Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question regarding hashing and move ordering

Author: Robert Hyatt

Date: 08:31:10 10/29/02

Go up one level in this thread


On October 29, 2002 at 11:29:40, Robert Hyatt wrote:

>On October 28, 2002 at 19:39:15, Nagendra Singh Tomar wrote:
>
>>I am using iterative deepening with transposition tables. Suppose at some
>>point in the match I have a board that is say stored in hash with stored depth
>>10. So when iterative deepening searches that position with increasing depth,
>>till depth 10 hash_probe will return the stored position and so alphabeta
>>returns instantly. Since I am not using some hash2pv kind of function, I get a
>>PV length of 1 with the best hash move being on the PV. But now when the
>>iterative deepening searches for depth 11, the (stored_depth>= depth) clause
>>is not true inside hash_probe and hence it does not return the stored
>>position.
>
>Just because the draft was insufficient to let you take the score as good and
>produce a cutoff, you _still_ use the hash move as the first move to search at
>this ply...
>
>
>>Now alphabeta is on its own with no PV to help it order the moves
>>(just some help from the hash inmove ordering). In such cases my program takes
>>an unusually *long* time to search the next iteration (11th in this example).
>>This reminds me of the days when I was not using iterative deepening and
>>searching to a fixed depth.
>>Is this normal behaviour, or dies it say something about the inefficiency of
>>the hash ?
>>What do we do to avoid this ?
>>
>>regds
>>tomar


Another point.  If you are talking about the case where you get a short PV at
depth
N and then depth N+1 is harder to search, look up "Internal Iterative Deepening"
which
will help in that particular case significantly.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.