[wbartley]

Just Play Bridge robot.

---

This has to be some sort of bug. There's no reason not to take the diamond finesse knowing the full distribution and knowing that North still has another diamond.

[johnu]

Use the full editor and click on the symbol that has a ♠ with reddish edges on both sides. That will bring up the hand editor where you can enter cards and bidding.

[virgosrock]

johnu, on 2017-November-01, 13:56, said:

Use the full editor and click on the symbol that has a ♠ with reddish edges on both sides. That will bring up the hand editor where you can enter cards and bidding.

johnu how do you invoke the hand editor. particularly with respect to money bridge hands which are available only here
http://www.bridgebas.../mbactivity.php

thanks vrock

[wbartley]

Took johnu's advice and entered the hand properly.

Thanks much for pointing this out!

[smerriman]

Taking the diamond finesse only works if you opened the bidding 1♥ (in second seat no less) with 8 AJ7654 9xxx A8. Perhaps you might do this on the rarest of occasions, but that would be pretty wild - to GIB, both are 0% chances. Making assumptions based on the bidding is an important part of the GIB logic; I agree that it often overassumes too much, but it's tough to call this one a bug.

[wbartley]

smerriman, on 2017-November-06, 01:16, said:

Taking the diamond finesse only works if you opened the bidding 1♥ (in second seat no less) with 8 AJ7654 9xxx A8. Perhaps you might do this on the rarest of occasions, but that would be pretty wild - to GIB, both are 0% chances. Making assumptions based on the bidding is an important part of the GIB logic; I agree that it often overassumes too much, but it's tough to call this one a bug.

If this isn't a bug, it's a design flaw. There is no way you will convince me that conceding down one when there are possible layouts where a different play succeeds is correct behavior. For example, had GIB gone up with the ace (a play that is, in my opinion, still incorrect design) you could at least argue that it took the opening bid into consideration hoping that the defender would mistakenly play his king under the ace.

If GIB uses a probabilistic model when determining which card to play, it should never assign a probability of zero to a layout that is physically possible. In this case, if it assigns the layout where North started with Kxx in diamonds a probability of .0000000001 and it assumes best play on both sides, it will play the queen or jack because all other layouts result in down one no matter which card it plays and thus the expected value of the play of the queen or jack is infinitesimally greater than the other cards.

[johnu]

smerriman, on 2017-November-06, 01:16, said:

Taking the diamond finesse only works if you opened the bidding 1♥ (in second seat no less) with 8 AJ7654 9xxx A8. Perhaps you might do this on the rarest of occasions, but that would be pretty wild - to GIB, both are 0% chances. Making assumptions based on the bidding is an important part of the GIB logic; I agree that it often overassumes too much, but it's tough to call this one a bug.

I call it a bug/design flaw/limitation/etc. Whatever you call it, fuzzy logic would lead to finessing, since playing low on the 10 is 1000% guaranteed to lose a trick, instead of finessing which has some chance of winning if a human South has made an odd bid. And in other auctions, even GIB makes bids that have a description that isn't closely related to the actual hand.

[Stephen Tu]

GIB certainly would defend/declare better if it would simulate vs. a broader range of opp's strength (+/- a K) and distribution to break ties among card choices. Ties sometimes are no longer ties when opp is off-shape or strength. But it would take maybe a lot more resources to calculate and maybe tricky to figure out when it should be simulating across a broader range.

[johnu]

Stephen Tu, on 2017-November-06, 16:28, said:

GIB certainly would defend/declare better if it would simulate vs. a broader range of opp's strength (+/- a K) and distribution to break ties among card choices. Ties sometimes are no longer ties when opp is off-shape or strength. But it would take maybe a lot more resources to calculate and maybe tricky to figure out when it should be simulating across a broader range.

In the end game, how many extra computer resources would it take to just play the queen or jack which can't possibly cost, instead of playing a random card like GIB seems to play (similar to passing unexpectedly when there isn't a bid in the bidding database). Yes, GIB would need to be reprogrammed which is almost certainly unrealistic, but that shows the futility of trying to patch a program that has reached the end of the line.

[wbartley]

This is a completely deterministic situation with a computationally manageable number of possible layouts of the outstanding cards. Unless you assign a probability of 0 to North holding Kx of diamonds, you should not get this play wrong. You can empirically determine the total number of unique possible layouts given what is known about the distribution of the outstanding cards and accurately estimate the amount of time necessary to do a full double dummy analysis. If that estimate falls below a certain threshold, assign a probability to each of the layouts given clues from the bidding and play, making sure not to assign a probability of zero to any possible layout. At this point it becomes a simple expected value problem. This is a five card ending with the distribution of the suits completely known. A full double dummy analysis of all possible positions has to be on the sub-millisecond scale.

If the number of possible permutations times the average amount of time to do a double dummy analysis exceeds your maximum threshold, you resort to sampling but you should generate more hands than you're going to do double dummy analysis on and rank them based on the probability analysis used in the determinative case, choosing only the most likely n hands.