[Cascade]

Dealer: North Vul: All Scoring: IMP

♠ KJ864 ♥ AQ7 ♦ Q3 ♣ T54

West North East South

 -     1♣    Pass  1♠
 Pass  2NT   Pass  3♦
 Pass  3NT   Pass  6NT
 Pass  Pass  Pass  

[barmar]

I know you should upgrade for a 5-card suit, but that does seem excessive.

[Jlall]

Probably just simulated hands and rated 6N to be the best bid, I doubt it was a bug.

[jdonn]

Yes he simulated but this might be a good place to have rules instead of simulation.

[Jlall]

jdonn, on Mar 2 2010, 03:34 PM, said:

Yes he simulated but this might be a good place to have rules instead of simulation.

Meh I think this would be the perfect time to simulate rather than have rules. I mean if you're not gonna simulate a signoff/invite/drive slam decision then you might as well scratch simulating all together.

[barmar]

Ari, when GIB does simulations like these, how many hands are usually dealt? I sometimes wonder if there's too much sampling error, because of the numerous times that GIB has blasted to slam and we're off two aces.

Then again, it may be me overbidding. It usually seems to happen with minors, I'll jump raise GIB to 5 and it will push to 6 because it thinks I'm showing a 21 count. Like the rest of us, it has holes in its system regarding minor suit constructive bidding. Sometimes its guess is right, sometimes not.

[Cascade]

I thought I had quite a good hand for the auction:

Dealer: South Vul: All Scoring: IMP

♠ QT ♥ KJ8 ♦ AK4 ♣ AJ632

Two honours in partner's suit, a five card suit, reasonable controls.

But slam had no play and would have been much worse without the ♠10.

I would be very surprised if the hand in the OP opposite a random 18-19 with only two spades would be on average good for slam.

[Cascade]

42% was my double dummy simulation for how often we had 12 tricks opposite a misfitting 18-19 HCP.

Perhaps I should have restricted it to 18 HCP GIB can invite.

Perhaps GIB thinks partner has 2=3=3=5 since with 4=4 maybe it thinks a diamond opening is mandatory.

[jdonn]

Jlall, on Mar 2 2010, 03:53 PM, said:

jdonn, on Mar 2 2010, 03:34 PM, said:

Yes he simulated but this might be a good place to have rules instead of simulation.

Meh I think this would be the perfect time to simulate rather than have rules. I mean if you're not gonna simulate a signoff/invite/drive slam decision then you might as well scratch simulating all together.

Wow don't agree at all. When there is a balanced hand opposite a balanced hand then HCP work very well, plus it's a really easy (relatively speaking) auction to create rules. The sims should be for when there is either an unfamiliar auction, competition, or lots of shape.

[nigel_k]

I don't know how many hands it can simulate in the time given, but I think you wouldn't need to simulate many to work out that pass beats 6NT and 4NT definitely beats 6NT.

The problem with simulation is playing double dummy, something like AJxxx opposite 10xx will make 4 tricks much more often than in real life. Double dummy analysis is bound to overbid slam hands so I tend to agree that resorting to rules, or maybe just requiring expected gain to be something larger than zero, would be a good idea. E.g. only bid slams when the IMP expectation is +1 or better.

This also highlights something I don't understand about GIB. In order to evaluate the gain/loss from bidding 4NT it has to first evaluate whether the simulated hand for opener is an accept or reject. Which requires yet another simulation. So there is a kind of exponential effect. It is not so bad here where each hand will have at most one decision to make, it could actually be a lot worse. How does it do that and still get through a decent number of examples?

[fred]

nigel_k, on Mar 2 2010, 11:21 PM, said:

This also highlights something I don't understand about GIB. In order to evaluate the gain/loss from bidding 4NT it has to first evaluate whether the simulated hand for opener is an accept or reject. Which requires yet another simulation. So there is a kind of exponential effect. It is not so bad here where each hand will have at most one decision to make, it could actually be a lot worse. How does it do that and still get through a decent number of examples?

It doesn't.

GIB doesn't do simulations within simulations. Instead it uses its bidding database to project how the bidding will continue. Errors and/or nonsense in the bidding database can cause a simulation to produce absurd results (which is one of the reasons why the work Ari is doing to improve the bidding database is so important).

Issues surrounding when and how to simulate are very interesting.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

[cherdanno]

Jlall, on Mar 2 2010, 03:53 PM, said:

jdonn, on Mar 2 2010, 03:34 PM, said:

Yes he simulated but this might be a good place to have rules instead of simulation.

Meh I think this would be the perfect time to simulate rather than have rules. I mean if you're not gonna simulate a signoff/invite/drive slam decision then you might as well scratch simulating all together.

I agree in theory but in practice I have seen GIB overbid towards slam very often on this auction (i.e. whenever the auction starts 1m-1M-2N but we don't have a fit).

[Winstonm]

Benjamin Franklin simulated this around 1786: a bird in hand....

[Cascade]

cherdanno, on Mar 3 2010, 05:12 PM, said:

Jlall, on Mar 2 2010, 03:53 PM, said:

jdonn, on Mar 2 2010, 03:34 PM, said:

Yes he simulated but this might be a good place to have rules instead of simulation.

Meh I think this would be the perfect time to simulate rather than have rules. I mean if you're not gonna simulate a signoff/invite/drive slam decision then you might as well scratch simulating all together.

I agree in theory but in practice I have seen GIB overbid towards slam very often on this auction (i.e. whenever the auction starts 1m-1M-2N but we don't have a fit).

I am not 100% sure but I think GIB has other problems on this auction like sometimes insisting on playing in a five-card major when partner denies three opposite.