Author: Robert Hyatt
Date: 08:04:33 01/25/01
Go up one level in this thread
On January 25, 2001 at 09:51:50, David Rasmussen wrote: >On January 25, 2001 at 09:34:05, Robert Hyatt wrote: > >>On January 25, 2001 at 08:20:26, David Rasmussen wrote: >> >>>Howdy. >>> >>>Inspired by the thread on extensions, I was wondering whether the idea of >>>negative extensions or reductions could be a good one. >>> >>>I mean, maybe many of the "unsound" pruning methods would be sounder if, instead >>>of just pruning, they just adjusted the resulting depth down. In that way, a >>>line would still be examined, only later. >> >> >>This is what null-move search does, in essence... > >Exactly, but null-move does it because it assumes that a null-move can't be very >good. I am talking about doing it in other cases as well, where we might think >that the move is not very good as well. Extensions extend exciting lines. I want >to reduce boring lines too. There are several well-known approaches to doing this... "razoring" is one. I (in general) don't particularly like extensions (or de-extensions) that depend on the value of alpha and/or beta. They tend to make the search unstable if you adjust either bound. But I am sure there must be ideas... ie something I tried a long while back was this: at a fail-low position (It appears that I will have to search all moves and none will be better than alpha) after I searched the first X% of the moves, I searched the remainder at a shallower (1 ply shallower) depth. Showed a lot of promise, but I didn't spend a lot of time testing, tuning and tweaking. The point is that if all moves are bad, why search them so deeply? Of course, you miss that brilliant tactical shot here and there. But you might get a ply deeper too, and find an even more brilliant shot...
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.