Computer Bridge programs -- Opening Bids and Takeout Doubles

[P***@yahoo.com]

I've been using both Lorne's Bridge Analyzer program and GIB (Ginsberg's Intelligent Bridge -- this is the "robot" on BridgeBase Online) program to simulate hands (for a problem that hasn't been presented in this forum yet). I might note that Lorne focused more on simulation than on bidding, while GIB was built for and focused much more on bidding. So, when they disagree, I'd be much more surprised if Lorne's program was "right." (By the way, did I ever say how much I love Lorne's program!)

In my analysis of a specific hand, I start with randomly dealt hands where my partner would open 1H and where overcaller would bid a takeout double. In order for the hands to qualify as candidate hands for my sample, I've chosen only hands where both programs agree that the auction would start auction "1H (X)".

So, I can see hands where the two programs will bid differently. Here are my observations:

(1) Lorne's program will tend to open 11 HCP hands with five card majors, where "rule of 20 + 2" will not be met. GIB will not open such hands. I agree with GIB here. However, it appears that GIB will not open 11 point hands at the one level unless there are three quick tricks. Here's an example of a hand that does meet "rule of 20 + 2" that GIB will not open:

T5 JT984 K8 AK73

I disagree with GIB and would always open this hand 1H.

(2) GIB will always pass an opening bid with 4-3-3-3 shape and less than 15 HCPs. Lorne's program will always bid a takeout double with 13-14 HCPs. Mike Lawrence, in his book "Complete Book of Takeout Doubles" gives some examples of hands where he believes it is appropriate to takeout double on this shape when the HCPs are "working" (outside of the opened suit). GIB would pass and not takeout double on the following Mike Lawrence's example hands:

AK4 J98 763 AQ84, over a 1D opening.
AQ7 KQ5 8652 KJ2, over a 1D opening.

(3) Lorne's program will open shapely 6-4 hands with 10 HCPs. GIB will pass such hands. Here is an example of a hand where Lorne's program will open 1H and GIB will pass. With NLTC of 6.0 and both majors, I would not even hesitate to open this 1H at any vulnerability and scoring.

K864 AQJT43 - 983

(4) Power Doubles -- In game forcing hands that need no partner support, Lorne's program will give a power double before proceeding...just like I would do. Here's an example of a hand where GIB elects to overcall 1S over a 1H opening rather than Power Double.

AKT97632 QT T AK

---------

In discussions of computer game programs, Chess is clearly an example of where computers and statistics have put programs on top of the game. The top PC chess program (Komodo) will do well against the top grandmasters, even in a game where the programs give them pawn-odds. Checkers has been solved by computer (Chinook), so that no human can win a game against the computer. Computers are the best in Othello (Zebra) and backgammon (BKG and others).

The bridge players I speak with proudly note that bridge has not had the great progress in computer play that chess and other games have. Their pride is as if to say that bridge is a much harder game for a computer to play than the other games. However, after observing the bidding of GIB (supposedly one of the best bridge programs) on the first two bids of various bridge hands, it appears (in my view) that it is because the same attention has not been devoted to bridge as to other games.

Your thoughts? (Lorne is especially invited to comment!)

[Robert Chance]

On Saturday, 1 October 2016 17:52:58 UTC+1, ***@yahoo.com wrote:

A significant difference between Bridge and Chess is that at Bridge you play the opponents, not just the cards. At chess, the right move is the right move is the right move, no matter who you are playing against. At bridge, if you have a close decision you will take what you know about your opponents into account.

This gives the bridge expert an edge over any computer program. From experience, you can learn a program's foibles, but, until there is a significant advance in programming techniques, a program cannot do the same to you.

[p***@gmail.com]

I don't think bridge poses any special problem. I think it would be easier than Go, in which a computer is world champion.

I would guess it would be avoided because bridge is less popular and the programming is "messy," with lots of time-consuming special cases. The other games are "cleaner," with much less distracting detail.

I don't see why any programming breakthrough would be necessary to deal with opponents tendencies. Indeed, I would think that a computer with its perfect memory would be able to do a better job than a human. It seems like a quite basic sort of learning.

