Author: Bruce Moreland
Date: 19:55:56 12/18/02
Go up one level in this thread
On December 18, 2002 at 22:36:27, Russell Reagan wrote: >On December 18, 2002 at 21:16:53, Bruce Moreland wrote: > >>You get into trouble when you are safe for the moment, but the computer has a >>longer term threat that you can't escape. The short depth null-move search >>won't detect the threat. > >In this case, I have to ask, will an extra two plies be able to detect that long >term threat? If it's a long term threat, I wouldn't think it would matter >whether you subtract two (or three) plies from the normal full width search. I will go into more detail. Let's say that your current depth is 5. You are going to make each of your moves, and then give black a 4-ply response. Before you do that, you do the null move thing. What this amounts to is you pass your move and allow your opponent a chance to search. What depth is this search? It depends upon the depth-reduction factor (R). R is how this saves time -- it takes advantage of the fact that moving twice is usually so potent that you don't need a lot of search to kill your opponent if that's possible. The typical value for R is 2. What this means is that the search that black is going to get to do will be 2 plies, which is 4 - 2. So they get a 2-ply search, which is a chance to kill you, and a chance for you to respond. After your response, they don't get a chance to kill you again, because this is a 2-ply search not an infinity-ply search. Some programs cannot detect non-capturing mates at the horizon, so if this program is like that, black could have a mate in one, and it would not be detected. So the knife stops inches from white's heart, white thinks he is fine, and this whole line is discarded as "awesome for white". Let's assume that we had not done the null move test. Now, for each of white's moves, black gets a four-ply shot at white. It is possible that in every case, black makes a threat, white cannot parry it, and black mates on the third move. There's an extra white move left over, but white is already dead. So white is dead, and white will figure this out, and fail low on this variation. Do you understand? If you don't, I will try again, because this is important. This kind of stuff happens a lot when playing against humans, who sacrifice stuff in order to mate you. I think it's the single worst problem that you'll have if you use traditional null move search with R=2. bruce
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.