I believe that hash should help at fixed depth search because programs may find the best move earlier thanks to hash cutoff(for example in fine70 programs can find Kb1 earlier than depth 26 thanks to hash cutoff).
I wonder if somebody investigated if using hash for pruning improve the rating at fixed depth search and what is the size of the advantage per depth.
Uri
what is the rating advantage of hash in fixed depth search
Moderators: hgm, Rebel, chrisw
-
- Posts: 10298
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
-
- Posts: 892
- Joined: Sun Nov 19, 2006 9:16 pm
- Location: Russia
Re: what is the rating advantage of hash in fixed depth sear
Earlier or later fixed depth search will reach horizon. There is almost no difference (except random factor of transposition table instability), if your do not consider time.
-
- Posts: 10298
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
Re: what is the rating advantage of hash in fixed depth sear
tranposition table instability is not a random effect and I think that if hash is used correctly hash should help not only to reach the same depth faster but also help to get better result in the same depth in part of the cases.Aleks Peshkov wrote:Earlier or later fixed depth search will reach horizon. There is almost no difference (except random factor of transposition table instability), if your do not consider time.
Uri
-
- Posts: 892
- Joined: Sun Nov 19, 2006 9:16 pm
- Location: Russia
Re: what is the rating advantage of hash in fixed depth sear
Instability is a difference between hashed result and research of the same node again is fundamental because of random difference in move ordering (killer/history tables moment states are "random" factors) between two separate searches of a single position at different time. Different move ordering may lead to different cut moves and different principal variation.
-
- Posts: 10298
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
Re: what is the rating advantage of hash in fixed depth sear
Different move ordering may lead to different cut moves but the point is that cutoff may lead also to finding the right move at smaller depth.Aleks Peshkov wrote:Instability is a difference between hashed result and research of the same node again is fundamental because of random difference in move ordering (killer/history tables moment states are "random" factors) between two separate searches of a single position at different time. Different move ordering may lead to different cut moves and different principal variation.
Uri