[jogs]

It's all about game theory. There are deterministic games and non deterministic games. Chess and Go are deterministic. These games are better for monte carlos. Bridge is a non deterministic game. The final result is not predetermined. Not good for monte carlos.

[Lorne Anderson]

Post by P***@yahoo.com
I've been using both Lorne's Bridge Analyzer program and GIB (Ginsberg's Intelligent Bridge -- this is the "robot" on BridgeBase Online) program to simulate hands (for a problem that hasn't been presented in this forum yet). I might note that Lorne focused more on simulation than on bidding, while GIB was built for and focused much more on bidding. So, when they disagree, I'd be much more surprised if Lorne's program was "right." (By the way, did I ever say how much I love Lorne's program!)
In my analysis of a specific hand, I start with randomly dealt hands where my partner would open 1H and where overcaller would bid a takeout double. In order for the hands to qualify as candidate hands for my sample, I've chosen only hands where both programs agree that the auction would start auction "1H (X)".
T5 JT984 K8 AK73
I disagree with GIB and would always open this hand 1H.
AK4 J98 763 AQ84, over a 1D opening.
AQ7 KQ5 8652 KJ2, over a 1D opening.
(3) Lorne's program will open shapely 6-4 hands with 10 HCPs. GIB will pass such hands. Here is an example of a hand where Lorne's program will open 1H and GIB will pass. With NLTC of 6.0 and both majors, I would not even hesitate to open this 1H at any vulnerability and scoring.
K864 AQJT43 - 983
(4) Power Doubles -- In game forcing hands that need no partner support, Lorne's program will give a power double before proceeding...just like I would do. Here's an example of a hand where GIB elects to overcall 1S over a 1H opening rather than Power Double.
AKT97632 QT T AK
---------
In discussions of computer game programs, Chess is clearly an example of where computers and statistics have put programs on top of the game. The top PC chess program (Komodo) will do well against the top grandmasters, even in a game where the programs give them pawn-odds. Checkers has been solved by computer (Chinook), so that no human can win a game against the computer. Computers are the best in Othello (Zebra) and backgammon (BKG and others).
The bridge players I speak with proudly note that bridge has not had the great progress in computer play that chess and other games have. Their pride is as if to say that bridge is a much harder game for a computer to play than the other games. However, after observing the bidding of GIB (supposedly one of the best bridge programs) on the first two bids of various bridge hands, it appears (in my view) that it is because the same attention has not been devoted to bridge as to other games.
Your thoughts? (Lorne is especially invited to comment!)

My program was wriiten originally in DOS around 1990, converted to
Windows in about 1999 and was always intended for simulation. The
bidding engine was added to give another way of biasing deals where you
wanted an opening bid or overcall but not as a way for developing and
auction to a final contract.

An example of what I might want to do is look at hands where partner
opens 1N and RHO bids 3D and you had a hand with a major and some range
of points and want to know if it is best to play 3M as to play, invite,
or GF or indeed play transfers. To do that you just need to deal hands
that are generally sensible rather than perfect examples for the bid
chosen so I have concntrated on that, together with options for
different methods such as strong club, various NT options, 4/5 card
majors, defesive overcall methods etc.

Also I set it up using what are common methods I have experienced so
with a 4333 13/14 count I would usuually double an opening in one of my
3 card suits but you should find checks and balances such as passing if
3-3 in unbid majors or wasted values in the suit opened.

As far as writing code to play the game I think it is currently too
difficult due to the imperfect info (you only see 26 cards for example)
and I have never tried. A computer bidding system may be possible (and
if interested look up the Cobra project) but as far as I know all
playing programs work via simulation rather than logic - ie they deal
and analyse a few bundred deals that match the info availble so far and
pick a play that works most often.

[jogs]

Post by Lorne Anderson
As far as writing code to play the game I think it is currently too
difficult due to the imperfect info (you only see 26 cards for example)
and I have never tried.

Actually each player can only see 13 cards. Hopefully partner is assisting by bidding his 13 cards. How often do any of us know partner's 13 cards during the auction?

