Author: Robert Hyatt
Date: 19:13:23 09/19/99
Go up one level in this thread
On September 19, 1999 at 20:05:50, Bas Hamstra wrote: >Thanks again. It is so logical, after reading your reply, as usual :) Don't know >what I was thinking. > >But wait, maybe I do. The last couple of plies, the nullmove search is no more >than a QSearch. Nullsearches for the last plies have in fact the *same* depth >(namely 0). When you put the normal searchdepth in the hashtable you don't >profit from that fact. But think about this. In _this_ position you did a null-move search and it failed high. If you reach this position again, with the same (or less) depth remaining, it also ought to fail high to a null-move search. And you can return without searching a thing... > >But I'll try it the normal way first and come back to this if necessary. > >(this is a great place!) > > >Regards, >Bas Hamstra. > > >On September 19, 1999 at 13:52:12, Robert Hyatt wrote: > >>On September 19, 1999 at 12:11:39, Bas Hamstra wrote: >> >>>Hello friends, >>> >>>When you do nullmove everywhere, it causes a lot of qsearches. Without nullmove >>>I have a good qnode/node rate, roughly around 5-10% or so. With nullmove that >>>worsens quite a bit, I see rates over 100% frequently. Is that normal? >>> >>>I read about nullhashing, a few times. I don't hash the result of a nullmove >>>search. To be more precise: when a position is cut by a nullmovesearch I don't >>>put that position in the hashtable, and just return Beta. Should I? And has that >>>got something to do with the bad qrates I'm seeing? >> >>can't answer that, but I don't see why you don't hash the result, as it is >>obviously the right answer in that position... I always do a hash store >>before I fail high, except when the fail high is after a hash probe, of >>course.. >> >> >> >> >>> >>>Can someone give a few pointers for nullhashing? Just putting a position that is >>>about to be cut by nullmove in the table, as a normal upperbound record, but >>>without a move? With flag "NULLHASH"? Record only to be used by nullmove, so >>>store with adjusted depth = depth-R ? That's how I would figure it, and I tried >>>it quickly, but without much result... >>> >> >>I store no move, a LOWER flag, and the normal depth.. >> >> >> >>>A second question: I don't store leaf-nodes at all, just see no point in that. I >>>would like to check if that's normal. A simple yes is enough :) >>> >> >>I don't. I used to. The only reason I don't is that it reduces hash table >>contention and lets me get away with smaller tables when memory is not as big >>as I'd like... >> >> >> >> >>>Regards, >>>Bas Hamstra.
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.