a****s
发帖数: 524 | 1 MP, BOTH
East:
S 962
H QJ987
D Q943
C 6
E S W N
P 1C X P
1H 2C 2N P
3D P 4D P
? | b***y
发帖数: 2804 | | p***r
发帖数: 20570 | 3 Partner's 4D is a very strange bid in MP. You didn't show much and he just
bid 4D to bypass 3NT, so he can't hold 3 hearts here and most likely he
should hold 5 diamonds. Also, he denies CA here because he didn't cuebid 4C.
So it can be something like: AKx Kx AKxxx Kxx. Now you should just bid 4H
to suggest a place to play.
Also, you can either play 3C over 2NT as a checkback with 5 H, or you can
play transfers over 2NT to show this shape easily. So you bid 3C to show 4+D
. Partner can still bid 3D to show fit and you then bid 3H to show 5+H and
partner can decide what to bid next.
【在 a****s 的大作中提到】
: MP, BOTH
: East:
: S 962
: H QJ987
: D Q943
: C 6
: E S W N
: P 1C X P
: 1H 2C 2N P
: 3D P 4D P
| b***y
发帖数: 2804 | 4 Since 3D is not forcing, using 4D here as invitational is very inefficient,
especially at MP. I would read 4D as forcing. But how can a hand that only
bids non-forcing 2NT now forces to game (and bypasses 3NT at MP)? It must be
a hand that is full of controls, and gets improved by the 3D bid. Even with
AKx/Kx/AKxxx/Kxx, bypassing 3NT is still strange, since this hand can bid
3S to show doubt on 3NT, but let partner decide. So the 4D bid shows strong
preference in a suit contract. I think that it may contain lots of diamonds
but miss the top honor, for example AK/Ax/KQxxxx/Kxx.
And yes, after 2NT, transfer works well, but many partnerships aren't
equipped with that. To play transfer, partnership must also decide where it
starts, for example if the opening were 1D, then over partner's 2NT, 3C
probably ought to be natural, not transfer. When each sequence has to be
treated differently, the details grow to an un-manageable degree for casual
partnership. The concept is good though, and for partnerships that are
willing to spend time on it, there will be benefits when it comes up.
【在 p***r 的大作中提到】
: Partner's 4D is a very strange bid in MP. You didn't show much and he just
: bid 4D to bypass 3NT, so he can't hold 3 hearts here and most likely he
: should hold 5 diamonds. Also, he denies CA here because he didn't cuebid 4C.
: So it can be something like: AKx Kx AKxxx Kxx. Now you should just bid 4H
: to suggest a place to play.
: Also, you can either play 3C over 2NT as a checkback with 5 H, or you can
: play transfers over 2NT to show this shape easily. So you bid 3C to show 4+D
: . Partner can still bid 3D to show fit and you then bid 3H to show 5+H and
: partner can decide what to bid next.
| m****r
发帖数: 6639 | 5 i don't think the chances of making 5d or 4h are good enough. i am passing
in MP. i probably bid 5d in imp.
i say that without any understand of what 2nt shows.
,
be
with
strong
diamonds
it
【在 b***y 的大作中提到】
: Since 3D is not forcing, using 4D here as invitational is very inefficient,
: especially at MP. I would read 4D as forcing. But how can a hand that only
: bids non-forcing 2NT now forces to game (and bypasses 3NT at MP)? It must be
: a hand that is full of controls, and gets improved by the 3D bid. Even with
: AKx/Kx/AKxxx/Kxx, bypassing 3NT is still strange, since this hand can bid
: 3S to show doubt on 3NT, but let partner decide. So the 4D bid shows strong
: preference in a suit contract. I think that it may contain lots of diamonds
: but miss the top honor, for example AK/Ax/KQxxxx/Kxx.
: And yes, after 2NT, transfer works well, but many partnerships aren't
: equipped with that. To play transfer, partnership must also decide where it
| b***y
发帖数: 2804 | 6 2NT probably just shows 19-21 balanced hand, or equivalent.
Cannot imagine 4D being passable. I am bidding 4H, but slightly worried
about missing slam...
【在 m****r 的大作中提到】
: i don't think the chances of making 5d or 4h are good enough. i am passing
: in MP. i probably bid 5d in imp.
: i say that without any understand of what 2nt shows.
:
: ,
: be
: with
: strong
: diamonds
: it
| c***n
发帖数: 1511 | 7 I think maybe the 3D was a little too strong, especially in MP and I assumed
that 2NT was not forcing.
P could have 4432 with 2 hearts and 4 clubs, 2NT seemed to be the best deal
for that hand.
Note that P didn't X 2c again, neither did he cue bid 3C, so 2NT was most
likely not forcing.
Given that, I'd have to assume that P's 4D got good Diamond match, so his
hand was probly 4243.
In MP I would pass, in IMP I might bet 5D, most likely not. | b***y
发帖数: 2804 | 8 Of course 2NT is not forcing, it just shows 19-21. 3D cannot be forcing
either, it is a weak bid, can be 5-5 in H-D, 0 HCP. That's why 4D is a very
odd bid. It must be based on lots of diamonds. | a****s
发帖数: 524 | 9 如果要我写个机器人叫牌程序，大致会是这样的：
参照随便一本桥牌入门读物，同伴2NT的期望值是20点，
我有大牌5点，单张3点，联手牌力期望值28点。
同伴在2NT后又继续叫牌，他的牌力应该是高限。所以我
期望完成5阶定约。最后选择：5D。
就这么简单。
至于同伴越过3NT暗示控制好，还有绝大多数大牌位置都在
开叫人后面等等有利因素，对于一个入门级的机器人来说，
实在太高难了，就不予考虑了吧。
举这个例子的意思是，叫牌的基础既不是感觉，更不是约定，
而首先是牌力估价的量化。只要有意识地运用一些简单到不能
再简单的量化手段，很多很多问题的结论都是非常显然的。
【在 a****s 的大作中提到】
: MP, BOTH
: East:
: S 962
: H QJ987
: D Q943
: C 6
: E S W N
: P 1C X P
: 1H 2C 2N P
: 3D P 4D P
| b***y
发帖数: 2804 | 10 量化的方式很容易编进程序，但是叫牌结果经常很臭，否则机器人早就成桥牌专家了。
不过作为叫牌的入门速成，还是可行的。 | | | c***n
发帖数: 1511 | 11 If this is IMP I would agree with you. But for MP game, wouldn't you want to
bet 2NT rather than 3D, even though 3D might be the safest contract?
This is why I think 3D tended have strong invitational hints to it.
It's actually pretty attemptive to bet with diamonds, the only concern is
there might be some wasted pts on clubs.
But if play NT and opps lead clubs, you might steal 9-10 tricks --- 2 clubs,
4 diamonds, 3-4 hearts, 1-2 spades.
very
【在 b***y 的大作中提到】
: Of course 2NT is not forcing, it just shows 19-21. 3D cannot be forcing
: either, it is a weak bid, can be 5-5 in H-D, 0 HCP. That's why 4D is a very
: odd bid. It must be based on lots of diamonds.
| p***r
发帖数: 20570 | 12 Well, my suggestion to most novice to intermediate players is to play a new
suit at 3 level as at least one round forcing facing partner's 2NT bid. Of
course you can construct hands that have good plays in 3D and no play in 2NT
. However, the basic goal of bridge bidding is to find the correct game, not
to sign off. If you play 3D as sign off, you may have huge problems to
handle H-D two suiters with gf to slam values.
very
【在 b***y 的大作中提到】
: Of course 2NT is not forcing, it just shows 19-21. 3D cannot be forcing
: either, it is a weak bid, can be 5-5 in H-D, 0 HCP. That's why 4D is a very
: odd bid. It must be based on lots of diamonds.
| p***r
发帖数: 20570 | 13 5D is usually a very unpopular contract in pairs. Here, 4H really should be
the clear MP bid. Also, partner almost surely holds H honor doubleton in
this sequence.
