[Furlan]

{comments}

---

[cloa513]

Furlan, on 2012-September-19, 19:24, said:

{comments}

---

That's because GIB does simulations filled with absolute rubbish. In this case it simulates that West has a high card point in his hand so it will gain from this bizarre play.

[Antrax]

This is indeed quote curious. GIB E knows his partner can't have anything, and knows declarer has an entry to the hearts. Normally when things like this happen it's because GIB simulates that he can take the contract two down, but in this case that seems pretty unlikely. It knows declarer must be 42(34) because of the missing minor-suit honors, so cashing the ♦A can't be wrong.

[TylerE]

For those who are more familiar with the GIB algorithms: It is my understanding that it basically does the simulations, and then picks a play on strict expected value. Would it be possible to make the play look at the median value as well?

I'm having a hard time putting it into works, but basically I'd like GIB, when it has two possible plays, to lean more towards the one that minimizes variance, rather than maximizes expected values. I would think that would tend to decrease how frequently it plays for unlikely occurrences.

Some heuristics might be helpful too, that adjust the expected values. (e.g. when defending, a line that is 100% for -1 should beat a line that is 80% for -1, 10% for -3, and 10% =.

[Stephen Tu]

This is an odd one. My home GIB cashes out, even when set to MP and a short thinking time. If I force it to lead low D when in with HK, it again grabs its ace on its 2nd chance when it should be even clearer. Is the GIB on BBO set so fast and examining so few deals that it only deals south say 4243 distribution hands where it can get away with ducking?