Author: Robert Hyatt
Date: 15:05:06 01/20/00
Go up one level in this thread
On January 20, 2000 at 16:35:53, Ricardo Gibert wrote: >On January 20, 2000 at 15:06:36, J. Wesley Cleveland wrote: > >>On January 19, 2000 at 21:01:47, Robert Hyatt wrote: >> >>>On January 19, 2000 at 19:24:37, Ricardo Gibert wrote: >>> >>>>[snip] >> >> >>>Feel free to refute this... but please do so with data rather than opinion. >>>It is easy to produce... >> >>It is also easy to hack crafty to check for null move failures. At the point in >>the code where you exit on null move cutoff, just set a flag and continue. Then, >>after you do the full search, if the null move would have produced a cutoff, and >> the full search did not, you have a null move failure. > >You don't need to do this. You can save yourself the trouble by noting that in >say for example in a particular game, null move is effective up to about move 60 >by going 10 ply, then you would expect null move to be effective up to about >move 55 by going 20 ply instead. Null move loses its effectiveness for about 8% >of those 60 moves. You don't need to consider the fact you are searching with >incrementally increasing depth, since the searches at lower depths are a drop in >the bucket compared to 20 ply search. So it is reasonable to surmise that with >an increase in search depth of 10 to 20, null move will lose its effectiveness >for a small, but significant percentage of the positions encountered in a game. > >Any test you do will probably reflect something like this, so why bother? Perhaps because it doesn't always work that way? Suppose you have checks to prevent doing null-moves in some positions. Then starting in a position where you will reach more of those positions as you go deeper does _not_ hurt the program, since it simply avoids doing the null-move search when it shouldn't do it.
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.