Singular Extensions...
Ed Schröder
[ razor ]
Simple question. Did the idea go away over time or is it getting more
attractive again as processing power increases ?
Obviously Ed could comment on that ;-)
-- Peter
Rebel does not use Singular Extensions anymore.
The loss of speed was to high even when I implemented the idea in a faster
way WITHOUT the "research" part (Just marking moves in the hash table as
singular and extending [ply++] in the *NEXT* iteration!)
Has anyone used this idea also?
Results?
Maybe the idea is interesting again since we have fast machines now.
- Ed Schroder -
Robert Hyatt
Ed Schröder (10065...@CompuServe.COM) wrote:
: [ razor ]
:
: Simple question. Did the idea go away over time or is it getting more
: attractive again as processing power increases ?
:
: Obviously Ed could comment on that ;-)
:
: -- Peter
:
:
: Rebel does not use Singular Extensions anymore.
:
: The loss of speed was to high even when I implemented the idea in a faster
: way WITHOUT the "research" part (Just marking moves in the hash table as
: singular and extending [ply++] in the *NEXT* iteration!)
:
: Has anyone used this idea also?
At first blush, this would seem to be somewhat different than singular
extensions. First, what if this move doesn't test "singular" when you
get to ply+1, you still extend because you aren't testing it there? Second,
what if it's singular *now* and you don't get to ply+1 to extend it, then
you get no information.
I'm not saying this is good, bad, or indifferent, just that it's
"different." I plan on playing with it again after Jakarta, just
to see what it does with newer hardware speeds...
: Results?
:
: Maybe the idea is interesting again since we have fast machines now.
:
The sad thing is that everything changes every two years or so. :)
: - Ed Schroder -
Jonathan Schaeffer
Ed Schroder <10065...@CompuServe.COM> writes:
>Rebel does not use Singular Extensions anymore.
>The loss of speed was to high even when I implemented the idea in a faster
>way WITHOUT the "research" part (Just marking moves in the hash table as
>singular and extending [ply++] in the *NEXT* iteration!)
>Has anyone used this idea also?
>Results?
I've had a similar experience in Chinook (checkers). "Pure" singular
extensions resulted in too much search overhead. Instead I implemented
a "modified" algorithm that included some shortcuts for finding singular
moves, marked them in the hash table and used the singularity informaiton
on the next iteration to extend the search.
I did some extensive experiments with this two years ago and concluded
that for checkers it was a win:
- the "modified" singular extensions program beat one not using it by a
reasonable margin
- the "pure" verison was a big loser: the trees were too large.