【在 a****s 的大作中提到】
: 如果要我写个机器人叫牌程序，大致会是这样的：
: 参照随便一本桥牌入门读物，同伴2NT的期望值是20点，
: 我有大牌5点，单张3点，联手牌力期望值28点。
: 同伴在2NT后又继续叫牌，他的牌力应该是高限。所以我
: 期望完成5阶定约。最后选择：5D。
: 就这么简单。
: 至于同伴越过3NT暗示控制好，还有绝大多数大牌位置都在
: 开叫人后面等等有利因素，对于一个入门级的机器人来说，
: 实在太高难了，就不予考虑了吧。
: 举这个例子的意思是，叫牌的基础既不是感觉，更不是约定，
| b***y
发帖数: 2804 | 14 Slam value is impossible, you wouldn't have responded only 1H in first place
. For game hands, you can still cuebid first. I know this is not idea (
transfer is a better treatment), but when partner makes a limited bid on
balanced hands, responder with weak hands often wants to place/improve the
contract. Just like after 1NT opening, responder would sign off with 0 HCP.
Now we have transfer, but in the old days, 1NT-2H is just a sign off.
【在 p***r 的大作中提到】
: Well, my suggestion to most novice to intermediate players is to play a new
: suit at 3 level as at least one round forcing facing partner's 2NT bid. Of
: course you can construct hands that have good plays in 3D and no play in 2NT
: . However, the basic goal of bridge bidding is to find the correct game, not
: to sign off. If you play 3D as sign off, you may have huge problems to
: handle H-D two suiters with gf to slam values.
:
: very
| b***y
发帖数: 2804 | 15 In MP, if partner opens 2NT, are you not transfer (then pass) with 5-6
hearts and 0 HCP? When you have shapely hand (but very weak), often a suit
contract is much better than NT. It is not just about safety, you are trying
to improve the contract. Your hand can be worthless in NT but worth
something in suits.
【在 c***n 的大作中提到】
: If this is IMP I would agree with you. But for MP game, wouldn't you want to
: bet 2NT rather than 3D, even though 3D might be the safest contract?
: This is why I think 3D tended have strong invitational hints to it.
: It's actually pretty attemptive to bet with diamonds, the only concern is
: there might be some wasted pts on clubs.
: But if play NT and opps lead clubs, you might steal 9-10 tricks --- 2 clubs,
: 4 diamonds, 3-4 hearts, 1-2 spades.
:
: very
| p***r
发帖数: 20570 | 16 I am talking about a general guideline, not this specific hand. A sign off
in D works when you hold very weak and distributional hands, and also
partner holds some sort of support. However, it takes away the very valuable
and natural bidding space for gameforcing hands. This kind of old day
conventions are simply against the common bridge logic, "Game and slam
investigations are more important than the best partials". So you may land
at a few good partials and misbid games and slams.
For this specific hand, your partner has denied 4 or more hearts, so the
value for being able to play in 3D or 3H is even less.
place
.
【在 b***y 的大作中提到】
: Slam value is impossible, you wouldn't have responded only 1H in first place
: . For game hands, you can still cuebid first. I know this is not idea (
: transfer is a better treatment), but when partner makes a limited bid on
: balanced hands, responder with weak hands often wants to place/improve the
: contract. Just like after 1NT opening, responder would sign off with 0 HCP.
: Now we have transfer, but in the old days, 1NT-2H is just a sign off.
| b***y
发帖数: 2804 | 17 OK. I suppose a simple way is just to play that repeating first suit (3H
here) is non-forcing, new suit is forcing. For 5-5 very weak hand you just
give up playing in 3D, instead you rebid 3H.
If that is the case, then partner's 4D is forced, not a slam try. But I
probably still bid 4H. Partner can always correct to 5D with an unsuitable
hand for hearts, since he knows I have exactly 5 H and at least 4 D. Passing
wouldn't be a crime, but a little timid.
【在 p***r 的大作中提到】
: I am talking about a general guideline, not this specific hand. A sign off
: in D works when you hold very weak and distributional hands, and also
: partner holds some sort of support. However, it takes away the very valuable
: and natural bidding space for gameforcing hands. This kind of old day
: conventions are simply against the common bridge logic, "Game and slam
: investigations are more important than the best partials". So you may land
: at a few good partials and misbid games and slams.
: For this specific hand, your partner has denied 4 or more hearts, so the
: value for being able to play in 3D or 3H is even less.
:
| p***r
发帖数: 20570 | 18 Of course, our opinions converge here, 4H is the best MP bid.
Passing
【在 b***y 的大作中提到】
: OK. I suppose a simple way is just to play that repeating first suit (3H
: here) is non-forcing, new suit is forcing. For 5-5 very weak hand you just
: give up playing in 3D, instead you rebid 3H.
: If that is the case, then partner's 4D is forced, not a slam try. But I
: probably still bid 4H. Partner can always correct to 5D with an unsuitable
: hand for hearts, since he knows I have exactly 5 H and at least 4 D. Passing
: wouldn't be a crime, but a little timid.
| a****s
发帖数: 524 | 19 GIBs are better than "experts", they have long been at the rank
of world class.
Can someone credibly claim he/she plays better than GIB?
I know Fred Gitleman says he cannot, and neither Zia.
On the matter of bidding, there's some room for improvement, but I don't
think GIB is *ANY* worse than an average expert.
GIB's bidding relies more on judgement, i.e. things can be quantified, than
on conventions.
The problem of most human players, is not only they don't know how to bid
naturally without "conventions", but also lack of the basic discipline when
playing with a highly disciplined partner.
【在 b***y 的大作中提到】
: 量化的方式很容易编进程序，但是叫牌结果经常很臭，否则机器人早就成桥牌专家了。
: 不过作为叫牌的入门速成，还是可行的。
| a****s
发帖数: 524 | 20 I tend to agree with your "bid more, have more" philosophy.
But seems one should be allowed more latitude in MP games, rather than
in IMP games.
Since in MP, importance of a plus outweighs that of how much a plus is, it
is sensible to emphasis on finding better partials.
In IMP, game swings dominate. it makes more sense to bid on values.
to
clubs,
【在 c***n 的大作中提到】
: If this is IMP I would agree with you. But for MP game, wouldn't you want to
: bet 2NT rather than 3D, even though 3D might be the safest contract?
: This is why I think 3D tended have strong invitational hints to it.
: It's actually pretty attemptive to bet with diamonds, the only concern is
: there might be some wasted pts on clubs.
: But if play NT and opps lead clubs, you might steal 9-10 tricks --- 2 clubs,
: 4 diamonds, 3-4 hearts, 1-2 spades.
:
: very
| | | p***r
发帖数: 20570 | 21 I have no idea what you are talking about.
It's commonly known that Gib is a weak declarer,
defender and bidder. I played way more hands with gib than anybody in this
board. Gib's bidding system is very very flawed with many bugs. Gib plays no defensive signals and is not programmed to have some very basic interference of previous carding, (like if partner wins A, he denies K) . Gib frequently makes mistakes in the final 5 card setups when declaring. I don't really think you understand how gib bids. It's just based on more than 1 thousand rules with some simulations. When many rules can be applied in a situation, those rules may conflict with each other and you may see something like gib first bids 1D later bids 4C to show a strong rebiddable club suit with 3 or less D. Also, gib's bidding is so bad sometimes. For example: with AKQJxx - AKQJx AK, it just bids 4S over opp's 3H opening because it's based on so called "total points", so without 36 total points, it would never bid 7S by itself.