[Lorne Anderson]

Post by jogs

Post by Lorne Anderson
As far as writing code to play the game I think it is currently too
difficult due to the imperfect info (you only see 26 cards for example)
and I have never tried.

Actually each player can only see 13 cards. Hopefully partner is assisting by bidding his 13 cards. How often do any of us know partner's 13 cards during the auction?

I was talking about play - bidding is a defferent issue and a lot easier
IMO.

[P***@yahoo.com]

I was talking about play - bidding is a different issue and a lot easier
IMO.

Thanks, Lorne, and I totally agree with you. I looked up the Cobra project and it is very interesting! Thanks for the reference.

[P***@yahoo.com]

GIB isn't just bad at bidding, it's also bad at play. Consider this hand:

Playing with spades as trump.
Dummy has KJ8 of diamonds and declarer has T43 of diamonds.
LHO leads the DA and another diamond.
Can you believe GIB actually finessed the DQ losing all three diamonds?

How can this be one of the "best" computer programs?

[jogs]

Post by P***@yahoo.com
Playing with spades as trump.
Dummy has KJ8 of diamonds and declarer has T43 of diamonds.
LHO leads the DA and another diamond.
Can you believe GIB actually finessed the DQ losing all three diamonds?
How can this be one of the "best" computer programs?

Sad, it could be the "best".
Poker without betting is easy to program. When one adds betting and reading opponents it get much, much harder.

[Will in New Haven]

Post by jogs

Post by P***@yahoo.com
Playing with spades as trump.
Dummy has KJ8 of diamonds and declarer has T43 of diamonds.
LHO leads the DA and another diamond.
Can you believe GIB actually finessed the DQ losing all three diamonds?
How can this be one of the "best" computer programs?

Sad, it could be the "best".
Poker without betting is easy to program. When one adds betting and reading opponents it get much, much harder.

Poker without betting is not poker. Betting, calling, raising and folding are all one _does_ in poker. Well, there's drawing but no one plays draw high and hardly anyone plays lowball draw seriously anymore, which is too bad.

The University of Alberta has done some serious work on poker. Nothing else to do in the prairie provinces.

[jogs]

Post by Will in New Haven
Poker without betting is not poker. Betting, calling, raising and folding are all one _does_ in poker. Well, there's drawing but no one plays draw high and hardly anyone plays lowball draw seriously anymore, which is too bad.
--
Will now in Pompano Beach

With lowball programming it without betting was quite useful. Unfortunately just by players playing tight the game became unbeatable. Those table fees are impossible to overcome when no one is giving action.

[f***@googlemail.com]

Post by P***@yahoo.com
Playing with spades as trump.
Dummy has KJ8 of diamonds and declarer has T43 of diamonds.
LHO leads the DA and another diamond.
Can you believe GIB actually finessed the DQ losing all three diamonds?
How can this be one of the "best" computer programs?

That's only a bad play if you think it's more likely the ace from was Ax than AQxxx. Human beings tend not to lead AQxxx and (some) like Ax as a lead. GIBs make ace leads far more often than people do (and are right more often than you might expect)

If you think that LHO would lead DA and another from all holdings including the DA, then (without seeing the rest of the hand which of course might have an impact) finessing is percentage.

[P***@yahoo.com]

Post by f***@googlemail.com

Post by P***@yahoo.com
Playing with spades as trump.
Dummy has KJ8 of diamonds and declarer has T43 of diamonds.
LHO leads the DA and another diamond.
Can you believe GIB actually finessed the DQ losing all three diamonds?
How can this be one of the "best" computer programs?

That's only a bad play if you think it's more likely the ace from was Ax than AQxxx. Human beings tend not to lead AQxxx and (some) like Ax as a lead. GIBs make ace leads far more often than people do (and are right more often than you might expect)
If you think that LHO would lead DA and another from all holdings including the DA, then (without seeing the rest of the hand which of course might have an impact) finessing is percentage.

The opening lead books by David Bird and Taf Anthias use computer simulation to show that opening leads of A's are underrated. However, even their computer analysis never called for an opening lead of A from AQxxx in an unbid suit in a suit contract. In this particular situation, GIB led the A from Ax (after having bid a five-card heart suit), but the GIB declarer finessed as if the A was led from AQxx. And GIB is expert at declarer play? It's hand #3 in the public spreadsheet offered in the "Another Counter-Intuitive Bidding Analysis" thread.

