1NT with 20 HCP
jogs
Finding pairs declaring in 1NT with exactly 20 HCP has not been
easy. Lucky to find 15-20 in 1NT with 20 HCP every 100 boards
searched.
After finding over 40 observations, the declarers have averaged about
7.2 tricks per play at 1NT. On some boards 1NT was declared by both
sides. That's about 2/3 of a trick above the expected 6.5 tricks.
Defending is harder for most players than declaring.
Andrew
Richard Pavlicek made an interesting post a few years ago where he
analyzed several thousand hands played at his table on OKB. He counted
both the double dummy tricks in each contract and the actual tricks
taken. He then looked at the difference by contract. 1NT was the
contract where declarer had the largest advantage versus double dummy.
The chart showed that:
* The declarer's advantage (over double dummy) was larger in NT than
in suits
* The declarer's advantage was larger at lower levels
Andrew
jogs
When both sides hold exactly 20 HCP, there's
lots of options to go wrong. Not easy, even for
experts to defend as a unit.
Hope to find 200+ obs within a few days. Maybe
the mean tricks for declarer will approach 7.
Qxxx(x)(x) and king offside occurs more often
when declarer holds the balance of points.
Maybe one day someone will be able to develop
a true single dummy program. A program which
plays the dummy statistically perfectly.
Dave Flower
>
> - Show quoted text -
If anyone does an analysis, I suggest that the number of bids be taken
into account. It is harder to defend 1NT all pass than 1D p 1S p 1NT
all pass
Dave Flower
Nick France
After Reading Dave's post you may also have to take into consideration
type of hands that the opponents will interfere over a 1NT opening.
These hands might be included in that first 6.5 trick estimate and,
without proof, I'd assume if they were excluded from expected tricks
that expect tricks would go up.
Nick France
KWSchneider
>
> - Show quoted text -
I keep saying that, properly configured, GIB plays the dummy
statistically perfectly. I've proven this in a number of posts but you
seem to ignore this.
If you want to stipulate conditions, I'll set up a 1000 hand
sumulation. You will need to stipulate the bidding as well.
1000 hands will take a day to run.
Kurt
jogs
Show the link. If GIB plays statistically perfectly,
it requires no simulation.
No bidding. Much too complex. Waiting for someone
to write a program which makes the best percentage
play against the entire vector space.
Fred.
Playing a weak no-trump, it's not all that unusual to find myself in
1NT with 17-19 HCP and I often find myself finishing the hand
wondering where all the tricks came from. On some of the flat broken
hands I seem to pick up a pattern where LHO pretty much has to give us
a trick on the opening lead, and at the same time creates a suit in
which I can exit withouthout giving up a trick. And, then, they have
to lead again.
Fred.
KWSchneider
>
> - Show quoted text -
I've had challenges issued on this forum and GIB passed them. If you
wish to provide a specific hand or hands to come up with the best SD
play, please do so. I've yet to have a hand given to GIB that failed
to find the "statistically correct play". I'm sure that there may be
one or two - but haven't found them yet. And if this is so tough to
find, it suggests that the vast majority of the deals are optimally
played.
Simulation is the right word - GIB will be required to play a series
of hands that are computer dealt to meet your simulation criteria.
Playing one hand is not relevant to answer the question you have
posed. Obviously, GIB will have to play defense as well, unless you
wish to play in real time.
Bidding will help all the players. So a 1m-1M-1N AP auction is more
descriptive than a 1N AP auction. Also, you probably will want to also
divide the 20points in buckets between declarer and dummy. Something
like 12-8, 14-6, 16-4. Obviously, playing 2/1 or SA, the first two
splits require an auction to get to 1N, so we probably need to
restrict shape for dummy as well. And finally, we need to restrict the
defense hands to those that will not be bidding at the 1level, nor
have an automatic double of 1N.
Kurt
Nick France
>
> - Show quoted text -
You are the only one that believes GIB or any computer plays a hand
statistically perfect. If this were true then people would rely on
computers to solve how to play a hand and that just isnt so.
