501 Essential Backgammon Problems???
Diceman
Problems' and have been told that there are certain flaws in some of
the problems, although i am enjoying the book, i would like to know if
anyone can point out these errors and what the appropriate moves should
be.
thanks in advance
Sent via Deja.com http://www.deja.com/
Before you buy.
Erik M Sørensen
"errors" and the appropriate moves /cubedecisions. All problems have been
evaluated by Snowie 3. I will later post a complete list.
Problem no. Correct move (p = makes the point,* = hits)
1 11 23
3 14
4 9 5*
7 23 11
23 5 and 3-points
24 5 7 18-p
32 21* 15
33 8 21
36 22 1*
42 13 4*
50 20 5*
53 11 3-p
64 12* 4
67 24* 1-p 11
74 8-p 7-p
75 1-p 23 7
83 22 7
101 7 8
102 22 23
103 3-p 5-p
107 8
129 5*10*
134 12
146 ND/T
148 ND/T
175 D/T
192 23* 8
Erik M Sørensen (emsu)
Diceman skrev i meddelelsen <8grqla$28f$1...@nnrp1.deja.com>...
hi
in Backgammon after waiting too long for delivery.
Diceman <scou...@hotmail.com> wrote in message
news:8grqla$28f$1...@nnrp1.deja.com...
Robi
like in Problem 1 is not a problem, but if you find something bigger, that
would be interesting for us. thanks.
Robi
Erik M Sørensen <e...@get2net.dk> schrieb in im Newsbeitrag:
7XvY4.281$5x....@news.get2net.dk...
Erik M Sørensen
favour. Let me hear from anybody that has made any intensive rollouts on the
positions.
Again let me point out that this is only a temporary list.
Problem no. Correct move Equity-diff
1 11 23 0,014
3 14 0,003
4 9 5* 0,004
7 23 11 0,003
23 5-p 3-p 0,097
24 5 7 18-p 0,003
32 21* 15 0,001
33 8 21 0,028
36 22 1* 0,040
42 13 4* 0,006
50 20 5* 0,035
53 11 3-p 0,064
64 12* 4 0,004
67 24* 1-p 11 0,005
74 8-p 7-p 0,005
75 1-p 23 7 0,066
83 22 7 0,007
101 7 8 0,054
102 22 23 0,012
103 3-p 5-p 0,014
107 8 0,025
129 5*10* 0,016
134 12 0,045
146 ND/T 0,014
148 ND/T 0,018
175 D/T 0,199
181 D/T 0,091
192 23* 8 0,004
Robi skrev i meddelelsen <3933...@news.datacomm.ch>...
Pip_Panther
news:3dRY4.243$Bh5....@news.get2net.dk...
> Right, here's the Snowie 3-ply evaluation of the equitydifference in its
own
> favour. Let me hear from anybody that has made any intensive rollouts on
the
> positions.
> Again let me point out that this is only a temporary list.
>
> Problem no. Correct move Equity-diff
> 101 7 8 0,054
>
>
>
Erik,
Thanks for your work on this. When you posted it I printed it out and left
the folded page in my book. I was looking over problems 85-110 today and
came across this one. Most of the other disagreements between Bill and
Snowie I'd look at and say... hmmm, ok... that makes sense, she might be
right. This one makes no sense to me though.
Was there a typo or does Snowie really think 13-8 13-7 is better? If so,
would anyone care to comment why?
Here is the position:
X to play (6 5)
+24-23-22-21-20-19-------18-17-16-15-14-13-+
| X O O O O O | | X |
| X O O O O O | | X |
| | | X |
| | | X |
| | | X |
| |BAR| |
| | | |
| | | |
| X | | O |
| O X X X | | O |
| O X X X | | X X O |
+-1--2--3--4--5--6--------7--8--9-10-11-12-+
Donald Kahn
Take a checker off the midpoint.
My Snowie rollout (cube centered) does not show a preference for 13/8
13/7. It makes it equal to 13/8 10/4.
13/7 is a bold stroke to equalize the position and win the game. But
only because you can't lose a gammon (Jacoby).
dk
Erik M Sørensen
101. (Allthough rollouts show that the alternatives are indeed close.)
You found one of the really funny problems. When analyzed on 3-ply 100 %,
huge searchspace in Snowie, the two alternatives 13/8 10/4 and 13/7 13/8 are
both only 2-ply evaluated. At that level Snowie prefers 13/8 10/4 by 0,154 -
that means huge blunder not to make that move.
When making a 3-ply evaluation, suddenly Snowie prefers 13/7 13/8 by 0,054
!!!
How can that be, and how often, and in what kind of positions, does 3-ply
evaluation differ so much from 2-ply evaluation ?? Anyone?
Pip_Panther skrev i meddelelsen ...
Pip_Panther
diagram. I corrected it below.