[Lorne Anderson]

Post by P***@yahoo.com

Post by f***@googlemail.com

Post by P***@yahoo.com
Playing with spades as trump.
Dummy has KJ8 of diamonds and declarer has T43 of diamonds.
LHO leads the DA and another diamond.
Can you believe GIB actually finessed the DQ losing all three diamonds?
How can this be one of the "best" computer programs?

That's only a bad play if you think it's more likely the ace from was Ax than AQxxx. Human beings tend not to lead AQxxx and (some) like Ax as a lead. GIBs make ace leads far more often than people do (and are right more often than you might expect)
If you think that LHO would lead DA and another from all holdings including the DA, then (without seeing the rest of the hand which of course might have an impact) finessing is percentage.

The opening lead books by David Bird and Taf Anthias use computer simulation to show that opening leads of A's are underrated. However, even their computer analysis never called for an opening lead of A from AQxxx in an unbid suit in a suit contract. In this particular situation, GIB led the A from Ax (after having bid a five-card heart suit), but the GIB declarer finessed as if the A was led from AQxx. And GIB is expert at declarer play? It's hand #3 in the public spreadsheet offered in the "Another Counter-Intuitive Bidding Analysis" thread.

This is a guess, based on my knowledge of how computer programmers are
likely to approach the problem. My guess is that GIB will create a
number of random deals that fit the auction with the ace in the hand on
led and the known cards in the right place. It will then look at
playing high or low on all those deals and choose the play with the best
overall outcome. I doubt very much if it has any other rules such as
leading an ace denies any holding with the Q but no K. Hence playing
low gains whenever the hand on lead has the Q and only loses if the
initial lead was a doubleton, whereas a human would know that a lead
from AQx(xx). is unlikely.

[P***@yahoo.com]

Post by Lorne Anderson
This is a guess, based on my knowledge of how computer programmers are
likely to approach the problem. My guess is that GIB will create a
number of random deals that fit the auction with the ace in the hand on
led and the known cards in the right place. It will then look at
playing high or low on all those deals and choose the play with the best
overall outcome. I doubt very much if it has any other rules such as
leading an ace denies any holding with the Q but no K. Hence playing
low gains whenever the hand on lead has the Q and only loses if the
initial lead was a doubleton, whereas a human would know that a lead
from AQx(xx). is unlikely.

That makes sense, Lorne, but it also shows that the programming of bridge is fairly elementary as compared to the programming of other games. In my experience conversing with other game programmers (I was a panelist, specializing in Othello, at an AAAI conference in 1987) it's not unusual to either build in such rules or to pick the random leads from a sample database of tournament occurrences, which would naturally set much higher probabilities of occurrence to the lead from an Ax as compared to a lead from AQxx.

Even if such databases of human tournament occurrences did not exist, neural network style learning might make use of developed databases from computer-vs-computer play. These things have been employed in other games (Michael Buro developed such a database in building his Othello program at the University of Paderborn in Germany -- see https://skatgame.net/mburo/log.html).

[rhm]

Post by P***@yahoo.com

Post by Lorne Anderson
This is a guess, based on my knowledge of how computer programmers are
likely to approach the problem. My guess is that GIB will create a
number of random deals that fit the auction with the ace in the hand on
led and the known cards in the right place. It will then look at
playing high or low on all those deals and choose the play with the best
overall outcome. I doubt very much if it has any other rules such as
leading an ace denies any holding with the Q but no K. Hence playing
low gains whenever the hand on lead has the Q and only loses if the
initial lead was a doubleton, whereas a human would know that a lead
from AQx(xx). is unlikely.