Good bidding judgment for gib? This is the most ridiculous claim I ever heard on this board.
than
when
【在 a****s 的大作中提到】
: GIBs are better than "experts", they have long been at the rank
: of world class.
: Can someone credibly claim he/she plays better than GIB?
: I know Fred Gitleman says he cannot, and neither Zia.
: On the matter of bidding, there's some room for improvement, but I don't
: think GIB is *ANY* worse than an average expert.
: GIB's bidding relies more on judgement, i.e. things can be quantified, than
: on conventions.
: The problem of most human players, is not only they don't know how to bid
: naturally without "conventions", but also lack of the basic discipline when
| b***y
发帖数: 2804 | 22 Yes, GIB is better than many "BBO experts", but still weaker than average C
flight player. "World class" is such a joke.
Even the dummy play (said to be the strongest skill that a robot can possess
), GIB routinely messes up at trick 10 or later. Of course bidding and
defense is even worse. Just think about the 5D bid that you said any robot
would do. Compared with 4H, choosing 5D is equivalent as choosing a play
line that cannot have a single winning case over a better line.
Of course, my comment is based on the experience with money-bridge GIBs. It
is possible that these GIBs are not of the best version, it can be a trial
version which is just buggy.
【在 a****s 的大作中提到】
: GIBs are better than "experts", they have long been at the rank
: of world class.
: Can someone credibly claim he/she plays better than GIB?
: I know Fred Gitleman says he cannot, and neither Zia.
: On the matter of bidding, there's some room for improvement, but I don't
: think GIB is *ANY* worse than an average expert.
: GIB's bidding relies more on judgement, i.e. things can be quantified, than
: on conventions.
: The problem of most human players, is not only they don't know how to bid
: naturally without "conventions", but also lack of the basic discipline when
| p***r
发帖数: 20570 | 23 wbridge5 and jack are much stronger I think. This year's winner is shark
bridge. I think it should also be a very decent program.
C
possess
It
【在 b***y 的大作中提到】
: Yes, GIB is better than many "BBO experts", but still weaker than average C
: flight player. "World class" is such a joke.
: Even the dummy play (said to be the strongest skill that a robot can possess
: ), GIB routinely messes up at trick 10 or later. Of course bidding and
: defense is even worse. Just think about the 5D bid that you said any robot
: would do. Compared with 4H, choosing 5D is equivalent as choosing a play
: line that cannot have a single winning case over a better line.
: Of course, my comment is based on the experience with money-bridge GIBs. It
: is possible that these GIBs are not of the best version, it can be a trial
: version which is just buggy.
| a****s
发帖数: 524 | 24 Sorry, I think anyone think GIB a weak declarer or defender is
simply lunatics.
GIB ranked 12th in the par contest in 1998, when the computing power were
much more moderate compared with today's hardware.
Sound you know GIB well, You must be familiar with these facts, then I am
just wondering on what ground one can call a player beat majority of
*top* human player in par contest, a weak player.
the only conclusion I see is they are lunatics, they don't know what they
are talking about.
no defensive signals and is not programmed to have some very basic
interference of previous carding, (like if partner wins A, he denies K) .
Gib frequently makes mistakes in the final 5 card setups when declaring. I
don't really think you understand how gib bids. It's just based on more than
1 thousand rules with some simulations. When many rules can be applied in a
situation, those rules may conflict with each other and: you may see
something like gib first bids 1D later bids 4C to show a strong rebiddable
club suit with 3 or less D. Also, gib's bidding is so bad sometimes. For
example: with AKQJxx - AKQJx AK, it just bids 4S over opp's 3H opening
because it's based on so called "total points", so without 36 total points,
it would never bid 7S by itself.
heard on this board.
【在 p***r 的大作中提到】
: I have no idea what you are talking about.
: It's commonly known that Gib is a weak declarer,
: defender and bidder. I played way more hands with gib than anybody in this
: board. Gib's bidding system is very very flawed with many bugs. Gib plays no defensive signals and is not programmed to have some very basic interference of previous carding, (like if partner wins A, he denies K) . Gib frequently makes mistakes in the final 5 card setups when declaring. I don't really think you understand how gib bids. It's just based on more than 1 thousand rules with some simulations. When many rules can be applied in a situation, those rules may conflict with each other and you may see something like gib first bids 1D later bids 4C to show a strong rebiddable club suit with 3 or less D. Also, gib's bidding is so bad sometimes. For example: with AKQJxx - AKQJx AK, it just bids 4S over opp's 3H opening because it's based on so called "total points", so without 36 total points, it would never bid 7S by itself.
: Good bidding judgment for gib? This is the most ridiculous claim I ever heard on this board.
:
: than
: when
| a****s
发帖数: 524 | 25 C flight player? either you know nothing about GIB or the meaning of joke
has to be redefined.
If GIB sucks, the result of money bridge would have been more random, But
there's clear evidence that good players win constantly in those games.
Good players don't like volatility, because they have a higher expectation
to win. But somehow they like to play higher stake money
bridge with GIB.
C
possess
It
【在 b***y 的大作中提到】
: Yes, GIB is better than many "BBO experts", but still weaker than average C
: flight player. "World class" is such a joke.
: Even the dummy play (said to be the strongest skill that a robot can possess
: ), GIB routinely messes up at trick 10 or later. Of course bidding and
: defense is even worse. Just think about the 5D bid that you said any robot
: would do. Compared with 4H, choosing 5D is equivalent as choosing a play
: line that cannot have a single winning case over a better line.
: Of course, my comment is based on the experience with money-bridge GIBs. It
: is possible that these GIBs are not of the best version, it can be a trial
: version which is just buggy.
| a****s
发帖数: 524 | 26 yeah, best of best.
club lead knock out your ace, two more round of clubs when they get in
will see you directly home.
many others, novices, beginners, don't know how to play MP, struggling
between 4D or 5D, will get either +150 or +600,
4H got your experts -100.
【在 p***r 的大作中提到】
: Of course, our opinions converge here, 4H is the best MP bid.
:
: Passing
| p***r
发帖数: 20570 | 27 par contest is very different from real bridge situations. Gib is not bad in
par contest simply because in par contest a lot of constraints can be
supplied. In real life, that's totally a different issue. You have to figure
out defender's possible holding all by yourself. Also, in par contest, you
just solve the problem. In real life bridge, your opps intentionally seek
your holes and weakness to take the maximum profit out of them.
Unfortunately, there are just way too many holes in gib's playing logic. For
example, gib never takes into account of playing history, so once it takes
a successful finesse, it wouldn't assume the finesse is successful in the
next round, that may cost gib a lot of imps in real life. However, this kind
holes would never show up in par contests. Also, gib never assumes defender
would rarely lead from A, so it put up K when dummy holds KJx and xx in
hand. So leading from Qxx would work very very well against gib. All these
kind of fundamental problems rarely would show up in par contests. Par
contests are just a unreal subset of real life bridge. In many situations,
par contests simply test the speed of players to construct possible layouts
to solve some certain end games under some certain constraints. Computer is
particular good in this area because of the ability to compute double dummy
many situations. This is in some sense an ability to do well in constructing
and solving double dummy problems, which computer is particularly good at.
However, bridge is not a game of double dummy. And the most complex part in
declaring is to get the right information from opps, eliminate the wrong
information and find the most probable layout and play accordingly. Computer
is particularly bad in this area because this requires very strong AI
designs.
