[myprac]

My link

GIB fairly often discards a clear winner as a falsecard, as on the second trick of this deal. These occurrences happen in situations where there is no plausible scenario where the discard could be a helpful unblocking play. I can't help wondering if there's a way to adjust GIB's logic to eliminate these plays or at least make them less frequent.

Here's my guess as to what's going on. GIB checks to see whether a falsecard is potentially harmful by creating a number of random deals. If the falsecard doesn't lose a trick in any of those random deals, GIB will falsecard -- or perhaps will randomly select between the normal play and the falsecard. If this is correct, you'll get a result like the one above whenever all (or most?) of the random deals place the next highest card (♣J) in partner's hand, making it appear safe for GIB to discard the ♣Q.

It seems to me there should be an override that prevents GIB from making these discards solely as a falsecard -- in other words, it shouldn't throw a card that would be the highest outstanding card in the suit unless it discovers the need for an unblocking play. In other words, it has to discover a situation where it takes more tricks by throwing the high card, not merely a situation where it takes as many tricks by throwing a high card.

[barmar]

I'm looking into this. As far as I can tell, GIB has no notion of falsecarding. It either plays the card that its defense algorithm says is best, or it signals honestly if that's appropriate, or it plays its lowest card.

My initial guess is that it played the Queen intentionally to avoid being end-played. Double dummy, the only way to set the contract is for South to return the ♣10 at trick 2, then North plays its third ♣ and South switches to a ♦. Once South plays the A on trick 2, the only way to set the contract is to assume that South has the J, so North needs to unblock. Then South can play his J and switch to a ♦.

[calm01]

GIB frequently falsecards as a defender when trumping in, preferring to ruff say with the Jack of trumps when the 4 or 7 of trumps will do nicely.

More often than not there is no alternative way of playing trumps as declarer nor for declarer to assume the defender has no lower trumps to trump again 'knowing' that the Jack was the defenders lowest trump.

So the false card normally fails to gain and sometimes loses by providing an otherwise impossible entry to dummy or just giving away a trump trick. A net losing strategy. I once made an impossible slam thanks to the trump 10 becoming an entry to dummy.

The GIB logic appears flawed (if logic is what you call it). If done by a human player the behaviour may fall into the category of grandstanding.

When down to two cards of equal value, GIB will usually discard the higher - again if human it would be showing off.

Perhaps GIB is programmed to appear to be an emotionally challenged palooka to make us mere humans feel right at home!

Calm01

[myprac]

barmar, on 2011-November-04, 13:36, said:

As far as I can tell, GIB has no notion of falsecarding.

Interesting. I have an earlier version of GIB's convention card (text description, printed about a year ago) which explicitly states, in the overview, "Gib does falsecard." Checking the current version I see that this statement has been removed. I'm pretty sure GIB still falsecards fairly often, but I'll pay closer attention now to see if some of the plays that looked to me like falsecards were actually signals or strategic plays.

[myprac]

barmar, on 2011-November-04, 13:36, said:

My initial guess is that it played the Queen intentionally to avoid being end-played.

If that's what North was thinking, North was giving West too much credit, because West went down at the tables where North pitched the lower card. I suppose it's possible, though, that North played through some such scenario in its random deals and that's how it came to make this play, which I attributed to a simple desire to make a falsecard. BTW, this deal is from playing robots on fast setting where they don't think as deeply as in tournament play.

[Bbradley62]

myprac, on 2011-November-04, 16:50, said:

... West went down at the tables where North pitched the lower card...

For the analysts: this South played the same as OP for the auction and first two tricks, and this North didn't pitch the ♣Q.

but...

It's the Battle of Pitched Queens! Check out the bizarreness on Trick 4.

---

This post has been edited by Bbradley62: 2011-November-06, 08:32

[Bbradley62]

duplicate post

[barmar]

Maybe cefero was playing with an advanced robot. myprac said he was using a dumb bot.

Although that doesn't explain why cefero's bot played the ♦Q under the K.

[myprac]

My link

Here's a good one. Explain why it makes sense to play the 9♠ at trick 2. It can't even qualify as a falsecard because it's played from dummy. The contract makes easily with the obvious play of the 7♠.

[barmar]

Was that a basic or advanced robot? I can reproduce it with small simulations, but not when I set the number of hands to the maximum. The only holding where it makes a difference is when North has Tx, and in a small sample this sometimes doesn't come up.

[cloa513]

barmar, on 2011-November-20, 22:12, said:

Was that a basic or advanced robot? I can reproduce it with small simulations, but not when I set the number of hands to the maximum. The only holding where it makes a difference is when North has Tx, and in a small sample this sometimes doesn't come up.

Why doesn't GIB default to play low if the actual card by its similations (very limited) shouldn't matter?

[barmar]

I'm not sure. There's a place in the code where it says that when dummy isn't winning the trick, it should play the lowest of equivalents. I'll need to do more investigation to figure out why it's not doing that.

[myprac]

barmar, on 2011-November-20, 22:12, said:

Was that a basic or advanced robot?

I'm not sure. I was practicing with the basic robot, and found this in the traveler when I wondered why my IMP score for the hand was so bad. (At my table, EW GIB made the contract, so GIB's mistake at the other tables gave me a big negative.) I've assumed (correct me if I'm wrong) that those paying extra to play advanced robot play the same hands and end up in the same traveler, and if so it seems possible, though unlikely, that the advanced robot committed this error.

It seems clear enough that GIB made this play after random deals failed to find Tx in the North hand, but the question is why even consider playing the 9? This is why I started the thread: it seems to happen fairly often that GIB trashes a stopper or other winning card, for no apparent reason except to insert randomness into its play. Randomness makes sense when choosing between true equivalents (which is the basis of the principle of restricted choice), but not when cards haven't been determined to be equivalent, and certainly not when playing from dummy.

[Bbradley62]

barmar, on 2011-November-20, 22:51, said:

There's a place in the code where it says that when dummy isn't winning the trick, it should play the lowest of equivalents.

Isn't the point here that the 7 and the 9 are not equivalents if the 8 hasn't been played?

[barmar]

Bbradley62, on 2011-November-21, 00:22, said:

Isn't the point here that the 7 and the 9 are not equivalents if the 8 hasn't been played?

The 8 is irrelevant, since the 9 loses to the 10. So unless the 10 is doubleton, there's no difference in trick-taking ability between the 7 and 9. So if none of the hands it deals out have Tx in North, the cards appear to be equivalent.

GIB has no other way to evaluate the potential trick-taking ability of cards. All it does is deal out sample hands and play them double dummy, comparing the expected result of each possible play. If a screw case doesn't show up in its sample, it won't try to avoid the problem it causes. That's why it doesn't find 100% plays, or take safety plays.