Nick France
jogs
Here's a hand straight out of a book.
Contract is 3NT. No opposition bidding.
S 73
H K8
D Q82
C KJ6542
S AKQ6
H A7
D AJ965
C Q8
Opening lead is small heart.
How would you play?
Humans are capable of thinking and changing
lines depending on how defenders play.
danc...@yahoo.com
>
> - Show quoted text -
Is this MPs or IMPs?
If it is IMPs I would play on diamonds. I would play on clubs at MPs
since more tricks available if the suit comes in.
jogs
Always IMPs. If I forget to state, it's IMPs.
KWSchneider
I'll run GIB based on
1) Assume South is declarer.
2) Assume bidding is 1D 2C 3N...
3) small heart lead [from 4+ hearts]
4) IMPS
GIB will change the play based on the play and holding of the
opponents as well, but obviously needs to have defending hands
simulated [consistent with the lead, carding and lack of bidding of
defenders] and the defenders use the bidding of the declarers. Or
should I report back on each decision? Which means that you would need
to feed me each play [kinda like oldtime computer chess]. Then I
wouldn't need to simulate the other hands.
After I report back on this, I'd like to the actual hands and
defensive plays to compare with how GIB would have handled the EXACT
hands.
Kurt
Fred.
>
> - Show quoted text -
If you really want to go for broke at match points, take the heart in
hand, lead toward the diamond queen, and if the queen holds, switch to
clubs.
Fred.
Fred.
>
> - Show quoted text -
I'm starting to think you mean something different by "perfect" than
the rest of us. While I agree that perfection, at least in the
limited sense of formal correctness, can exist, I don't believe that
it can be demonstrated by testing.
If you were to test the program will 1000 hands, and amd get an expert
panel to agree that the program took the best possible line on each
hand, that will be a very impressive accomplishment (actually, just
getting the panel to agree on 1000 hands one way or the other would be
an impressive accomplishment). However, there are two very real
possibilities. One is that everyone on the panel fell into the same
mental trap in evaluating one of the hands. The other possibility is
that there is a problem which will only be disclosed on hand 1001, or
perhaps 10001.
Keep in mind that a program with only 10 if statements has slightly
more than 1000 distinct paths though the code, and, thus, require in
excess of 1000 cases for exhaustive testing.
And, without maligning the program in any way, there are good reasons
to believe that the program in question could not be formally
correct. Programs which deal with complex issues, in a manner similar
to the people who deal with those issues need to find means of
reducing the number of possibilities to consider. In the case of this
program I suspect it uses Monte Carlo (simulation) methods to
_estimate_ probablities. While these estimates have a high
probability of being close to the actual probabilities they have some
chance of being markedly off and suggesting the wrong course.
Fred..
jogs
Humans are capable of trying lines against
the entire vector space. Often only necessary
to test against 3 or 4 distributions. The trick
is humans can determine which are the critical
distributions.
KWSchneider
>
> - Show quoted text -
Perfect to me is the play that has the most chances to achieve the
required result - in IMPS, namely the contract. No more, no less.
There is one play [there may be a series of equal plays in simple
contracts] that provides the "best" line based on the outstanding
cards and frequency-weighted distributions.
I suppose that I should go back and revisit all of Pavlichek's
challenges and see how many that GIB gets right. Having said that,
perhaps I should throw a couple of Kantar's challenges from the ACBL
Bulletin at it.
Kurt
Gary Logan
Win in hand and play ace of diamonds.
If all follow, a low diamond to the queen ensures nine tricks.
If rho shows out on second diamond, switch to clubs to
ensure at least two tricks in each minor.
Fred.
By the way, I don't disbelive that GIB can make a perfect play in the
sense you give, or even that it may do so quite consistently.
However, to me, to be a perfect system, it needs to make the pefect
play every time. And for me to believe something is a perfect system,
it not only has to have made perfect plays to date, but there has to
be good reason for me to believe that it will never fail in the
future.
Fred.