I have to admit I would never have played this one over the board. When
Snowie does 2 and 3 plies that means 2 or 3 exchanges of moves from each
player, right? I think I read in another thread that 1 ply is a move by each
side. Either way, it has to be something about X's position after O misses.
Otherwise it's an ace point game with X likely to have 4 checkers on 24.
X has greatly superior timing. But 3 rolls instead of 2? Could it be that if
O doesn't escape and he doesn't pay for better timing by getting points made
on his head, his three checkers on 12 are 21 pips away from his inner table?
In about 3 rolls his board starts to collapse while waiting to jump X's
prime. If X's timing holds up, his two checkers on 24 can probably wait
until O clears his outfield. Also, X will probably get a shot or two at a
6,something joker as O comes down.
<dross snipped out>
"Erik M Sørensen"
> Yep - no typo here (phew :) )
I am the king fo tpyos!
>- Snowie actually prefers 13/7 13/8 on problem
> 101. (Allthough rollouts show that the alternatives are indeed close.)
> You found one of the really funny problems. When analyzed on 3-ply 100 %,
> huge searchspace in Snowie, the two alternatives 13/8 10/4 and 13/7 13/8
are
> both only 2-ply evaluated. At that level Snowie prefers 13/8 10/4 by
0,154 -
> that means huge blunder not to make that move.
> When making a 3-ply evaluation, suddenly Snowie prefers 13/7 13/8 by 0,054
> !!!
> How can that be, and how often, and in what kind of positions, does 3-ply
> evaluation differ so much from 2-ply evaluation ?? Anyone?
>
> >Pip's Post
> >Was there a typo or does Snowie really think 13-8 13-7 is better? If so,
> >would anyone care to comment why?
> >
> >Here is the position:
> >
> >X to play (6 5)
> > +24-23-22-21-20-19-------18-17-16-15-14-13-+
> > | X O O O O O | | X |
> > | X O O O O O | | X |
> > | | | X |
> > | | | X |
> > | | | |
hretter
Donald Kahn <don...@easynet.co.uk> schrieb in im Newsbeitrag:
CMBwOb4wHr37mE...@4ax.com...
> 13/7 is a bold stroke to equalize the position and win the game. But
> only because you can't lose a gammon (Jacoby).
>
> dk
Donald, you can't lose a gammon???
After the bold move Volatility is sky high and O is solid favorite, so
double/take looks mandatory.
Did you do cubeful rollouts?
Harry
JP White
> When
> Snowie does 2 and 3 plies that means 2 or 3 exchanges of moves from each
> player, right? I think I read in another thread that 1 ply is a move by each
> side.
Not exactly. 1-ply doesn't include any 'moves' at all.
Snowie's help File tells us the following
1-Ply evaluation:
A 1-Ply evaluation is directly the output of the neural net. That means Snowie
is evaluating the position with its neural nets and the output is the 1-Ply
cubeless
evaluation. Snowie then converts it into a cubeful equity. This is the fastest
evaluation type.
2-Ply evaluation:
A 2-Ply evaluation is computed by looking one move ahead. That means that Snowie
is playing the best move for each of the possible 36 dice rolls, the resulting
positions are then evaluated 1-Ply and the average of these 36 evaluations is
the result of the 2-Ply evaluation. Notice that Snowie plays with the doubling
cube after
all the 36 dice rolls to produce an accurate cubeful evaluation of the position.
If Snowie has to make a checker play based on 2-Ply evaluations it will first
compute the 1-Ply evaluation for all legal plays, then it will reevaluate the
best
candidates with a 2-Ply evaluation. This is the role of the search space to
select which moves will be reevaluated for the 2-ply.
3-Ply evaluation:
A 3-Ply evaluation is computed by looking two moves ahead. That means that
Snowie is playing the best move for the first 36 possible dice rolls and then
again for
the following 36 possible dice rolls, the resulting positions are then evaluated
1-Ply and the average of these 1296 evaluations is the result of the 3-Ply
evaluation. As
for the 2-ply evaluation, the cube is turned if Snowie thinks it has a double
within this sequence.
JP
--
JP White
jp.w...@nashville.com
Pip_Panther
So just to clarify my understanding from reading your response I take it
that:
1 ply = Evaluate the position for Snowie's next move
2 ply = For all of Snowie's possible moves evaluate the position after
her opponent makes a move.
3 ply = For all of the combinations of 2-ply, evaluate the position
after Snowie moves in response to her opponent's move.
So, the three ply portion of NOT getting hit in the evaluation, Snowie
thinks something similar to:
One ply: Which looks prettiest 13/8,13/7 or 13/8,10,4?
Two ply: IF I take 8,7 Does O get to hit? (15 of 36) 6,3 hits twice.
Three ply: Oops, O missed!, Now, 27 rolls cover, 20 of them build
a 5prime, pip count about the equal, timing belongs to X
Is that perhaps why 3ply shows such a difference?