【在 a****s 的大作中提到】
: Sorry, I think anyone think GIB a weak declarer or defender is
: simply lunatics.
: GIB ranked 12th in the par contest in 1998, when the computing power were
: much more moderate compared with today's hardware.
: Sound you know GIB well, You must be familiar with these facts, then I am
: just wondering on what ground one can call a player beat majority of
: *top* human player in par contest, a weak player.
: the only conclusion I see is they are lunatics, they don't know what they
: are talking about.
:
| p***r
发帖数: 20570 | 28 Man, good players play with gib to win money not because of gib's level. No
matter what level gib is, good players have an intrinsic advantage in all
areas to beat the bad players. I play a few two cents games almost every day
and make about $200 every month. I also frequently follow the gib bug
reporting forum and I know how buggy gib is. I have to say nobody in this
board really know gib better than me.
【在 a****s 的大作中提到】
: C flight player? either you know nothing about GIB or the meaning of joke
: has to be redefined.
: If GIB sucks, the result of money bridge would have been more random, But
: there's clear evidence that good players win constantly in those games.
: Good players don't like volatility, because they have a higher expectation
: to win. But somehow they like to play higher stake money
: bridge with GIB.
:
: C
: possess
| p***r
发帖数: 20570 | 29 Can you show what your partner holds?
【在 a****s 的大作中提到】
: yeah, best of best.
: club lead knock out your ace, two more round of clubs when they get in
: will see you directly home.
: many others, novices, beginners, don't know how to play MP, struggling
: between 4D or 5D, will get either +150 or +600,
: 4H got your experts -100.
| b***y
发帖数: 2804 | 30 The point is, 4H isn't final, it offers a choice between 4H and 5D.
I can understand PASS. But if you are going to bid, 5D is just terrible,
unless your partner cannot be trusted to make right decision. (and judging
from the odd 4D bid, that may well be the case).
【在 a****s 的大作中提到】
: yeah, best of best.
: club lead knock out your ace, two more round of clubs when they get in
: will see you directly home.
: many others, novices, beginners, don't know how to play MP, struggling
: between 4D or 5D, will get either +150 or +600,
: 4H got your experts -100.
| | | b***y
发帖数: 2804 | 31 Money bridge doesn't require GIB to be good. The only requirement is that
GIB be consistent, either consistently good, or consistently bad. That will
eliminate the randomness and give good (human) players an edge.
I still think the GIB used in money bridge must be some trial version with
lots of bugs. I've seen the GIB defense going wrong at trick 12, GIB pitches
a winning card and keeps a card in a suit that declarer has shown out. I
think even a person who just learned bridge for 10 minutes could do better
than that. I also do not believe this is the GIB that they actually sent out
for computer bridge tournament.
So it may be that we are talking about different GIBs. One commercial (I
think when it cost $70, it cannot be *that* bad), one free version that no
one is maintaining (and the free one is what gets used on BBO money bridge).
【在 a****s 的大作中提到】
: C flight player? either you know nothing about GIB or the meaning of joke
: has to be redefined.
: If GIB sucks, the result of money bridge would have been more random, But
: there's clear evidence that good players win constantly in those games.
: Good players don't like volatility, because they have a higher expectation
: to win. But somehow they like to play higher stake money
: bridge with GIB.
:
: C
: possess
| a****s
发帖数: 524 | 32 S AQJ
H Ax
D AJxx
C Axxx
what you suggest over partner's 3D?
【在 b***y 的大作中提到】
: The point is, 4H isn't final, it offers a choice between 4H and 5D.
: I can understand PASS. But if you are going to bid, 5D is just terrible,
: unless your partner cannot be trusted to make right decision. (and judging
: from the odd 4D bid, that may well be the case).
| a****s
发帖数: 524 | 33 There had been experiments for robots playing against human players in a
tournament setup.
196 boards, they finished in the middle of a eight-pair field consist of
seven top Dutch pairs. lost by a small margin overall.
The fact is robots have similar performance in tournaments as in the par-contest.
I don't see how your argument can stand against this.
in
figure
you
For
takes
kind
【在 p***r 的大作中提到】
: par contest is very different from real bridge situations. Gib is not bad in
: par contest simply because in par contest a lot of constraints can be
: supplied. In real life, that's totally a different issue. You have to figure
: out defender's possible holding all by yourself. Also, in par contest, you
: just solve the problem. In real life bridge, your opps intentionally seek
: your holes and weakness to take the maximum profit out of them.
: Unfortunately, there are just way too many holes in gib's playing logic. For
: example, gib never takes into account of playing history, so once it takes
: a successful finesse, it wouldn't assume the finesse is successful in the
: next round, that may cost gib a lot of imps in real life. However, this kind
| b***y
发帖数: 2804 | 34 Anything but 4D...
Is partner GIB? If I am playing with GIB, I won't bid 4H over his 4D. GIB
cannot understand it anyway.
【在 a****s 的大作中提到】
: S AQJ
: H Ax
: D AJxx
: C Axxx
: what you suggest over partner's 3D?
| a****s
发帖数: 524 | 35 please explain "anything", why "anything" is better than 4D here.
Now I give two hands, you are not playing with GIB.
Perhaps you can just throw away that 4D card from every bidding boxes you
care to use.
【在 b***y 的大作中提到】
: Anything but 4D...
: Is partner GIB? If I am playing with GIB, I won't bid 4H over his 4D. GIB
: cannot understand it anyway.
| p***r
发帖数: 20570 | 36 If I play support double, I'd just bid 3H to show honor doubleton in H and
doubts in 3NT.
If I don't play support double, I'd bid 3S to show S value here and doubts
in 3NT because of C weakness. Committing to 4D is rather bad here for MP IMO
. Also, 4D should deny CA.
Also, the field is likely to play 3NT in this hand IMO.
For single dummy, 4H is not a bad MP contract at all. When 4H makes, it
usually beats those 5D guys. It requires H 3-3, or DK , DKx on side when H
is 4-2.
Even against those 3NT pairs, 4H would usually go down less.
【在 a****s 的大作中提到】
: S AQJ
: H Ax
: D AJxx
: C Axxx
: what you suggest over partner's 3D?
| p***r
发帖数: 20570 | 37 That's Jack, not gib. Also, human players are not quite familiar with Jack's
game. Also their game is not tuned to play against Jack's weakness. Still,
Jack's game is way stronger than gib.
contest.
【在 a****s 的大作中提到】
: There had been experiments for robots playing against human players in a
: tournament setup.
: 196 boards, they finished in the middle of a eight-pair field consist of
: seven top Dutch pairs. lost by a small margin overall.
: The fact is robots have similar performance in tournaments as in the par-contest.
: I don't see how your argument can stand against this.
:
: in
: figure
: you
| b***y
发帖数: 2804 | 38 Anything beyond 3NT is bad. 3H can show honor doubleton, since the 2NT tends
to deny 3 hearts (otherwise double 2C). 3S would be fine, showing some
doubt about the strain. Even 3NT is acceptable, since that is the most
likely game (and pays extra dividend in MP). I think 3NT is a little
committal though, but 4D is neither here nor there with this hand. In MP you
just have to try hard to keep 3NT as a possibility, you need a strong
reason to go beyond it, this hand just doesn't warrant it.
Of course sometimes bad bidding doesn't cost you, or very infrequently it
can even land you into a superior contract, what is commonly known as 狗屎运
. You can try passing with a 20-hcp hand and find that you are the only one
in the room who doesn't go minus, but that doesn't mean passing is right.
