Help
This seems crazy to me. The bid hint specifically said X is takeout. So robot should play me for at least 4 clubs. Did it really think 3D was going down on its measly 3 HCP + 4 ♦ to the jack? 3♦X+2 was a terrible reults. Seems to me it should support spades with 2+ or bid clubs with 3+.
Page 1 of 1
GIB passes clear takeout double
#1
- Group: Full Members
- Posts: 65
- Joined: 2012-April-01
- Gender:Male
Posted 2023-July-24, 17:44
This seems crazy to me. The bid hint specifically said X is takeout. So robot should play me for at least 4 clubs. Did it really think 3D was going down on its measly 3 HCP + 4 ♦ to the jack? 3♦X+2 was a terrible reults. Seems to me it should support spades with 2+ or bid clubs with 3+.
#2
- Group: Advanced Members
- Posts: 4,625
- Joined: 2014-March-15
- Gender:Male
Posted 2023-July-24, 19:58
GIB does understand it to be a takeout double, and the default database response is indeed 4♣.
GIB runs a simulation to see whether pass would work better, and concludes that it does for a few reasons.
When I ran it through the old version of GIB, it simulated 16 possible deals.
In 7 of the hands, East held a doubleton heart. The database tells East to *always* pull the double to 3♥ when holding 2 hearts. Not unsurprisingly, defending 3♥ works out better than playing in 4♣, so all of these will result in a win for the pass.
This is of course silly, even for robots, because in reality East is going to simulate too and see that passing is better on a number of those hands. But all extrapolations for the auction are based on the database definitions, full stop.
But even of the other 9 hands, only a single one had 3♦ making, on the nose. (Note it's using double dummy, so even if it had looked at your hand, it would have found 9 tricks in diamonds, not 11).
This is heavily outweighed by all of the other deals; all of the other 8 had 3♦x defeated, while 4♣ was going down (with one exception; the one time 4♣ made, 3♦x was down 3, so an even bigger win for passing).
On top of that, this is one of the special cases where every time 4♣ goes down in a simulation, it scores it as if the opponents had doubled, even though it never would actually be doubled. (See my lengthy post on how GIB simulations work here, under Going deeper #3).
While pass would have won in this simulation either way, all of the hefty penalties it assigned to 4♣ going down meant that it concludes pass actually wins *by a country mile*.
The bit about faulty extrapolation and auto-doubled undertricks aside, the hands it simulated all looked appropriate for what you may hold, so it may well be the case that even without these, its "measly 3 HCP and Jxxx of diamonds" really are enough to pass with.
GIB runs a simulation to see whether pass would work better, and concludes that it does for a few reasons.
When I ran it through the old version of GIB, it simulated 16 possible deals.
In 7 of the hands, East held a doubleton heart. The database tells East to *always* pull the double to 3♥ when holding 2 hearts. Not unsurprisingly, defending 3♥ works out better than playing in 4♣, so all of these will result in a win for the pass.
This is of course silly, even for robots, because in reality East is going to simulate too and see that passing is better on a number of those hands. But all extrapolations for the auction are based on the database definitions, full stop.
But even of the other 9 hands, only a single one had 3♦ making, on the nose. (Note it's using double dummy, so even if it had looked at your hand, it would have found 9 tricks in diamonds, not 11).
This is heavily outweighed by all of the other deals; all of the other 8 had 3♦x defeated, while 4♣ was going down (with one exception; the one time 4♣ made, 3♦x was down 3, so an even bigger win for passing).
On top of that, this is one of the special cases where every time 4♣ goes down in a simulation, it scores it as if the opponents had doubled, even though it never would actually be doubled. (See my lengthy post on how GIB simulations work here, under Going deeper #3).
While pass would have won in this simulation either way, all of the hefty penalties it assigned to 4♣ going down meant that it concludes pass actually wins *by a country mile*.
The bit about faulty extrapolation and auto-doubled undertricks aside, the hands it simulated all looked appropriate for what you may hold, so it may well be the case that even without these, its "measly 3 HCP and Jxxx of diamonds" really are enough to pass with.
#3
- Group: Full Members
- Posts: 65
- Joined: 2012-April-01
- Gender:Male
Posted 2023-July-24, 20:46
smerriman, on 2023-July-24, 19:58, said:
GIB does understand it to be a takeout double, and the default database response is indeed 4♣.
GIB runs a simulation to see whether pass would work better, and concludes that it does for a few reasons.
GIB runs a simulation to see whether pass would work better, and concludes that it does for a few reasons.
GIB should read Mel Colchimiro's book, especially the part about passing doubles for penalty. GIB is so unpredictable with higher level doubles it makes one doubt whether you can ever double anything over 2N without it being taken wrong or messed up. Check out my other post for a GIB short-circuit that worked in my favor but shouldn't. I often wonder what would happen if Jack Bridge played the current version of GIB. I see much more consistency from Jack.
#4
- Group: Advanced Members
- Posts: 2,770
- Joined: 2006-January-30
Posted 2023-July-25, 13:35
I agree with GIB on this one - it's not just the 4 diamonds, but the 5 hearts and zero spades. This hand screams defense.
#5
- Group: Advanced Members
- Posts: 2,871
- Joined: 2018-July-04
- Gender:Male
- Location:Australia
Posted 2023-July-25, 17:05
What would 4C by South mean - no need for a takeout given the existing auction - always a risk of being raised to 5 I guess
This is one of my never risk a takeout into game with GiB situations
This is one of my never risk a takeout into game with GiB situations
Page 1 of 1