Gnu insults my perfect cube handling

[Paul]

Gnu says that I had made no cube errors in this money game until I
rolled 52 as shown below. Gnu then fined me 0.099 for not doubling.

Surely that analysis is nonsensical. How on earth can I gain my
market? The play-on is obviously correct because I've got no market
gainers, and my gammon chances (although small) are not zero.





GNU Backgammon Position ID: btsBDBDYtg0HAA
Match ID : UQkVAAAAAAAA
+13-14-15-16-17-18------19-20-21-22-23-24-+ O: gnubg
| X | | O O O O O | 0 points
| X | | O O O O O |
| X | | O O |
| | | |
| | | |
v| |BAR| |
| | | |
| | | |
| | | |
| O X X X | | X X X | Rolled 25
| O X X X | | X X X O | 0 points
+12-11-10--9--8--7-------6--5--4--3--2--1-+ X: Silvia (Cube: 2)

[Bradley K. Sherman]

Paul <peps...@gmail.com> wrote:
> Gnu says that I had made no cube errors in this money game until I
>rolled 52 as shown below. Gnu then fined me 0.099 for not doubling.
>
>Surely that analysis is nonsensical. How on earth can I gain my
>market? The play-on is obviously correct because I've got no market
>gainers, and my gammon chances (although small) are not zero.
>

> +13-14-15-16-17-18------19-20-21-22-23-24-+ O: gnubg
> | X | | O O O O O | 0 points
> | X | | O O O O O |
> | X | | O O |
> | | | |
> | | | |
>v| |BAR| |
> | | | |
> | | | |
> | | | |
> | O X X X | | X X X | Rolled 25
> | O X X X | | X X X O | 0 points
> +12-11-10--9--8--7-------6--5--4--3--2--1-+ X: Silvia (Cube: 2)

I'll guess that it's because in lots of games you will leave a
shot on the 13-pt, rather than break your prime, and that gives
O some winning chances, or that you will have to break the prime
giving O a chance in the race. I would not have doubled, however.

--bks

[Paul]

On Feb 10, 10:21 pm, b...@panix.com (Bradley K. Sherman) wrote:

Doubling is quite obviously an error, though maybe a minor one,
because it needlessly throws away X's gammon chances. If X leaves a
shot on the 13 point and O hits it, X just doubles O out. That's what
I mean by "no market gainer".

[Tim Chow]

On Feb 10, 4:57 pm, Paul <pepste...@gmail.com> wrote:
>  Gnu says that I had made no cube errors in this money game until I
> rolled 52 as shown below.  Gnu then fined me 0.099 for not doubling.
>
> Surely that analysis is nonsensical.  How on earth can I gain my
> market?  The play-on is obviously correct because I've got no market
> gainers, and my gammon chances (although small) are not zero.

It does seem that GNU's evaluation is off base here. I'll try to
remember to see what eXtreme Gammon thinks.

As for whether you have any market gainers, what if you roll 21? I
guess you'd play 9/8 9/7? Then if O rolls 61 and runs with 23/16, it
seems you've regained your market.

---
Tim Chow

[Paul]

No, I would keep the 6 prime with all rolls. 21 plays 13/12 13/11.
Gnu's strong board is actually a weakness because it enables me to
keep my prime.

Let's continue this challenge, Tim.

I will play 13/12 13/11. Please give yourself a roll and a play that
allows you to take my double on the next turn.

Paul

[badgolferman]

I have seen other instances similar to this with gnubg. Often times
gnu dings me for not doubling also but I tend to play on and eventually
get the gammon. However there are occassions when I have ignored gnu
and it has got the perfect roll to crush me. Maybe it was trying to
warn you to be happy with two points knowing what was about to
happen.......

[Tim Chow]

On Feb 11, 3:58 am, Paul <pepste...@gmail.com> wrote:
> I will play 13/12 13/11.  Please give yourself a roll and a play that
> allows you to take my double on the next turn.

Good point. Perhaps if I roll 11 and hit 2 or 3 blots? But even then
it's not clear that it's a take.

Anyway, I took a look at this with the bots. First of all, my version
of GNU says that it's slightly too good, on 2-ply evaluation. So that
makes me wonder if you're using a very old version of GNU?

Secondly, eXtreme Gammon version 1.21 on 4-ply evaluation (equivalent
to GNU 3-ply evaluation) thinks it's a tiny error to double (something
like 0.006) but then it changes its mind upon rollout, and concludes
that it's too good. But it's clear that its evaluation is suspect;
for example, after the variation above (13/12 13/11 followed by snake
eyes), on evaluation, XG thinks that O is doing well, barely an
underdog!

