hi,
do you know incompatibility or probleme to use this two techniques on same time?
in rare case: i have a bad result
PS: my table has two info, lower_bound and upper_bound, not score and flag.
thx
aspiration window and transposition table
Moderators: hgm, Rebel, chrisw
-
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: aspiration window and transposition table
I use both and don't have any problems...hcyrano wrote:hi,
do you know incompatibility or probleme to use this two techniques on same time?
in rare case: i have a bad result
PS: my table has two info, lower_bound and upper_bound, not score and flag.
thx
-
- Posts: 1822
- Joined: Thu Mar 09, 2006 11:54 pm
- Location: The Netherlands
Re: aspiration window and transposition table
It seems like Shredder and Rybka 3.0 use several aspiration windows to search with, but i'm not a rather big fan of it. Researches can take forever sometimes at crucial points in the game, and you HAVE to give it time then to resolve, which means in short you might lose precious time.hcyrano wrote:hi,
do you know incompatibility or probleme to use this two techniques on same time?
in rare case: i have a bad result
PS: my table has two info, lower_bound and upper_bound, not score and flag.
thx
I remember a case of Falcon playing Shredder in 2004 and Shredder vs Diep in 2004 Tel Aviv. That last case was a drop of +1.0 to close to draw score, which took Shredder nearly an hour.
Vincent
-
- Posts: 1822
- Joined: Thu Mar 09, 2006 11:54 pm
- Location: The Netherlands
Re: aspiration window and transposition table
That's because you didn't use LMR for a long time yetbob wrote:I use both and don't have any problems...hcyrano wrote:hi,
do you know incompatibility or probleme to use this two techniques on same time?
in rare case: i have a bad result
PS: my table has two info, lower_bound and upper_bound, not score and flag.
thx
-
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: aspiration window and transposition table
I've been using it at least 5 years now, starting in a very early 20.1 or 20.2 version.diep wrote:That's because you didn't use LMR for a long time yetbob wrote:I use both and don't have any problems...hcyrano wrote:hi,
do you know incompatibility or probleme to use this two techniques on same time?
in rare case: i have a bad result
PS: my table has two info, lower_bound and upper_bound, not score and flag.
thx
Why would this matter for an aspiration window set at the root? And how would an aspiration window at the root cause any sort of trans/ref table problems? I've been using aspiration search windows since the early 80's... and I have never seen them cause a trans/ref problem. It is just a different window bound, if it is wrong in one direction the search fails in that direction. This is no different than a normal non-aspiration search where you find a lower-bound after searching the first move, and use an upper bound that is +1 higher, and then you search with that window. You can still fail high, and have the exact same trans/ref table usage. I don't see how it can possibly hurt a thing.
-
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: aspiration window and transposition table
The problem is, had it _originally_ started out with that lower window bound, it would _still_ have taken an hour to resolve via search. You use the aspiraton to artificially prune away (via alpha/beta cutoffs) parts of the tree that produce scores outside your expectation. If you are wrong, you have to relax the bound, but the overhead is only what you have wasted to this point to find out that the window was too tightly constrained, you _still_ have to search the three with the relaxed window and this is no less efficient the second time around than if you had the window right the first time. And in fact, it is probably better because you did learn something about move ordering as you discovered that too many scores lie above or below the aspiration window.diep wrote:It seems like Shredder and Rybka 3.0 use several aspiration windows to search with, but i'm not a rather big fan of it. Researches can take forever sometimes at crucial points in the game, and you HAVE to give it time then to resolve, which means in short you might lose precious time.hcyrano wrote:hi,
do you know incompatibility or probleme to use this two techniques on same time?
in rare case: i have a bad result
PS: my table has two info, lower_bound and upper_bound, not score and flag.
thx
I remember a case of Falcon playing Shredder in 2004 and Shredder vs Diep in 2004 Tel Aviv. That last case was a drop of +1.0 to close to draw score, which took Shredder nearly an hour.
Vincent
-
- Posts: 186
- Joined: Mon Jan 21, 2008 2:07 pm
- Location: Russia
Re: aspiration window and transposition table
I was tested the Rybka 3 with and without aspiration window. Without aspiration window the node count was increased for 1.65 times on depth 12 halfmoves. And the more depth, the greater the difference.