【在 a****s 的大作中提到】
: please explain "anything", why "anything" is better than 4D here.
: Now I give two hands, you are not playing with GIB.
: Perhaps you can just throw away that 4D card from every bidding boxes you
: care to use.
| a****s
发帖数: 524 | 39 Be it Jack or GIB, the point is focus on things that can be quantified give
you result.
's
,
【在 p***r 的大作中提到】
: That's Jack, not gib. Also, human players are not quite familiar with Jack's
: game. Also their game is not tuned to play against Jack's weakness. Still,
: Jack's game is way stronger than gib.
:
: contest.
| a****s
发帖数: 524 | 40 West has
S AQJ
H Ax
D AJxx
C Axxx
So, armed with the most advanced conventions of your choice, incredibly fine
judgement (at least compared with robots, or anyone in sight, for that
matter), and superior knowledge of how match-point game should be played,
you experts are likely to end up with 3NT, a contract that has no play,
after seeing both hands! | | | b***y
发帖数: 2804 | 41 Good players will try for 3NT, but east hand should reject it, and settle
for 4H. On this particular hand, 4D doesn't affect the result, since East
has an obvious 4H bid anyway. But that doesn't mean 4D is right. It is just
a bad bid that doesn't cost this time. | p***r
发帖数: 20570 | 42 I'd play in 4H.
1C x p 1H
2C 2N p 3C (4+D)
p 3D(fit in D) 3H(5+H)
p 3S( S value, doubts in 3NT, concerns in C) p 4H(choice of games)
all pass.
I think 5D is only slightly better than 4H in MP (because the overall chance
to make 5D is slightly higher than 4H) and 4H can often allow you to win a
top when it makes.
Here, 4D is simply a rather bad bid because you bypass 3NT. What if your
partner holds something like xx QTxx QTxxx Kx? You really think you want to
play in 5D here in MP? 3NT here would always beat 5D no matter what.
fine
【在 a****s 的大作中提到】
: West has
: S AQJ
: H Ax
: D AJxx
: C Axxx
: So, armed with the most advanced conventions of your choice, incredibly fine
: judgement (at least compared with robots, or anyone in sight, for that
: matter), and superior knowledge of how match-point game should be played,
: you experts are likely to end up with 3NT, a contract that has no play,
: after seeing both hands!
| p***r
发帖数: 20570 | 43 I don't know what you are talking about. You think my sequence can't
quantify the total strength and gib's sequence does? This is getting
ridiculous. Do you really know that there are just many systemic flaws in
gib's system that doesn't "quantify" anything? Gib's system is a very buggy
one, it's really a huge surprise to see you claim it's a good one.
give
【在 a****s 的大作中提到】
: West has
: S AQJ
: H Ax
: D AJxx
: C Axxx
: So, armed with the most advanced conventions of your choice, incredibly fine
: judgement (at least compared with robots, or anyone in sight, for that
: matter), and superior knowledge of how match-point game should be played,
: you experts are likely to end up with 3NT, a contract that has no play,
: after seeing both hands!
| a****s
发帖数: 524 | 44 Brilliant Skill.
Bid that leads to a lay-down game on a known 4-4+ fit, is bad bid but good
luck.
Bid a failing game on a known 5-2 fit. is good bid and bad luck.
I used to think all things have their limit, but I was wrong. obviously BS
doesn't have one. BTW, BS means Brilliant Skill.
settle
just
【在 b***y 的大作中提到】
: Good players will try for 3NT, but east hand should reject it, and settle
: for 4H. On this particular hand, 4D doesn't affect the result, since East
: has an obvious 4H bid anyway. But that doesn't mean 4D is right. It is just
: a bad bid that doesn't cost this time.
| a****s
发帖数: 524 | 45
chance
a
The only thing you have confirmed here is you are no better a bidder than
GIB. Holding xx QTxx QTxxx Kx, after
1C x p 1H
2C 2N p ?
you suggest 3D could be a plausible choice. for what? missing grand in
diamonds? Even the dumbest robot would have bid 3NT with that in every
single *IMP* game.
Robots have bugs, just like people make mistakes. But the difference is that
robots have no ego, therefore bugs are easily fixed.
【在 p***r 的大作中提到】
: I'd play in 4H.
: 1C x p 1H
: 2C 2N p 3C (4+D)
: p 3D(fit in D) 3H(5+H)
: p 3S( S value, doubts in 3NT, concerns in C) p 4H(choice of games)
: all pass.
: I think 5D is only slightly better than 4H in MP (because the overall chance
: to make 5D is slightly higher than 4H) and 4H can often allow you to win a
: top when it makes.
: Here, 4D is simply a rather bad bid because you bypass 3NT. What if your
| p***r
发帖数: 20570 | 46 You bid 3NT over 2NT and found partner happen to hold Qx AKx AKxxx ATx, or
Ax AKJ KJxxx ATx.
Everybody plays in their 10 card fit in D making 5 or 6, you go down in a no
play 3NT after an obvious S lead.
Is there anything related to this bidding failure? Poor bidding judgment?
It's not even about judgment at all. A basic bidding principle is to bid
your hand naturally and don't miss any 10 card fit games.
All those who only think judgment is the most important thing in bridge
bidding are my best clients in this money game.
【在 a****s 的大作中提到】
:
: chance
: a
: The only thing you have confirmed here is you are no better a bidder than
: GIB. Holding xx QTxx QTxxx Kx, after
: 1C x p 1H
: 2C 2N p ?
: you suggest 3D could be a plausible choice. for what? missing grand in
: diamonds? Even the dumbest robot would have bid 3NT with that in every
: single *IMP* game.
| b***y
发帖数: 2804 | 47 This is beyond bridge discussion. No need to get all worked up.
If the result of this single board leads you to believe that 4D then 5D is
the right bids in all similar situations, go ahead with your quantifying
bids, no one is stopping you.
But if you still want to have meaningful technical discussion, at least you
can try to use more accurate words. For example 5D isn't really a lay-down,
and 4H isn't hopeless. And if either hand has H10, 4H will be favorite, but
I know of no bidding system that can quantify trump 10, not even the robot
system.
【在 a****s 的大作中提到】
: Brilliant Skill.
: Bid that leads to a lay-down game on a known 4-4+ fit, is bad bid but good
: luck.
: Bid a failing game on a known 5-2 fit. is good bid and bad luck.
: I used to think all things have their limit, but I was wrong. obviously BS
: doesn't have one. BTW, BS means Brilliant Skill.
:
: settle
: just
| a****s
发帖数: 524 | 48 I love this, just keep trying.
They have entered the bidding, club twice, and managed to miss the master
suit, spades, NINE of them. Not to mention with 3 good hearts and
a weak side suit, partner may have cued show doubt of strain.
Then comes the fun part, when defending 3NT, suddenly somehow Spade becomes
the "obvious" lead.
神出鬼没啊。
What puzzled me more is, with singleton in enemy's suit, you scoff at the
idea of 5D, but with Kx doublton, suddenly it becomes a smart thing to do.
LOL.
Oh, I see, because you can see all the cards!
I agree, you can beat anyone, who doesn't have the magic ball.
no
【在 p***r 的大作中提到】
: You bid 3NT over 2NT and found partner happen to hold Qx AKx AKxxx ATx, or
: Ax AKJ KJxxx ATx.
: Everybody plays in their 10 card fit in D making 5 or 6, you go down in a no
: play 3NT after an obvious S lead.
: Is there anything related to this bidding failure? Poor bidding judgment?
: It's not even about judgment at all. A basic bidding principle is to bid
: your hand naturally and don't miss any 10 card fit games.