Perhaps someone with eXtreme Gammon version 2 can comment on how it
sees this position.

---
Tim Chow

[Paul]

I would think this position is a big pass however you play the 11.
But I've been proved wrong so many times in these discussions. All I
can say is that I would pass. That doesn't mean that the 11 is not
an interesting question, even if I'm right. It's possible that some
plays of 11 lead to D/P and some lead to too good.

At DMP, the roll of 11 seems like an extremely difficult problem.
Clearly O should get to the edge of my prime. But how many blots
should O hit? [O's winning prob is by no means tiny but, in my
judgment, far too small for a money take.] O's winning plan is for X
to enter quickly, fail to jump the 5 prime and then for X to run out
of timing and crunch X's own prime. Having many checkers on the bar
gives timing for X on one hand but, on the other hand, makes it more
likely for X to get trapped. I would hit only one blot but it's
basically a 50/50 for me between hitting one blot or hitting two
blots. Hitting 3 seems very bad to me, but I don't consider myself an
expert on this type of position.

You say you wonder if I'm using a very old version of gnu. It is not
the case that I'm using a very old version of gnu. I had two concerns
but used the defaults for everything else. I wanted gnu to play as
strongly as possible so I changed the play setting (I think it's
grandmaster). I also wanted to prep for live play so I took out the
board numberings and the pip count.
I did not think to change the analysis which was on expert level. I
have now changed the analysis to grandmaster level (the strongest
other than user-defined).

What I found interesting was that, before this change, gnu rarely
criticised its own play despite the play being grandmaster and the
analysis being only expert. Before looking into this, I was puzzled
by the fact that gnu sometimes criticised its own play. Now I find
myself puzzled by the fact that gnu didn't criticise its own play more
often.

Paul

[Stick]

XG realizes this position is too good ... but then again so does GNU
on any real analysis that matters. If you give the bot no real
lookahead you can't blame it for not knowing any better. You
shouldn't analyze on 0 ply.

Stick

[Grunty]

Paul wrote:
> Doubling is quite obviously an error, though maybe a minor one,
> because it needlessly throws away X's gammon chances.
> If X leaves a shot on the 13 point and O hits it,
> X just doubles O out.

Aren't you doing the same you criticise?
At present, if X leaves a shot on the 13 point and O hits it, I think X is too good to double.

Now I'm going to make an attempt at calculating the position.

Silvia will leave a shot with 21 and 11. But we have to add 65 and 64, making it 7 rolls (20%).

Then if Gnu hits (20 * 0.3 = 6%), Silvia will enter from the bar: 6% * 0.3 = 1.8%

Now we can imagine she will have 3 or 4 rolls to throw a 6 before breaking her prime, while Gnu will likely have enough time to go with its two walkers around.

Odds Silvia will get a six to jump Gnu's board.
0.3 + (0.7 * 0.3) + (0.4 * 0.3) + (0.1 * 0.3) = 0.66%

So Silvia's prime-breaking chances are:
1.8 * (1 - 0.66) = 0.6%

And Gnu's winning chances are then:
0.6 * 0.3 = 0.18%
as it will have to throw its number too.
Addmitedly, this is a simplification, since Silvia could manage to get her 6 right after Gnu gets its own.

Now let's match this odds to Silvia's tiny gammon odds in the original position. If gammons are better than 0.09% (half the price she risks by playing on) then Silvia is too good to double.

Phewww... any corrections are welcome.

[Tim Chow]

On Feb 12, 7:36 am, Paul <pepste...@gmail.com> wrote:
> I did not think to change the analysis which was on expert level.  I
> have now changed the analysis to grandmaster level (the strongest
> other than user-defined).

GNU 3-ply isn't unequivocally stronger than GNU 2-ply. Have a look at
this chart:

http://www.bgonline.org/forums/webbbs_config.pl?read=114355

To summarize, GNU 3-ply's checker-play seems to be slightly stronger
than GNU 2-ply. GNU 2-ply also has a tendency to double late and take
too deeply in certain positions, notably holding games. However, GNU
3-ply makes a lot of wrong passes. Given that GNU 3-ply takes a lot
more compute time, I rarely bother with GNU 3-ply any more.

GNU 4-ply is notably stronger than either 3-ply or 2-ply but it's so
slow that I use it only for a few selected analysis decisions where I
want a second opinion after looking at what 2-ply says.

---
Tim Chow