KWSchneider
>
> - Show quoted text -
But the purpose of the "experiment" is to determine over the course of
"many" hands [1000?] whether 20point combined HCP hands make a certain
number of tricks as declarer. As long as we can feel comfortable that
99% [or some large number] of the deals are played "perfectly" then we
can assume that the mean value of tricks taken is truly representative
of SD play.
Kurt
Fred.
>
> - Show quoted text -
If you add the sometimes qualifier to that statement I'll agree with
it. Humans are often capable of simple and conclusive analyses in
complex situations. But, humans also often run into situations at the
bridge table. and elsewhere, which they cannot adequately analyze
(that's the nice way of saying they are clueless). I think the
opening lead against a contract of 1NT is frequently a good example.
Fred.
Fred.
I think claiming pefection based on a series of trials is like
claiming immortality because nothing has managed to kill you yet. Not
only is there the possibility of the claim being refuted in the
future, but there is the systemic expectation that eventually it will
be.
Just as healthy people may live a very long time before demonstrating
their mortality, well made systems may go a long time between
demonstrations of their imperfection.
Fred.
KWSchneider
>
> - Show quoted text -
Ran GIB - took HA, cashed one round of spades and then lead DA
followed by low diamond.
Kurt
jogs
The statement was referring to declarer play
not opening lead, which is often a blind opening
lead. Did not claim the humans would always
find the optimum line. Only said humans are
able to find good lines examining distributions
while Kurt's SS method is still using the brute
force method used in the early chess programs.
OldPalooka
How did GIB do so well in the world championship par contest despite
blowing a board because it misunderstood the opponents bidding? Set
up for slow play, GIB easily handles a sure trick hand such as this.
jogs
First, there's no need to invent your own
vocabulary. In game theory the best strategy
is optimum strategy.
The line by Gary Logan is mostly correct but
not complete.
...
Win in hand and play ace of diamonds.
If all follow, a low diamond to the queen ensures nine tricks.
If rho shows out on second diamond, switch to clubs to
ensure at least two tricks in each minor.
No need to switch to clubs if diamonds are 1-4 to
the right.
Also if on the diamond ace one player shows out
play on clubs.
....
This line only fails if both diamonds are 5-0 and
clubs are 4-1.
KWSchneider
>
> - Show quoted text -
If you would like me to run GIB using multiple scenarios [ie
defender's distributions] that is fine. It can prodice a fully
annotated complete line provided I set up the defenders for the
different shapes that a "full" line needs to examine.
The "brute force" method is surely the only method since, unlike
chess, you have to check a wide variety of what-if scenarios based on
potential defender's distributions.
Kurt
KWSchneider
>
> - Show quoted text -
I believe taking one round of spades slightly improves your chances by
eliminating potential exit cards when the spades split 6-1. Or so GIB
says...
Kurt
KWSchneider
>
> - Show quoted text -
Thank you for your post - I ran GIB at 40% of capability because at
100% it blew up my computer for this hand. I've seen no evidence that
GIB will not find the statistically "best" play. What it will NOT do,
is find the best psychological play when there is no hope of making
the contract.
Kurt
Fred.
>
> - Show quoted text -
I wasn't really disagreeing with you. I thought you meant
"sometimes".
Fred.
Fred.
My point is that 99%, or whatever, is not perfect. It might be really
good in this context. But certainly wouldn't be in others. I've
gone through fully cost-justified efforts to reduce 0.01% failure
rates to 0.001%. And, in some situations 0.001% is sheer garbage,
Fred.
KWSchneider
As an engineer, I'm fully aware of Six Sigma and five, six or even
seven "nine" uptime SLA's, so I understand that "perfect" is relative.
If your access to the internet is only up 99% of the time, you will
experience downtime for over 5000 minutes per year - certainly not
acceptable.
Since we are averaging 1000+ results together to determine a mean
result, we can expect the mean result to be within 3 digit accuracy as
far as tricks are concerned. However, we are only looking at two digit
significance - anything more will be irrelevant.