: All those who only think judgment is the most important thing in bridge
: bidding are my best clients in this money game.
| w****b
发帖数: 623 | 49 folks, calm down. Having differences in opinion shouldn't be the end of the
world, not even the friendships.
This sequence is somewhat analogus to the following (although you may
disagree) with no interference:
1NT 2D/H
2H/S 3m
in which responder showed 2 suiter, a major and a minor. This is gf, but I
don't think there is a general consensus on slam interest -- some may just
bidding this with 11HCP and 5-4.
In such situations, unless I have special agreement with pd about the
serious slam interest, the structure I prefer to play is a 4m bid, which
passes 3NT, promises both m and M support. This way, you can always stop at
the 4M. If you have just minor support, bid the 4oM to show support. And if
you really really need to surpass 3NT (like a 5 or 6 bagger in that minor
with primary values in other suits, you still have a chance to do it later).
Here, as far as I'm concerned, all new suits are forcing (since we don't
play xfer here I assume), so we are in similar ground: responder showed 2
suits, and 3 strains are possible: Major, minor games or 3NT. In fact there
are even more reason to play what I suggested above, since here, responder
did not necessarily suggest a 5 card M suit, so it'll be perfect to have
doubler show 3M and 4+m support by surpassing 3NT directly.
Now let's say, we know doubler cannot have 3card H support, I'm afraid I'll
side with arrows. Since you have rather limited strength, and quite marginal
values in game, in MP, making the game would be huge already and there won'
t be a majority to play in 5-2 major fit. And even if pd does have Xx in H,
with correct defense, you often find that you need H3-3, and this is not to
count the possibility of pd having 5+D. | w****b
发帖数: 623 | 50 Now back to the GIB issue. I'm a programmer so I hope I can make some
comments.
A program like that usually requires two things: the engine, and data. You
need both to be competitive. If the engine is bad, no perfect data can save
you. If the engine is good, but the data is insufficient, at least you have
a future.
GIB's engine is not based on exhaustive searches or complicated branching/
backtrack algorithms. It's based on statistics. It generates a number of
hands that fit the criteria of the info so far, and run simulations. This is
the debatable point.
First, due to speed constraints, in all the GIBs you guys played against, it
probably runs too less simulations (10 hands?), this makes the result quite
random at times, and sometimes laughable. But this isn't a fundamental flaw
.
Sometimes you can attribute the flaw to the constraints it currently has.
But again I don't think that is a fundamental flaw either. It's just
insufficient insight and data.
I'm not quite sure if GIB always runs simulation, or only when it comes to
no man's land. My thinking process, in the bidding, at least in early stages
, I don't need any simulation and construction of pd's hands to make
decisions. If GIB just uses this to overcome the insufficient data, then
that should be fine.
The problem with GIB is it does not have a "learning mode". Relying on
humans to enter the rules will always be restricted since the guy who knows
how to enter rules may not (and likely is not) a top guy. So the constraints
will always be flawed. It needs to develop an engine to figure out rules
based on a set of opponents (obviously you need to turn on and off the mode
so you don't learn from random players). In other words, it has the random
hands -> decision process, but not the decision -> rule feedback. | | | m****r
发帖数: 6639 | 51 ginsberg himself claims to be good enough to play fro england.
he said that to me.
save
have
is
【在 w****b 的大作中提到】
: Now back to the GIB issue. I'm a programmer so I hope I can make some
: comments.
: A program like that usually requires two things: the engine, and data. You
: need both to be competitive. If the engine is bad, no perfect data can save
: you. If the engine is good, but the data is insufficient, at least you have
: a future.
: GIB's engine is not based on exhaustive searches or complicated branching/
: backtrack algorithms. It's based on statistics. It generates a number of
: hands that fit the criteria of the info so far, and run simulations. This is
: the debatable point.
| p***r
发帖数: 20570 | 52 4H is actually way better than a H 3-3 fit.
Suppose you win CA, play DJ which is likely to force out DK. then you ruff C
return, play HQ let it run, assume it works, then you take S finesse, cash
HA, play D to take another S finesse, your chance is indeed pretty good,
because in this bidding sequence, declarer is likely to hold most of the
HCPs when his partner passed the takeout double and declarer is aceless.
Also, if DJ is ducked, you can just play DA and play the 3rd D to force it
out. Later, as long as you take the H finesse and S finesse, you are still a
favorite to make 4H.
Also, if you make 4H, the payoff can be pretty good.
I have been trying to be nice here.
the
【在 w****b 的大作中提到】
: folks, calm down. Having differences in opinion shouldn't be the end of the
: world, not even the friendships.
: This sequence is somewhat analogus to the following (although you may
: disagree) with no interference:
: 1NT 2D/H
: 2H/S 3m
: in which responder showed 2 suiter, a major and a minor. This is gf, but I
: don't think there is a general consensus on slam interest -- some may just
: bidding this with 11HCP and 5-4.
: In such situations, unless I have special agreement with pd about the
| p***r
发帖数: 20570 | 53 There are some fundamental problems in gib bidding. The largest problem is
not its system, its simulation, but it's hand evaluation. Gib's hand
evaluation is based on total points. In trumps or no trumps, it always
assumes that with 25 total points, it would force to game, with 31 total
points, it would force to slams (based on double dummy analysis, 31 total
points can often be enough, however, we all know 33 HCP is usually good).
That's why you see gib sometimes becomes incredibly crazy when holding
shapely hands even bidding no trumps. This fundamental flaw of gib bidding
would bring it no where close to expert level, no matter how good the rules
are or how good the simulations are.
I have studied gib's system for so many years, made constant contributions
to gib bug reporting forum and made a lot of testings.
In my opinion, a sound hand evaluation scheme should be a hybrid of high
card points, total tricks, total losers, controls, trump qualities, side
suit qualities and how well the hands fit with each other. This requires a
lot of work and sometimes also involves good AI designs. For example a good
database on your opps' long term and short term behavors in certain
situations, which is very computing intensive. Therefore, in the near future
, computer's bidding is quite hopeless to reach the level of human experts.
save
have
is
【在 w****b 的大作中提到】
: Now back to the GIB issue. I'm a programmer so I hope I can make some
: comments.
: A program like that usually requires two things: the engine, and data. You
: need both to be competitive. If the engine is bad, no perfect data can save
: you. If the engine is good, but the data is insufficient, at least you have
: a future.
: GIB's engine is not based on exhaustive searches or complicated branching/
: backtrack algorithms. It's based on statistics. It generates a number of
: hands that fit the criteria of the info so far, and run simulations. This is
: the debatable point.
| b***y
发帖数: 2804 | 54 Ginsberg probably hasn't touched GIB for long time.
We also have people in this forum claiming good enough to be world champions
, so what? :-) Ginsberg is not even British...
【在 m****r 的大作中提到】
: ginsberg himself claims to be good enough to play fro england.
: he said that to me.
:
: save
: have
: is
| o*******n
发帖数: 6500 | 55 谁说自己和世界冠军一个水平的？
champions
【在 b***y 的大作中提到】
: Ginsberg probably hasn't touched GIB for long time.
: We also have people in this forum claiming good enough to be world champions
: , so what? :-) Ginsberg is not even British...
| j*******e
发帖数: 2168 | 56 nod. I think "machine learning" is the key to make the software better.
save
have
is
【在 w****b 的大作中提到】
: Now back to the GIB issue. I'm a programmer so I hope I can make some
: comments.
: A program like that usually requires two things: the engine, and data. You
: need both to be competitive. If the engine is bad, no perfect data can save
: you. If the engine is good, but the data is insufficient, at least you have
: a future.
