Andrew Paik wrote:
> Hi Everyone,
> On the surface, this seems like a really easy question. What is
> currently believed to be the best way to play the different opening
> rolls? I thought I knew this stuff, but when I read the ACM/TD-Gammon
> article, they said that TD-Gammon had `changed the way experts play'.
> For instance, they said that slotting the five was now generally
> considered inferior to splitting the back men to use the 1 for some of
> the opening rolls. This was news to me, but then I'm not one of the
> experts. Since there are experts here, I thought I would ask...
It is quite true that rollout results from three backgammon playing
computer programs (Expert Backgammon, TD-Gammon, and Jellyfish) have
given us new insights into opening rolls and other phases of the game.
Before taking any of these as gospel, there are several things to keep in
1) The strengths and weakness of the programs. For example, a program
which is weak in backgame play might downgrade early slotting plays in
its rollouts since these plays will probably lead to backgames more often
when the slotted blots are hit.
2) The rollouts do not take into account access to the cube, which might
make a difference.
3) There can be quite a bit of variance in the rollouts due to luck, even
with large sample sizes.
4) The rollouts do not take into account individual personalities of
players playing. A player may well do better with an "inferior" opening
play if it suits his personal style.
5) If playing a match, the match score may affect the value of different
Now, on the what I believe is an accurate synopsis of the 15 possible
2-1: The slotting play 13/11, 6/5 and the splitting play 24/23, 13/11,
the two most common plays, seem to be about equal. Nothing else is a
3-1: 8/5, 6/5 is obviously the only play.
4-1: The splitting play 24/23, 13/9 has come out clearly superior to the
slotting play 13/9, 6/5. Probably the reason is that with the builder on
the 9 point there are so many good pointing numbers next turn anyway that
you don't need the 5 point slotted.
5-1: The splitting play 24/23, 13/8 has come out a bit better than the
slotting play 13/8, 6/5. A third less common alternative, 24/18, came
out clearly worse.
6-1: The obvious 13/7, 8/7 is correct. Magriel's experiment of 13/7, 6/5
3-2: The splitting play 24/21, 13/11 came out a bit better than building
with 13/10, 13/11.
4-2: 8/4, 6/4 of course.
5-2: The normal play for years has been 13/11, 13/8. However the newer
splitting play, 24/22, 13/8, (shunned because of the crushing 5-5 threat)
has come out a bit better. The slotting play of 13/8, 6/4 (which used to
be my choice) did not survive the rollouts -- it was clearly inferior.
6-2: The splitting play of 24/18, 13/11 comes out fairly clearly
superior. Running with 24/16 is 2nd, but the run isn't far enough.
Slotting with 13/5 (a common choice several years ago) was definitely in
4-3: The building play of 13/10, 13/9 and the common splitting play of
24/20, 13/10 were just about tied. The alternative split of 24/21, 13/9
was only a little behind.
5-3: The simple 8/3, 6/3 is clearly best. The once common 13/10, 13/8
has been found vastly inferior.
6-3: The splitting 24/18, 13/10 comes out best, but the running play of
24/15 is not too far behind.
5-4: Splitting with 24/20, 13/8 and building with 13/9, 13/8 come out
quite close (that builder on the 9 point is powerful), with the split
generally a tiny bit better. 24/15 is weaker still.
6-4: Both running with 24/14 and splitting with 24/18, 13/9 are about
equal. However the once laughed at 8/2, 6/2 has reared its head as a
serious contender and comes out about equal with the other choices --
nice play to try if you get familiar with it, since your opponent
probably won't be.
6-5: The simple 24/13 is clearly better than any other possibilities.