Why does Gib-defender always play highest of "equals"

[Stefan_O]

Why does Gib as defender so often play the highest of "equal" (according to its double-dummy analysis) cards?

 

Here one gross example from money-bridge of what disasters sometimes come out of this ill-designed approach.

 

I can only recommend Gib be re-programmed to always play the lowest --- at least as long as they are not touching cards -- because it evidently has no way to judge if a higher card may be harmful or not.

 

 

[hv=http://www.bridgebase.com/tools/handviewer.html?lin=pn%7CHumanVictim%2Cstefan_o%2CExpertRobot%2CDummyRobot%7Cst%7C%7Cmd%7C4S347HJD457TC2567T%2CS9TKAH3QD9JKC489A%2CS28JH257TD368C3JQ%2C%7Crh%7C%7Cah%7CBoard+2%7Csv%7Cn%7Cmb%7C1H%7Can%7CMajor+suit+opening+--+5%2B+%21H%3B+11-21+HCP%3B+12-22+total+points+%7Cmb%7Cp%7Cmb%7C1S%7Can%7COne+over+one+--+4%2B+%21S%3B+6%2B+total+points%7Cmb%7Cp%7Cmb%7C3H%7Can%7COpener+rebid+--+6%2B+%21H%3B+3-+%21S%3B+17-20+total+points+%7Cmb%7Cp%7Cmb%7C4N%7Can%7CBlackwood+%28H%29+--+2%2B+%21H%3B+4%2B+%21S%3B+15%2B+total%7Cmb%7Cp%7Cmb%7C5C%7Can%7CZero+or+three+key+cards+--+6%2B+%21H%3B+3-+%21S%3B+17-20+total+points+%7Cmb%7Cp%7Cmb%7C5N%7Can%7CQuery+kings.+Have+all+keycards.+Does+not%7Cmb%7Cp%7Cmb%7C6C%7Can%7CKing+of+C+--+6%2B+%21H%3B+3-+%21S%3B+%21CK%3B+17-20+total+points+%7Cmb%7Cp%7Cmb%7C7N%7Can%7C2%2B+%21H%3B+4%2B+%21S%3B+21%2B+HCP%3B+%21HQ%7Cmb%7Cp%7Cmb%7Cp%7Cmb%7Cp%7Cpg%7C%7Cpc%7CD3%7Cpc%7CDQ%7Cpc%7CD4%7Cpc%7CD9%7Cpg%7C%7Cpc%7CH4%7Cpc%7CHJ%7Cpc%7CHQ%7Cpc%7CH7%7Cpg%7C%7Cpc%7CH3%7Cpc%7CHT%7Cpc%7CHA%7Cpc%7CS3%7Cpg%7C%7Cmc%7C13%7C%0A%0A]600|450[/hv]

[johnu]

I've complained about GIB randomly sacrificing high cards which could win a trick later for years and also suggested that GIB play the lowest (or random card from equals) card in these situations.

 

The "reason" is that double dummy, the ♥10 will never take a trick so it can't cost to play it. Obviously bridge isn't played double dummy by most humans or GIBs even if they have a complete count of the hand.

[smerriman]

Happens all the time sadly :(

 

A similar idea applies to GIB declarers who have a 100% line, and a line which works if the bidding can be 100% trusted, and it takes the latter and goes down, rather than assigning tiny probabilities to exceptions and breaking ties that way.

 

These types of things have been mentioned so many times, it's clear that BBO aren't capable of fixing them.

[johnu]

A similar idea applies to GIB declarers who have a 100% line, and a line which works if the bidding can be 100% trusted, and it takes the latter and goes down, rather than assigning tiny probabilities to exceptions and breaking ties that way.

 

Another problem is that GIB doesn't do enough sampling (or the right kind of sampling). Take a 4-0 trump break missing J10xx that you can pick up in 1 direction. That's ~10% for a 4-0 break, but 10%/2 = 5% (1 in 20 chance) for a 4-0 break in the right direction. It's very likely that a small sampling won't test an example where there is that 4-0 break, so GIB won't play to pick that up. Ideally, GIB should use suit distribution probabilities to do modify the random sampling and do some weighting on the results to reflect the odds. I'm not holding my breath on that one since that seems like a major change in the play engine.

[steve2005]

I don't think Gib is always playing high it is playing random cards when double it doesn't matter.

If you check other people playing the hand the same thing may be happening but it is because the random seed is the same.

 

Gib plays random cards to confuse you. Since supposedly it doesn't matter which card is played i don't see how this will confuse you.

[johnu]

I don't think Gib is always playing high it is playing random cards when double it doesn't matter.

If you check other people playing the hand the same thing may be happening but it is because the random seed is the same.

 

Gib plays random cards to confuse you. Since supposedly it doesn't matter which card is played i don't see how this will confuse you.

 

I'm sure declarer was very confused by the random ♥10 but apparently not confused enough to duck the trick :P If you know GIB will play random spot cards, why would you ever be confused? The problem isn't that declarer is confused but that GIB defender has eliminated a potential guess into a 100% play.

[steve2005]

The problem isn't that declarer is confused but that GIB defender has eliminated a potential guess into a 100% play.

What's worse is Gib is depending on opponents bidding and sampling. Gib can never be sure that what card it play doesn't matter even against the best players in the world.. But that is how it is programmed.

[Stefan_O]

What's worse is Gib is depending on opponents bidding and sampling. Gib can never be sure that what card it play doesn't matter even against the best players in the world.. But that is how it is programmed.

 

I don't get what you mean here, sorry...

 

"depending on opponents bidding" --

All decent declarers, of course, take opps bidding into consideration -- but I guess you mean something else?

 

And, do you mean

a) "depending on opponents bidding and (its own) sampling" --- sure, sampling is part of Gibs core algorithm -- do you think that is bad?

or

b) "depending on opponents bidding and (opps') sampling" --- the sentence is grammatically ambiguous... not sure which one is intended...

?