Problem with GNU here?

[mont...@lycos.com]

I just installed GNU on a new computer and I noticed something very
strange. The hint feature seems to be off in a major, but inexplicable
way. Here is an example position:

GNU Backgammon Position ID: sOfBAyDYzuABBg
Match ID : cAkPAAAAAAAA
+13-14-15-16-17-18------19-20-21-22-23-24-+ O: GNU
| X O | | O O X | 0 points
| X O | | O O X |
| X O | | O |
| X O | | O |
| | | |
v| |BAR| | (Cube: 1)
| | | |
| O | | |
| O | | X |
| O X | | X X X | Rolled 63
| O X | | X X X O | 0 points
+12-11-10--9--8--7-------6--5--4--3--2--1-+ X: monty

13-10, 13-7 is +.329, with 566 wins, 162 gammon wins, 3 BG wins, 79
gammon losses, and 2
BG losses. 13-4 is +.293, with 569 wins, 164 gammon wins, 3 BG wins,
64 gammon losses, and 2 BG losses. This was at 2 ply. How is this
evaluation possible?

[mont...@lycos.com]

Yet in this similar position, GNU likes 13-4 over 13-7, 13-10 by .242
(+.926 to +.683):

GNU Backgammon Position ID: sOfBAyBsO+DBAA

Match ID : cAkPAAAAAAAA
+13-14-15-16-17-18------19-20-21-22-23-24-+ O: GNU

| X O X | | O O | 0 points
| X O X | | O O |

[Nige M]

I am not well versed in GNU techno issues. Here however, is a low-down on
the pos.

In his book: "Backgammon Boot Camp" the main man Walter Trice discusses the
topic of game plans. In this pos, Monty has all three game plans at his
disposal, although each is weighted differently according to the subtleness
of the pos.

For instance, after the roll Monty will be 4 pips up with Gnu on roll - so
an even race.

Most us know how difficult it is to prime a lone checker, however this
option does exist. As to its weighting, we'll return to it later.

Then there is the attack option. For a successful attack, Monty requires
additional forces to bear pressure on the backyard chap. This may take a
couple of rolls, and with each passing roll the threat that GNU will escape
increases.

The paint is now on the canvas, all we need to do now is interpret it.

After 13/4 Monty has the race and attack as his prime options, with little
if any priming capacity. I would estimate these to be weighted evenly.

13/7, 13/10 activates the priming option, with a subtle element of attack
with the chap on the bar-pt. While 3 game plans are better than two, the
risk of being hit reduces the weighting of each game plan. Combined, they
are less than after 13/4.

The advantages when it works makes the decision a close one.

A Snowie rollout indicates an equity of .012

Nigel M.

<mont...@lycos.com> wrote in message
news:1153732191....@p79g2000cwp.googlegroups.com...

[Nige M]

A continuation from the last thread.

The attack option here is weighted greater than the prime option by a clear
margin. In fact, 13/7, 13/10 is obnoxious to say the least.

Nigel M
<mont...@lycos.com> wrote in message
news:1153733144....@p79g2000cwp.googlegroups.com...

[mont...@lycos.com]

Thanks for those 2 posts. Now the question becomes (I guess) is there
a bug in the hint feature in this version of GNU I downloaded recently?

[Adam Tansley]

monty:

"Now the question becomes (I guess) is there a bug in the hint feature
in this version of GNU I downloaded recently?"

The answer to this is no.

In the first example you give,
at 2 ply, 13/10 13/7 is better by 0.0369
at 3 ply, 13/4 is better by 0.0269
at 4 ply, 13/4 is better by 0.0190
in a rollout, 13/4 is better by 0.0307

Thus, the two moves are very close, and this kind of minor disagreement
between different plies, and rollouts, is ubiquitous.
I can't see why these results make you think there may be a bug, and
concluding that "The hint feature seems to be off in a major, but
inexplicable way" would seem to be off in a major, but inexplicable
way.

