[steve2005]

My link
Wide open in hearts and it cant take all the tricks
This was a just declare so bidding is Gibs
I think its AJx has to overtake trick 1 and lead back the J

[johnu]

Good defense by GIB. If you were in a slam, you would have gone down.

[smerriman]

GIB doesn't make any assumptions that the lead is from a strong suit or that West must have the ten of hearts.

Under that assumption, why GIB plays as it does is pretty straightforward.

On the first round, it plays the J because that is guaranteed to be at least as good as the Ace double dummy; it assumes it'll know when to overtake on the second round (standard double dummy flaw).

On the second round, it plays low because there are too many occasions when overtaking costs a trick at MPs (eg every time declarer has Txx). It comfortably finds the overtake at IMPs, where there's nothing to lose.

[pigpenz]

so GIB would never lead Q from KQ109...?
I gave up on robots, but sometimes get stuck playing with
them when someone leaves a game....they suck!

[johnu]

pigpenz, on 2022-June-29, 18:14, said:

so GIB would never lead Q from KQ109...?
I gave up on robots, but sometimes get stuck playing with
them when someone leaves a game....they suck!

Nah, Q lead is so avant-garde. But maybe the tried and tested 4th best 9.

[sfi]

smerriman, on 2022-June-27, 19:28, said:

GIB doesn't make any assumptions that the lead is from a strong suit or that West must have the ten of hearts.

Under that assumption, why GIB plays as it does is pretty straightforward.
[...]

Now all we need is a justification for the HQ rather than the H9. Does GIB really treat the Q, 10 and 9 as equals in this situation?

[smerriman]

sfi, on 2022-June-29, 19:56, said:

Now all we need is a justification for the HQ rather than the H9. Does GIB really treat the Q, 10 and 9 as equals in this situation?

Well, they are 100% equals after the J has been played.. sure, if West were human, you would play the 9 so that East would have choice but to overtake with the Ace, but that's not something that double dummy analysis could ever tell a robot.

[sfi]

smerriman, on 2022-June-29, 21:03, said:

Well, they are 100% equals after the J has been played.. sure, if West were human, you would play the 9 so that East would have choice but to overtake with the Ace, but that's not something that double dummy analysis could ever tell a robot.

So in practice do the robots have a 2/3 chance of defeating the contract or a 0% chance?

[smerriman]

0% - it'll always continue with the Q; I believe when choosing amongst equals it follows strict rules based on leading conventions etc. Sometimes that rule is to randomise (when following to a restricted choice situation), but in this case is high from equals.

[sfi]

smerriman, on 2022-June-29, 22:55, said:

0% - it'll always continue with the Q; I believe when choosing amongst equals it follows strict rules based on leading conventions etc. Sometimes that rule is to randomise (when following to a restricted choice situation), but in this case is high from equals.

I don't believe it randomises in restricted choice situations either - I'm yet to go wrong playing the robot to play the higher (or highest) of equals there. So maybe it just always plays the top of a sequence.

[smerriman]

Oh that's right, I forgot the forum post where we concluded it always played the higher card. But yeah, distinguishing equals would require recursing into partner's hand which adds a significant amount of computation time. Far too much for GIB at least.

[pescetom]

smerriman, on 2022-June-30, 01:53, said:

Oh that's right, I forgot the forum post where we concluded it always played the higher card. But yeah, distinguishing equals would require recursing into partner's hand which adds a significant amount of computation time. Far too much for GIB at least.

Recursion is usually more of a drag on programmer's CPU than that of the computer.
Could GIB not in any case maintain a table of equivalence which updates trick to trick with minimal computation time?
Just programming laziness I would think.

[smerriman]

pescetom, on 2022-June-30, 14:38, said:

Recursion is usually more of a drag on programmer's CPU than that of the computer.
Could GIB not in any case maintain a table of equivalence which updates trick to trick with minimal computation time?
Just programming laziness I would think.

I'm not sure what you mean. GIB doesn't have a problem computing equivalence. We're talking about *distinguishing* equivalent cards - ie figuring out that playing a low card from equals here will make it easier for East to find the correct defense than playing high from equals.

The only way to do that would be recursively - rather than just simulating deals from West's perspective, for every such deal, you have to move into another player's seat and then run a second simulation to see what ideas they might come up with if they had that hand. This recursion results in an exponential increase in the number of simulations run, and the time it takes to perform the calculations would immediately skyrocket.

[thepossum]

Out of curiosity I put this through another major Bridge program

Same auction, led King to East's 4, then switched to diamond, East unblocked hearts during subsequent play, and finally led back the Jack but sadly West had discarded one to many hearts and the contract was made exactly

EDIT forced a lead of the Q on trick two, sadly also blocked - I realise it already has been blocked by the play of the 4 and the Queen - my bad
Will report back if I find anything that defeats the contract.