For example, let's assume that we obtain a mean result of 6.25 tricks
DD and 6.75 tricks SD for a 20point 1N contract over 1000 deals. Since
the accuracy of the 6.75 may be off by .005 tricks due to "imperfect"
play by the GIB engine on one or two hands, it is hardly noteworthy.
And that assumes that the "imperfect" play costs a trick.
Kurt
Thomas Dehn
Meanwhile, at the table, in 1NT, I hold
the C Jx in hand opposite the C xxx in dummy,
and I play a small C from dummy at trick 2, to discourage
a C switch.
RHO well-known international player decided not
to split his KQ, and LHO well-known international player
naturally ducked his CA.
Try to simulate that.
Thomas
KWSchneider
>
> - Show quoted text -
You continue to denigrate SD play with the rare example. So which of
these plays is "correct"? The declarer play or the defense? If GIB is
playing all of the hands, I would suspect that the ploy would not have
worked.
Quoted verbatim from my earlier post regarding SD simulation:
"What it will NOT do, is find the best psychological play when there
is no hope of making
the contract."
Yes, GIB will not find the very rare "out of the box" play that MAY
make a contract as declarer. But then you see these on every hand
don't you?
Kurt
jogs
That's not denigrating SD play. It's an example
of how DD is NOT a proxy for play by real
human beings. They are two different populations.
I interested in how many tricks real players make
while declaring 1NT, not how much they make if
everyone played optimally. The difference between
the two population means is the bias favoring the
declarer. At 1NT it may be as great as nearly one full
trick.
KWSchneider
>
> - Show quoted text -
Unfortunately there is no such thing as "real" people. There are
experts and beginners, and everything in between. There are 2/1
people, SA people and Precision people [systems affect the bidding,
hence defense and lead].
There are "tired" people [in finals of NA championships] and people
who don't really give a damn [BBO speedballs in many cases].
There is a mishmash of results from a wide variety of events and
capabilities. What makes you think that is representative of your
expectation at the table. Isn't THAT what we are trying to come up
with?
Oh - and please do not confuse SD play with DD play - which your
opening sentence seems to do.
Kurt
jogs
You don't have a program which play truly SD.
Such a program would state the optimum line
when given the opening lead, dummy and
declarer's hand. No monte carlo would be
necessary. Pretty sure no such program
has been written.
This mishmash of random play is what one sees
when playing in the real world.
KWSchneider
>
> - Show quoted text -
Your statement is incredibly naive. True single dummy engines like GIB
base the choice of their opening play on "optimum" line with the
parameters given - ie opening lead, dummy and declarer. However, this
"optimum" line is decanted as each card is played and improved to a
point when at some juncture, there is a "perfect" line for success [ie
one that cannot fail, see Kantar's last column in the Bulletin] or we
arrive at NO line for success.
I truly don't understand your obstinance. You apparently have no clue
as to how GIB works.
You gave me a hand to run with GIB, and it found a slightly better
line than your "expert" [essentially the same line nonetheless] and
you continue to gripe about SD by now saying it does not represent
"real people".
Make up your mind - either we want TRUE SD play, where the optimal
decision is made at each trick, or we don't. And if we don't, we
simply have a crapshoot where the results are of little scientific
value.
Kurt
jogs
GIB did not give a superior line. There was no point
in playing the spade ace. Declarer didn't have entries
into the dummy. You gave a totally incomplete line.
Your idea of SD includes DD. Which means it isn't
playing like an expert. Real people don't use monte
carlo for solving dummy problems.
Give a month and year, if you want me to read an
article from the Bulletin.
KWSchneider
>
> - Show quoted text -
Complete lines depend on each play - and is a tree. If you would like
a complete line to what level of depth do you want? It becomes a
pascal's triangle based on level.
As far as playing one round of spades, how do you know that it isn't
superior? How do you know that it may set-up a squeeze or end-play in
some exotic distribution - and since it does not cost anything, it
makes sense to do at trick one.
However, you haven't addressed my major point - why are you stressing
over the accuracy of SD, when you wouldn't accept EXACT SD anyway?