By the way, in describing the evaluation results you talk of "... 566
wins, 162 gammon wins ...". 556 wins out of how many trials? It means
nothing on it's own. Turns out you meant a 0.556 probability of
winning. Are you aware of this?

tansley

[mont...@lycos.com]

Anyone who uses GNU's hint feature knows that it means 556 out of 1000,
because it will then give the "plain game" losses as 444. And to me,
.037 is not "piddling," but perhaps the important point to remember
here is that 2 ply may be a lot less useful than at least I thought it
was.

[Hardy Hübener]

What I do not understand are the equites presented by GNUBG in the
example given by Monty. It should either be

Wins + Wins(g)*2 + Wins(bg)*3 - Losses - Losses(g)*2 - Losses(bg)*3

or - as Jacoby rule is in force -

Wins - Losses

(While Wins shown by GNUBG are wins, including gammons and backgammons.
So one has to subtract the gammons wins to get the single wins. Same for
the gammons.)

In the example given by Monty the equity for 13/10 13/7 should be 0,2153
(or 0,1314 if you don't count gammons or backgammons due to the Jacoby
rule), not 0,3294 as given by GNUBG.

Hardy ;-)

--
Hardy's Backgammon Pages: www.hardyhuebener.de (Last update: May 2006)

[Raccoon]

There is nothing wrong with the hint feature. If I understand your
question, it is:

How can GNUBG 2-ply analysis say 13/10 13/7 is the better play (but
only by 0.036) when the alternative play 13/4 wins more games, wins
more gammons and loses fewer gammons?

It's possible because the cubeless game outcome breakdown (wins,
losses, gammons) are only part of the evaluation story. The other part
is how you and your opponent can use the cube after either play. 2-ply
evaluation believes that you gain enough equity through more
advantageous use of the cube after 13/10 13/7 to offset 13/4's
apparently more favorable game outcome breakdown.

[mont...@lycos.com]

I thought it might be something like that, but had no idea what to call
it. Let's try to put it in "everyday language" now. After the 13-7,
13-10 play, if my opponent doesn't hit or escape, and I can make the
bar point and establish a strong position, he still has to take the
cube. However, on 13-4, again assuming he doesn't roll something like
6,5, there aren't nearly as many rolls that give me a strong double on
the next roll assuming his position remains about the same, except that
he gains some pips in the race. Is this the point?

[muratk]

mont...@lycos.com wrote:

> Let's try to put it in "everyday language" now. After the 13-7

You cocksucking human scums are bad clowns... :((

MK

[mont...@lycos.com]

MK:

I'm trying to understand how GNU can come to this evaluation. There is
either a bug, or else there is a reason for it, even if you think GNU's
dice are "corrupt," or whatever you want to call it. Are you unable to
understand this point?

[Gregg Cattanach]

There is no 'reason' in MK, i.e. rational thought is impossible for him.
It's like trying to reason with a starfish. Please do not ask him
questions. It just irritates him and you certainly won't get any answers.

--
Gregg C.

[mont...@lycos.com]

Here's another one of these positions:

GNU Backgammon Position ID: dduAAQaNbRswAA
Match ID : cAkXAAAAAAAA
+13-14-15-16-17-18------19-20-21-22-23-24-+ O: GIO111
| X | | O O O O O O | 0 points
| X | | O O O O |

| | | O |
| | | |
| | | |
v| |BAR| | (Cube: 1)
| | | |
| | | |
| | | |

| O X X X | | X X O X | Rolled 65
| O X X X | | X X O X X | 0 points
+12-11-10--9--8--7-------6--5--4--3--2--1-+ X: monty64

GNU 2 ply hint has 16-11, 16-10 as +.367, whereas 7-1, 7-2 is +.362,
but according to the breakdown, it is worse in every way, except for
backgammons for and against, where it is equal to the "lesser" move.

[Philippe Michel]

The equities are cubeful while the breakdown percentages are cubeless.

After 16/11 16/10, O's cubes tend to be somewhat inefficient. In fact,
after X: 16/11 16/10 ; O rolls 6-X and hits ; X dances, O may even be "too
good" and have a cubeful equity *lower* than the cubeless one, because of
the Jacoby rule.