That makes sense, Lorne, but it also shows that the programming of bridge is fairly elementary as compared to the programming of other games. In my experience conversing with other game programmers (I was a panelist, specializing in Othello, at an AAAI conference in 1987) it's not unusual to either build in such rules or to pick the random leads from a sample database of tournament occurrences, which would naturally set much higher probabilities of occurrence to the lead from an Ax as compared to a lead from AQxx.
Even if such databases of human tournament occurrences did not exist, neural network style learning might make use of developed databases from computer-vs-computer play. These things have been employed in other games (Michael Buro developed such a database in building his Othello program at the University of Paderborn in Germany -- see https://skatgame.net/mburo/log.html).

The problem is much more complex.
A computer can deal out hands randomly which fits some constraints from the bidding, giving the opening leader the diamond ace. Though a big challenge this is the easy part.
But to do it properly it would also have to simulate whether a computer (Let's forget about humans and human tendencies and prejudices) would lead the diamond ace from the simulated hands.
The computer should not be interested in the question how likely it is, that opening leader was a priory dealt the queen as well, but also how likely he would lead the ace if that was the case compared to when it was not.
This would require more simulations based on what the opening leader knows and what he would lead based again on bidding constraints.
To my knowledge no Bridge playing program does this type of second level simulation. It would give the robot information how a robot as an opponent would likely behave.
Then we come to the third problem assume the diamond holding was as described: Dummy has KJx and declarer Txx.
A standard DD simulation program would always assume declarer will take the right decision, which works at the table. Result: It does deter the robot from leading the suit in the first place.
But single dummy this strategy (declarer always the right thing) is impossible.
It is easy to see that this type of simulation does not lead perfect results, and it requires so much computer power, that this type of simulation will not be feasible with sufficient large samples in the foreseeable future, assuming the computer is supposed to play a card within a given time frame.

[P***@yahoo.com]

Post by rhm

Post by P***@yahoo.com

Post by Lorne Anderson
This is a guess, based on my knowledge of how computer programmers are
likely to approach the problem. My guess is that GIB will create a
number of random deals that fit the auction with the ace in the hand on
led and the known cards in the right place. It will then look at
playing high or low on all those deals and choose the play with the best
overall outcome. I doubt very much if it has any other rules such as
leading an ace denies any holding with the Q but no K. Hence playing
low gains whenever the hand on lead has the Q and only loses if the
initial lead was a doubleton, whereas a human would know that a lead
from AQx(xx). is unlikely.

That makes sense, Lorne, but it also shows that the programming of bridge is fairly elementary as compared to the programming of other games. In my experience conversing with other game programmers (I was a panelist, specializing in Othello, at an AAAI conference in 1987) it's not unusual to either build in such rules or to pick the random leads from a sample database of tournament occurrences, which would naturally set much higher probabilities of occurrence to the lead from an Ax as compared to a lead from AQxx.
Even if such databases of human tournament occurrences did not exist, neural network style learning might make use of developed databases from computer-vs-computer play. These things have been employed in other games (Michael Buro developed such a database in building his Othello program at the University of Paderborn in Germany -- see https://skatgame.net/mburo/log.html).

The problem is much more complex.
A computer can deal out hands randomly which fits some constraints from the bidding, giving the opening leader the diamond ace. Though a big challenge this is the easy part.
But to do it properly it would also have to simulate whether a computer (Let's forget about humans and human tendencies and prejudices) would lead the diamond ace from the simulated hands.
The computer should not be interested in the question how likely it is, that opening leader was a priory dealt the queen as well, but also how likely he would lead the ace if that was the case compared to when it was not.
This would require more simulations based on what the opening leader knows and what he would lead based again on bidding constraints.
To my knowledge no Bridge playing program does this type of second level simulation. It would give the robot information how a robot as an opponent would likely behave.
Then we come to the third problem assume the diamond holding was as described: Dummy has KJx and declarer Txx.
A standard DD simulation program would always assume declarer will take the right decision, which works at the table. Result: It does deter the robot from leading the suit in the first place.
But single dummy this strategy (declarer always the right thing) is impossible.
It is easy to see that this type of simulation does not lead perfect results, and it requires so much computer power, that this type of simulation will not be feasible with sufficient large samples in the foreseeable future, assuming the computer is supposed to play a card within a given time frame.

