, although with lower expectation of success rate, if done in the pass out seat.

 

I think that I would like to see some sort of a decision tree along the following lines:

 

Is it a sequence catered for in the database? If yes, apply database rules. If not:

 

Are we in a cue bid (or otherwise known to be in a fit of fewer than 7 cards)? If yes, exclude pass from the options, if no, include it (unless excluded later)

 

If not (in cue bid etc): Have we established GF values? If yes, are we in game? If no, then exclude pass from the options. If yes, include pass as option

 

Etc

 

Not complete, but I won't go further, but you get the drift.

 

Eventually you end up with a population of possible calls which may or may not include pass (and indeed some other bids), as viable alternatives from which to choose by sim

[1eyedjack]

Returning to Antrax's earlier question, I speculate that without relying on sims, if in response to a cue bid the "default" (ie what to do if you have no rule and cannot think what to do) is to bid the next bid up (where currently the default is pass) then your expected net long term profit will be increased (or loss reduced) contrasted with the current default treatment of pass, and that by a country mile.

 

I am not suggesting that this (bidding the next bid up) is the optimal default treatment, only that it is an obvious improvement on the current state. As such, I would not recommend the devotion of programmer resources to making such a wholesale change throughout the GIB system. If you are going to devote any resources to it, you might as well go the extra and make a half hearted attempt at installing something closer to optimal.

 

I simply seek to prove that a substantial improvement of sorts is quite trivial to devise.

[iandayre]

Did everyone read Georgi's response? His "fix" is "not use such high cuebids" ????? The fix should be that cue bidder's partner cannot pass such cuebids. With his "fix" if a human partner cuebids, GIB is still going to be passing. Definitely not the "fix" that was needed.

[1eyedjack]

Did everyone read Georgi's response? His "fix" is "not use such high cuebids" ????? The fix should be that cue bidder's partner cannot pass such cuebids. With his "fix" if a human partner cuebids, GIB is still going to be passing. Definitely not the "fix" that was needed.

 

Yes that was my point in post #4 (and the subsequent posts in thread). Doesn't hurt to reinforce it.

[georgi]

For all ( or until it's found flaw ) of the cases where GIB tends to make cuebids there are continuations.

 

When there is a flaw like this, it's probably that global cuebid is over expected continuations.

 

There might be hundreds of sequences and using cuebid is not always the same meaning. Cuebid could be short/stopper/asking/values/etc

 

I.e. here, replying on sequent TOX with cuebid on level 6 is not designed or rather expected. It's just flaw.

 

There can't be full descriptive sequences and full up to 7NT rules where everything has meaningful and spotless way.

 

But, the more common and frequent some auction is, GIB should be more equipped to provide accurate explanation and corresponding hand.