: GIB's engine is not based on exhaustive searches or complicated branching/
: backtrack algorithms. It's based on statistics. It generates a number of
: hands that fit the criteria of the info so far, and run simulations. This is
: the debatable point.
| a****s
发帖数: 524 | 57 Perhaps he is a Scot?
champions
【在 b***y 的大作中提到】
: Ginsberg probably hasn't touched GIB for long time.
: We also have people in this forum claiming good enough to be world champions
: , so what? :-) Ginsberg is not even British...
| b***y
发帖数: 2804 | 58 He is an American professor in computer science, but he graduated from
Oxford. He has abandoned GIB long ago.
【在 a****s 的大作中提到】
: Perhaps he is a Scot?
:
: champions
| m****r
发帖数: 6639 | 59 he may have abandoned GIB, but as far as i can understand from what people
said on this board, the basic concept has not changed at all.
【在 b***y 的大作中提到】
: He is an American professor in computer science, but he graduated from
: Oxford. He has abandoned GIB long ago.
| b***y
发帖数: 2804 | 60 Of course the basic concept is the same, since nothing further has been done
...
10 years ago it may beat other computer programs, but not any more.
【在 m****r 的大作中提到】
: he may have abandoned GIB, but as far as i can understand from what people
: said on this board, the basic concept has not changed at all.
| | | a****s
发帖数: 524 | 61 本来不是说这手牌的。原来的意思是同伴的5个点加单张够叫5D了。“数点都不会么？
GIB都会的”。4H不是本手，有些追求顶分的味道。顶分是可遇而不可求的。做人要正
直，人品好，自然会有人来送礼的。
对于这手牌，4D即使不是世界上最正直的叫品，起码也是最正直的之一。Ace多（已经
不能再多了），位置好，等等等等，就不必罗嗦了。
冒充一把内行，说说我认为高手是怎么思考的。且不管同伴出两套是基于牌力还是牌型
，参考叫牌过程，如果同伴有牌型，必定是草花单缺。为什么呢？
“黑桃套哪去啦？”如果同伴黑桃短，敌人就有9张黑桃，叫牌过程还会是这样的么？
认识到这一点，你就知道无将定约是没法打的。这样的推理，对高手来说，应该是自动
的。
接下来算算赢墩，假定同伴是3=5=4=1的白皮，这手牌打方块定约也几乎一定能取得9墩
。基于这个考虑，朴素地4D建议10墩牌，叫的是同伴牌力的期望值。
没有约定，也没有啥逼叫不逼叫的。简单的牌力估价和逻辑推理。 | b***y
发帖数: 2804 | 62 我真的糊涂了。莫非发贴的目的不是诚心问问题，而是心中已经有现成的答案，只是故
意考考大家来的？ | m****r
发帖数: 6639 | 63 what do you mean "nothing further has been done"?
combining "ginsberg has abandoned it" and the fact that there is a bug
tracking system, it sounds like alot is been done since he left the project.
done
【在 b***y 的大作中提到】
: Of course the basic concept is the same, since nothing further has been done
: ...
: 10 years ago it may beat other computer programs, but not any more.
| b***y
发帖数: 2804 | 64 I happen to know the person who tracks the bug reports. It is mainly in
product maintaining mode, not developing mode. Feature-wise, nothing new.
【在 m****r 的大作中提到】
: what do you mean "nothing further has been done"?
: combining "ginsberg has abandoned it" and the fact that there is a bug
: tracking system, it sounds like alot is been done since he left the project.
:
: done
| m****r
发帖数: 6639 | 65 ic. so, in that sense, my initial statement about the "author" is good
enough (at least he claims to be) is still valid, although he's not being
around.
【在 b***y 的大作中提到】
: I happen to know the person who tracks the bug reports. It is mainly in
: product maintaining mode, not developing mode. Feature-wise, nothing new.
| w****b
发帖数: 623 | 66 In my humble opinion, both system and hand evaluation fall into the "rules
and constraint" part, not necessarily engine. A better hand evaluation is
simply some additional rules. I'm not totally sure if GIB allows such rules
to be expressed and inserted without changing the engine. I'd imagine it
should. If this is the case, then the engine is fine but needs maintenance.
If not, then this is pretty a hopeless project.
Even to human players, hand evaluation is a deep and evolving subject.
Personally in my earlier years of bridge, I simply used abstrated rules to
perform hand evaluation. But later I started to construct sample hands and
simulate the play. That is good news and bad news. It helped me to land into
more precise contracts often, but it also caused some rather idiotic
incidence. So in a way, I do think like GIB, and I cannot say if GIB would
be able to achieve worldclass level if it's pushed to its perfection, but
not a problem to beat me -- here I'm only talking about bidding, not dummy
play or defense.
rules
【在 p***r 的大作中提到】
: There are some fundamental problems in gib bidding. The largest problem is
: not its system, its simulation, but it's hand evaluation. Gib's hand
: evaluation is based on total points. In trumps or no trumps, it always
: assumes that with 25 total points, it would force to game, with 31 total
: points, it would force to slams (based on double dummy analysis, 31 total
: points can often be enough, however, we all know 33 HCP is usually good).
: That's why you see gib sometimes becomes incredibly crazy when holding
: shapely hands even bidding no trumps. This fundamental flaw of gib bidding
: would bring it no where close to expert level, no matter how good the rules
: are or how good the simulations are.
| o*******n
发帖数: 6500 | 67 我觉得不是什么考大家吧
而是印证想法
这应该是正常的吧
对于一手牌，我们总会有自己的看法(我的理解就是所说的现成的答案)
但是想听听别人的意见
别人理由充分，能够说服我改变想法，对我来说总是好的，我学习到了新东西嘛
别人理由不充分，被我说服了，也是可能的嘛
主要是不要吵起来，那就得不偿失了 :-)
【在 b***y 的大作中提到】
: 我真的糊涂了。莫非发贴的目的不是诚心问问题，而是心中已经有现成的答案，只是故
: 意考考大家来的？
| b***y
发帖数: 2804 | 68 如果别人不同意就是BS，那就没有“印证”一说了吧，干脆公布标准答案就是了。
【在 o*******n 的大作中提到】
: 我觉得不是什么考大家吧
: 而是印证想法
: 这应该是正常的吧
: 对于一手牌，我们总会有自己的看法(我的理解就是所说的现成的答案)
: 但是想听听别人的意见
: 别人理由充分，能够说服我改变想法，对我来说总是好的，我学习到了新东西嘛
: 别人理由不充分，被我说服了，也是可能的嘛
: 主要是不要吵起来，那就得不偿失了 :-)
| b***y
发帖数: 2804 | 69 All we know is that he wrote the software before, he no longer maintains it.
This has nothing to do with him being good enough or not. Your statement is
neither valid nor invalid, it is irrelevant. :-)
【在 m****r 的大作中提到】
: ic. so, in that sense, my initial statement about the "author" is good
: enough (at least he claims to be) is still valid, although he's not being
: around.
| b***y
发帖数: 2804 | 70 Actually, I suspect that GIB's bidding is mainly sample based. Sometimes you
see very strange thing, for example after you open 1NT, GIB would transfer
to 2H, but then you find it has a balanced hand with only 4 hearts.
In general, there should be initial rules to follow, for example if you have
13 HPC with 5 spades, you should open 1S. There is no need to construct
sample hands to decide what to open, it is all part of the bidding system
and you have to trust that the system is designed well. Problems arise when
rules are conflicting. For example you hold 1363 shape and 19 HCP, you open
1D partner responds 1S. Now the bidding system doesn't cover this situation,
all possible bids (2H/2NT/3C/3D) are flawed. Now constructing sample hands
would be useful. But if the robot doesn't even know which possible bids to
choose from, then having sample hands may not help much, it may instead
decide to bid something totally off the chart.