The point I was trying to make was not that the computer should do the simulations within the program (very time consuming!), but that a huge database of games should be used to develop the program's rules to lead to the conclusion, for example, that leading from AQxx is unlikely relative to leading from Ax. Most of the great computer programs in other games make use of massive databases for training of pattern recognition (via statisical regression or neural network training). Rather than simulation, I'd favor a table lookup of probabilities of database occurrences to address this problem. And as long as the database is developed from good players (including good computer programs vs themselves), the probabilities will be reasonable.

[Barry Margolin]

Post by P***@yahoo.com
The point I was trying to make was not that the computer should do the
simulations within the program (very time consuming!), but that a huge
database of games should be used to develop the program's rules to lead to
the conclusion, for example, that leading from AQxx is unlikely relative to
leading from Ax.

They definitely need something like this.

GIB, like humans, practically never underleads an ace against a suit
contract. Yet I can't count the number of times I've seen it play K from
KJ(x)(x) in dummy. No human would do this, since they "know" that RHO
has the ace, but GIB doesn't make this inference.

[jogs]

Post by Barry Margolin
They definitely need something like this.
GIB, like humans, practically never underleads an ace against a suit
contract. Yet I can't count the number of times I've seen it play K from
KJ(x)(x) in dummy. No human would do this, since they "know" that RHO
has the ace, but GIB doesn't make this inference.
--
Barry Margolin
Arlington, MA

That may be true with the masses. At a regional Von Der Porten underlead an ace in a part score suit contract(mps) against me. It was the killing lead.

[Barry Margolin]

Post by jogs

Post by Barry Margolin
They definitely need something like this.
GIB, like humans, practically never underleads an ace against a suit
contract. Yet I can't count the number of times I've seen it play K from
KJ(x)(x) in dummy. No human would do this, since they "know" that RHO
has the ace, but GIB doesn't make this inference.

That may be true with the masses. At a regional Von Der Porten underlead an
ace in a part score suit contract(mps) against me. It was the killing lead.

"Practically never" is not the same as "never". Lawrence's book on
opening leads spends most of the book saying that it's the worst option
in most cases, waiting until one of the last chapters to discuss the
rare cases where you should try it.

Don't forget that partner will also not expect it. If you underlead an
ace, and he sees a queen in dummy, he probably won't put up his king,
since he expects it to be taken by declarer's presumed ace.

[Bruce Evans]

Post by Barry Margolin

Post by jogs

Post by Barry Margolin
They definitely need something like this.
GIB, like humans, practically never underleads an ace against a suit
contract. Yet I can't count the number of times I've seen it play K from
KJ(x)(x) in dummy. No human would do this, since they "know" that RHO
has the ace, but GIB doesn't make this inference.

That may be true with the masses. At a regional Von Der Porten underlead an
ace in a part score suit contract(mps) against me. It was the killing lead.

"Practically never" is not the same as "never". Lawrence's book on
opening leads spends most of the book saying that it's the worst option
in most cases, waiting until one of the last chapters to discuss the
rare cases where you should try it.
Don't forget that partner will also not expect it. If you underlead an
ace, and he sees a queen in dummy, he probably won't put up his king,
since he expects it to be taken by declarer's presumed ace.

Actually, it is trivial for partner to put up the K if you play that the
lead of a low card shows an honor. Since the Q is in dummy, when you have
the K then partner is marked with the A. The honor shouldn't be the J
even if you play that the J is an honor, since a lead from an unsupported
J is much worse than a lead from an unsupported A.

If you play that the lead of a low card is just count, then good defenders
can still often guess if the lead was from the A, based on whether the
bidding indicates such a lead and whether you need the leader to have the
A to beat the contract. At matchpoints, overtricks matter so the decision
is harder.

Bruce

[KWSchneider]

The problem is much more complex.=20
A computer can deal out hands randomly which fits some constraints from the=
bidding, giving the opening leader the diamond ace. Though a big challeng=
e this is the easy part.=20
But to do it properly it would also have to simulate whether a computer (Le=
t's forget about humans and human tendencies and prejudices) would lead the=
diamond ace from the simulated hands. =20

This is almost "chicken and egg". First, GIB the opening leader takes the auction, his hand shape, and simulates a significant number of hands (say 250) that meets these conditions, then analyzes which lead makes the most sense for each hand - ie it ranks them. This is what GIB does.