Kurt
jogs
stressing
- over the accuracy of SD, when you wouldn't accept EXACT SD anyway?
>
> Kurt
I'm not rejecting the accuracy of SD. I'm rejecting it
as a valid way to solve the problem.
I ask for the average height of six grade boys. You
survey 1000 six grade girls and insist the average
of your survey is the average height for six grade
boys.
jogs
Yes, the complete tree. Or at least the relevant part of the tree.
You didn't give it.
Thomas Dehn
Those plays are not "rare". They are inherent
to good card play. Create losing options for
opponents. As declarer, be aware of their carding
methods, and select the small cards you play
accordingly. If a defender will have to make
a key decision, such as whether to cover an
honor or not, force him to make that
key decision early. Etc. pp.
Those opportunities exist both for declarer and
for defenders.
Here a defensive problem as an example.
N/S vul. Dealer W. IMPs.
You, E, hold
QJ
Ax
Axxxx
xxxx
Bidding:
W N E S
2H(1) p p 3S(2)
p 4S
(1) weak-two, at this vulnerability approximately 4-9 HCP
(2) decent hand with 6+ spades
W leads the HK, and dummy comes down with:
K9x
xxx
KJTx
xxx
Plan the defense.
Thomas
KWSchneider
I need East's 13th card and to remove dummy's 14th card...
Kurt
KWSchneider
Sorry eyesight problems...
Herb
Are you just trying to make be feel better about misreading/miscounting
a hand in another earlier thread? ;)
- Herb
KWSchneider
>
> - Show quoted text -
I'm sympathetic to your postion - somehow I've developed a problem
reading the quantity of "x's". And certainly 6 and 8 are starting to
look similar. Perhaps I need a new font?
Being 58 doesn't help...
Kurt
KWSchneider
I've run GIB and it overtakes HK, leads DA and then returns low
heart...
Kurt
Thomas Dehn
That is the correct defense. Good to see that GIB finds it.
At the table, partner won the HQ, and then
did not lead another heart, because he had
only a small spade, and thus "nothing to promote".
And that is one of the big differences between
declaring and defending - at defense, it usually
takes two to find the best play.
Thomas
KWSchneider
Although I agree that this is best defense, unless partner has Tx, I
don't see how this provides us a trick unless the restricted choice
decision has changed. Irrespective of whether I ruff with an honor
when partner returns the 13th heart or if we let declarer try the suit
himself, it is still the same decision.
No doubt the DA first, clearly demands a heart return but it certainly
appears that your partner can decide whether a spade in his hand can
be promoted.
Kurt
Thomas Dehn
Yes, it is not that easy to see. Especially,
it is not easy to see for partner who holds
just a small singleton spade.
The point is that the spades are
K9x
x QJ
ATxxxxx
There is no restricted choice there, declarer
has ten spades, not nine. He just draws trumps,
and makes 4S.
After another round of heart has been played,
ruffed with a H honor, and overruffed with the
SA, declarer has:
K9x
?? ??
Txxxxx
Declarer now has to guess spades, with the
Jx resp. Qx still out.
Thomas
KWSchneider
More likely the spades start:
K9x
xx QJ
ATxxxx
where there is a restricted choice, although with the bidding and the
drop of an honor first round, most would play for 2-2.
After the ruff and overruff, we have:
K9x
xx J
Txxxx
On the lead of a trump from declarer, It seems to me that nothing has
changed - still restricted choice and with the bidding the drop is
still the play.
Kurt
Thomas Dehn
I am telling you that at the table declarer had seven spades ;-).
I considered seven spades to be likely - would he
reopen 3S rather than 2S on something like
ATxxxx,xx,Qx,AKQ? Maybe not.
Then, in case declarer really has only six spades, he
might have Axxxxx, and partner has the Tx. Then you
have an actual trump promotion.
Anyways, that wasn't my point. My point is that
such plays are easier for declarer because dummy
will always cooperate.
Thomas
KWSchneider
Understood - hopefully GIB has gotten your seal of approval...
Kurt