> >X to play (6 5)
> > +24-23-22-21-20-19-------18-17-16-15-14-13-+
> > | X O O O O O | | X |
> > | X O O O O O | | X |
> > | | | X |
> > | | | X |
> > | | | |
> > | |BAR| |
> > | | | |
> > | | | |
> > | X | | O |
> > | O X X X | | O |
> > | O X X X | | X X O |
> > +-1--2--3--4--5--6--------7--8--9-10-11-12-+
The rollout must have been cubeless like Harry pointed out. I don't see O's
current chances for a gammon so good that X would take on a 41% shot at
making it nearly a sure thing.
Dammit, where is that Visa with the $380 left on it?
Long sigh,
Pip
JP White
> Thanks JP.
> So just to clarify my understanding from reading your response I take it
> that:
I lifted the description directly from Snowie's help file. I do not know the
internal workings of Snowie. However I am happy to interpret it for you (for
what it's worth). There is some ambiguity in my mind. Maybe Mr. Snowie can
answer your questions more accurately.
>
> 1 ply = Evaluate the position for Snowie's next move
Looks like what it says to me.
>
> 2 ply = For all of Snowie's possible moves evaluate the position after
> her opponent makes a move.
This sounds correct. But what is it evaluating? The equity of the position
prior to Snowie's next dice roll?
Also the implication is that it evaluates non doublets (e.g. 1-2 and 2-1)
separately. Hmm how could the result differ? Why not evaluate the 21 dice
combinations?
>
> 3 ply = For all of the combinations of 2-ply, evaluate the position
> after Snowie moves in response to her opponent's move.
>
Looks right to me.
>
> So, the three ply portion of NOT getting hit in the evaluation, Snowie
> thinks something similar to:
> One ply: Which looks prettiest 13/8,13/7 or 13/8,10,4?
> Two ply: IF I take 8,7 Does O get to hit? (15 of 36) 6,3 hits twice.
> Three ply: Oops, O missed!, Now, 27 rolls cover, 20 of them build
> a 5prime, pip count about the equal, timing belongs to X
>
> Is that perhaps why 3ply shows such a difference?
>
The further Snowie can look ahead, one would assume the better the analysis.
However if it's initial 1-ply evaluations are way off, it may never do 2 and 3
ply analyses on some of the candidate plays. It doesn't evaluate each candidate
3-ply, just the ones it likes the best at first glance. I have noticed in some
positions you can force it evaluate a low ranking candidate (i.e. the one I
played) at 3-ply and it jumps way up the list, sometimes to the top! More often
than not it gets it right though, and my play stinks!
There is mention in the help file of 4-ply 5-ply and 6-ply evaluations. These
can be specified as a special sort of mini-rollout for one or more candidate
plays. One could argue that 2-ply and 3-ply evaluations are very very
abbreviated rollouts. Rollouts of positions seldom play to the end of the game
(unless you tell Snowie to do so as a full rollout). The way I see it, rollouts
are just n-ply evaluations and Snowie varies what n is dependent on the
positions it comes across and the parameters you supply. Of course I could be
totally wrong on this.
>
> Dammit, where is that Visa with the $380 left on it?
>
It's probably in the mail box. 1.9% (introductory) APR, sign here to terminate
your financial future.
When you get the VISA card, send the number to Michael Crane, one of his
personas is collecting credit card numbers for some reason.
--
JP White
jp.w...@nashville.com
Donald Kahn
Of course you are right. I don't know where that Jacoby brainstorm cme
from.
I got to work and did careful, cubeful rollouts. 3888 with 3 ply
lookahead and the cubefulls with 2 ply.
First some benchmarks for 24/13 and 24/18 13/5:
1. 24/13 Cubeless:
1.9 36.9 61.9 38.1 8.6 .4 = .528 cubeless equity for o.
Indicates .547 equity if the cube is centered and .779 if x owns it.
Cubeful: No double .523 double/take .777
2. 24/18 13/5 Cubeless:
1.9 37.2 59.5 40.5 10.5 .4 = .471 cubeless equity for o.
.444 equity with center cube, and .649 if x owns it.
Cubeful: No double .451 double take .645
Both these, for money are given recommendation double and take.
Now to the real contenders.
3. 13/7 13/8 cubeless
2.2 27.6 58.2 41.8 10.8 .6 = .349
.311 equity with centered cube, .395 if x owns it.
Cubeful: No double.327 double take .388
Double, take recommended
4. 13/8 10/4 Cubeless:
1.2 20.1 59.7 40.3 12.5 .8 = .273
.389 equity with centered cub, .253 if x owns it.
Cubeful: No double .391 double take .253
No double recommended.
I have to believe the numbers and what they seem to say is this:
If x exposes a blot, it is worth doubling because if you hit it your
gammon chances go up considerabley and you lose your market if you
wait and then hit.
In case 4, the "normal" play, the old wisdom of not giving up the cube
in a non-volatile prime vs prime game seems to remain in force.
One of the best problems ever posted here. Very deep.
dk