GIB the declarer now does the same. Leading or underleading an Ace isn't relevant to the discussion, since GIB the leader and GIB the declarer essentially don't have "rules" (other than GIB the leader has some "standard" programmable leads like 4th best against notrump, A from AK, highest of 3 touching honors, low from honor of partner's bid suit, singleton against trump, etc). If A from Ax makes the most sense, it will be lead. If A from AQx makes most sense, it will be lead.

The computer should not be interested in the question how likely it is, tha=
t opening leader was a priory dealt the queen as well, but also how likely =
he would lead the ace if that was the case compared to when it was not. =20
This would require more simulations based on what the opening leader knows =
and what he would lead based again on bidding constraints.=20

This is what GIB does - I've seen it take 5 minutes (i7) to play to trick one when set at its highest quality setting.

Kurt

[Steve Willner]

Post by Lorne Anderson
I was talking about play - bidding is a defferent issue and a lot
easier IMO.

Matt Ginsberg thought the opposite. He abandoned GIB when he realized
how hard bidding is. (I am claiming time coincidence here, not cause
and effect, but I wouldn't rule out the latter.)
One example doesn't prove GIB is bad overall. It did very well in that
par contest, for example. And how sure are you that GIB's play in the
example was wrong?

One problem GIB has is knowing how to interpret human bidding. That may
extend to human play as well. When it comes to raw technique, though,
everything I've seen suggests GIB is pretty good.

Nothing I've seen suggests GIB has any clue how to bid.

[KWSchneider]

I've been using both Lorne's Bridge Analyzer program and GIB (Ginsberg's In=
telligent Bridge -- this is the "robot" on BridgeBase Online) program to si=
mulate hands (for a problem that hasn't been presented in this forum yet). =
I might note that Lorne focused more on simulation than on bidding, while =
GIB was built for and focused much more on bidding. So, when they disagree=
, I'd be much more surprised if Lorne's program was "right." (By the way, d=
id I ever say how much I love Lorne's program!)

I've been using GIB for more than 13 years now, and I wouldn't characterize it as "built for and focused much more on bidding". GIB is a world-class declarer, and a crappy bidder - since all of the bids (and leads) are rule-based. My experience has been predominantly using the CLI engine, not the GUI.

I'd be interested in learning more about how you use GIB, as DD is not its predominant strength.

Kurt

[P***@yahoo.com]

Post by KWSchneider
I've been using GIB for more than 13 years now, and I wouldn't characterize it as "built for and focused much more on bidding". GIB is a world-class declarer, and a crappy bidder - since all of the bids (and leads) are rule-based. My experience has been predominantly using the CLI

I use GIB only for bidding and not for play. I screen random hands for certain starting bids -- and I make sure that GIB will bid the random hands according to the starting bids before the hands become candidates to test.

But I wanted to see how its SD play did versus DD (which I solved for using Lorne's program). All of the results will be stored in a public spreadsheet, along with my own "blind bids" based on my own system...which is clearly better than GIB's bidding system for the competitively bid hands being tested.

After I publish, I'll invite people to correct my bidding so that the bidding will be based on "best forum advice."

[Barry Margolin]

Post by P***@yahoo.com
I've been using both Lorne's Bridge Analyzer program and GIB (Ginsberg's
Intelligent Bridge -- this is the "robot" on BridgeBase Online) program to
simulate hands (for a problem that hasn't been presented in this forum yet).
I might note that Lorne focused more on simulation than on bidding, while GIB
was built for and focused much more on bidding. So, when they disagree, I'd
be much more surprised if Lorne's program was "right." (By the way, did I
ever say how much I love Lorne's program!)

Just wondering which version of GIB you're using? The download version,
which hasn't been updated in years, or the version on BBO, which has had
many updates to its bidding rules database?