Using sample construction as hand evaluation method, that is not something
new. Virtually all experts do that. It is call "elastic evaluation". But
this must be done in a well-constructed framework in first place, GIB may
have serious flaws in that area.
【在 w****b 的大作中提到】
: In my humble opinion, both system and hand evaluation fall into the "rules
: and constraint" part, not necessarily engine. A better hand evaluation is
: simply some additional rules. I'm not totally sure if GIB allows such rules
: to be expressed and inserted without changing the engine. I'd imagine it
: should. If this is the case, then the engine is fine but needs maintenance.
: If not, then this is pretty a hopeless project.
: Even to human players, hand evaluation is a deep and evolving subject.
: Personally in my earlier years of bridge, I simply used abstrated rules to
: perform hand evaluation. But later I started to construct sample hands and
: simulate the play. That is good news and bad news. It helped me to land into
| | | m****r
发帖数: 6639 | 71 i made my statement in response to wimp's post about "the guy who enters
rules maybe not be a top guy".
it.
is
【在 b***y 的大作中提到】
: All we know is that he wrote the software before, he no longer maintains it.
: This has nothing to do with him being good enough or not. Your statement is
: neither valid nor invalid, it is irrelevant. :-)
| b***y
发帖数: 2804 | 72 Well, wimp isn't wrong: maybe he's not a top guy (but maybe he is). We also
know that he claims he is good, but again, maybe he is, maybe he isn't. Just
like anyone can claim to be good enough for world class, but maybe yes,
maybe no. I don't want to be judgmental.
【在 m****r 的大作中提到】
: i made my statement in response to wimp's post about "the guy who enters
: rules maybe not be a top guy".
:
: it.
: is
| p***r
发帖数: 20570 | 73 Let me set it straight. Most constraints set by gib are based on total
points. A certain bid usually means a certain amount of total points and
some shape and suit quality information. Then the rules are just based on
total points. Here comes the key problem, both no trumps and trumps are
based on the same total points system. The total points are based on HCP and
distributions. It is simply absurd to count a void as a 3 total point asset
in no trumps. However, gib does that in every hand. So as long as it counts
25 combined total points in two hands, it forces to game. From this bad
starting point, it is simply absurd to claim that gib's bidding is anything
close to an advanced bridge player's bidding, not even experts.
Of course gib's bidding can be improved. However, that requires important
implementation of total tricks, total losers and more. Without these basic
concepts, gib is simply hopeless to bid correctly in many situations. For
example, in gib's RKC sequence, 5NT is for K asking. However, gib would
never know when to bid 7 even if 13 tricks are in hand because its rules are
written in terms of total points. Simulations in theory should work here.
However, it still has many bugs in its simulation functions and it has a
hard time to generate good samples when the hands are very strong. Also,
simulation is not a sure thing. Most bidding decisions are made based on 10-
20 hands simulations. Also, there are many bugs and problems in the
constraint set up in simulations.
As I said, I have been trying to be nice in this whole thread. However,
arrows' ignorance on gib system and stubbornness are really something unseen
before.
rules
.
into
【在 w****b 的大作中提到】
: In my humble opinion, both system and hand evaluation fall into the "rules
: and constraint" part, not necessarily engine. A better hand evaluation is
: simply some additional rules. I'm not totally sure if GIB allows such rules
: to be expressed and inserted without changing the engine. I'd imagine it
: should. If this is the case, then the engine is fine but needs maintenance.
: If not, then this is pretty a hopeless project.
: Even to human players, hand evaluation is a deep and evolving subject.
: Personally in my earlier years of bridge, I simply used abstrated rules to
: perform hand evaluation. But later I started to construct sample hands and
: simulate the play. That is good news and bad news. It helped me to land into
| p*********6
发帖数: 679 | 74 The following doc might be interesting to some people here - shown in BBO
news at login today.
http://doc.bridgebase.com/lobbynews/gib_descriptions.html | b***y
发帖数: 2804 | 75 The concept is nice:
"GIB starts by finding the matching bid in its bidding rules (we call this
the "book bid"). If simulations are allowed, it then makes some adjustments
to its hand (adding a card to each suit, adding/subtracting a few total
points) and finds the book bids for those similar hands."
This is actually what human experts do as well. First you narrow down
possible bids allowed by the system, when there are more than 1, then you do
"elastic evaluation", which basically constructs worst/average/ best hands
that partner can have with the bidding so far, then use that to guide the
bidding decision.
But in reality, GIB sometimes chooses a "non-book" bid, which suggests
either a bug in the program or ignorance of the programmers.
【在 p*********6 的大作中提到】
: The following doc might be interesting to some people here - shown in BBO
: news at login today.
: http://doc.bridgebase.com/lobbynews/gib_descriptions.html
| p***r
发帖数: 20570 | 76 There are still many holes in the rules. Anyway, they are improving them.
adjustments
do
hands
【在 b***y 的大作中提到】
: The concept is nice:
: "GIB starts by finding the matching bid in its bidding rules (we call this
: the "book bid"). If simulations are allowed, it then makes some adjustments
: to its hand (adding a card to each suit, adding/subtracting a few total
: points) and finds the book bids for those similar hands."
: This is actually what human experts do as well. First you narrow down
: possible bids allowed by the system, when there are more than 1, then you do
: "elastic evaluation", which basically constructs worst/average/ best hands
: that partner can have with the bidding so far, then use that to guide the
: bidding decision.
| p*********6
发帖数: 679 | 77 On BBO news - A new version of GIB (Version 20) released on 12/15.
Main Changes:
-Added/improved many rules for takeout doubles and continuations.
-Improved GIB's ability to choose between minor or major suits after a -2
level opening or overcall, including unusual NT.
-Improved rules on penalty and support redoubles.
-After partner gives a limit raise, GIB will now refrain from bidding to an
illogical level.
-Made GIB smarter after the 1NT - Stayman - 2 Major - 3 Other Major sequence
. GIB should now differentiate better between game and slam interest hand
types.
-Improved GIB's weak 2 opening bids.
-GIB will now bid unusual 1NT more often.
-GIB will now bid according to the LAW (of total tricks, by LC) more often.
-Some alerts were buggy. For instance, sometimes gives wrong explanation of
suit lengths. These are now fixed.
-Added some rules to help GIB bid when his opponent overcall competitively.
-Fixed a bug where GIB was not recalculating points after a passed hand
doubles, then bids a suit. GIB now knows about this trick, so watch out
tourney players!
-Improved the Lebensohl 2NT sequence.
-Fixed a bug where after 1 Major - 1NT - 2 Major, if a takeout X was made,
GIB raised with an inappropriate hand.
-After a 1NT rebid by partner, if GIB holds 7+ in major and max invite
values (ie, unbalanced hand), it will now bid game instead of only invite.
-Adjusted GIB's hand evaluation and bidding after a strong 3NT opening.
-Adjusted GIB's hand evaluation for quantitative 4NT invite.
Improved 1NT - Minor stayman points/hand evaluation and continuations.
-Added new rules to handle minor suit opening followed by a reverse. GIB now
uses Lebensohl to show a weak responding hand.
-Improved the Fourth suit forcing auction rules.
-Improved GIB's cuebidding logic.
-Improved GIB's 1NT - 3NT raise bidding logic.
-Improved GIB's Cappelletti bidding logic.
-Some other bug fixes and tweaks. |
|