[P***@yahoo.com]

Post by Barry Margolin

Post by P***@yahoo.com
I've been using both Lorne's Bridge Analyzer program and GIB (Ginsberg's
Intelligent Bridge -- this is the "robot" on BridgeBase Online) program to
simulate hands (for a problem that hasn't been presented in this forum yet).
I might note that Lorne focused more on simulation than on bidding, while GIB
was built for and focused much more on bidding. So, when they disagree, I'd
be much more surprised if Lorne's program was "right." (By the way, did I
ever say how much I love Lorne's program!)

Just wondering which version of GIB you're using? The download version,
which hasn't been updated in years, or the version on BBO, which has had
many updates to its bidding rules database?
--
Barry Margolin
Arlington, MA

I ordered the CD from Bridge World within the past month, as I didn't find a "download" version available online. Could it be that the BBO version has updated bidding rules that aren't being offered in the version that's for sale? Doesn't sound like a great business plan to me...but it might be possible! Thanks for the info, Barry.

[Barry Margolin]

Post by P***@yahoo.com

Post by Barry Margolin

Post by P***@yahoo.com
I've been using both Lorne's Bridge Analyzer program and GIB (Ginsberg's
Intelligent Bridge -- this is the "robot" on BridgeBase Online) program to
simulate hands (for a problem that hasn't been presented in this forum yet).
I might note that Lorne focused more on simulation than on bidding, while GIB
was built for and focused much more on bidding. So, when they disagree, I'd
be much more surprised if Lorne's program was "right." (By the way, did I
ever say how much I love Lorne's program!)

Just wondering which version of GIB you're using? The download version,
which hasn't been updated in years, or the version on BBO, which has had
many updates to its bidding rules database?
--
Barry Margolin
Arlington, MA

I ordered the CD from Bridge World within the past month, as I didn't find a
"download" version available online. Could it be that the BBO version has
updated bidding rules that aren't being offered in the version that's for
sale? Doesn't sound like a great business plan to me...but it might be
possible! Thanks for the info, Barry.

The BBO service is the core of the company. The software CDs are now
very much a side business, with very little resources devoted to them.

I think we updated GIB 2-3 years ago, but we don't do regular updates of
them like we do the online version.

[KWSchneider]

Post by Barry Margolin
I think we updated GIB 2-3 years ago, but we don't do regular updates of
them like we do the online version.
--
Barry Margolin
Arlington, MA

Barry - I have GIB dating from 2002 (6.1.0), and updates through the website. It seems to me the last update I could find is 6.2.0 from 2009. If there was an update 2-3 years ago, I'd certainly like to get it.

I'm a registered owner of GIB.

Kurt

[P***@yahoo.com]

Post by KWSchneider

Post by Barry Margolin
I think we updated GIB 2-3 years ago, but we don't do regular updates of
them like we do the online version.
--
Barry Margolin
Arlington, MA

Barry - I have GIB dating from 2002 (6.1.0), and updates through the website. It seems to me the last update I could find is 6.2.0 from 2009. If there was an update 2-3 years ago, I'd certainly like to get it.
I'm a registered owner of GIB.
Kurt
--
Posted by Mimo Usenet Browser v0.2.5
http://www.mimousenet.com/mimo/post

My version is also 6.2.0, and I just got it within the past month.

[John Hall]

Post by KWSchneider

Post by Barry Margolin
I think we updated GIB 2-3 years ago, but we don't do regular updates of
them like we do the online version.

Barry - I have GIB dating from 2002 (6.1.0), and updates through the
website. It seems to me the last update I could find is 6.2.0 from
2009. If there was an update 2-3 years ago, I'd certainly like to get
it.
I'm a registered owner of GIB.

All of the above applies to me too. If there have been more recent
updates to the version that's on BBO (where I'm a registered user
although I don't play there), is there any way that I can download them?

[Barry Margolin]

Post by KWSchneider

Post by Barry Margolin
I think we updated GIB 2-3 years ago, but we don't do regular updates of
them like we do the online version.
--
Barry Margolin
Arlington, MA

Barry - I have GIB dating from 2002 (6.1.0), and updates through the website.
It seems to me the last update I could find is 6.2.0 from 2009. If there was
an update 2-3 years ago, I'd certainly like to get it.
I'm a registered owner of GIB.

I think we may have only updated the version that's included in the BBO
application.