Author: Uri Blass
Date: 02:40:42 11/13/02
Go up one level in this thread
On November 13, 2002 at 05:07:48, Sune Fischer wrote: >On November 13, 2002 at 01:10:23, Uri Blass wrote: > >>The point is that it may be better to use average of only 80% of the time for >>the expected move and the rest of the time for other moves when you do not use >>the same time for all of the other moves. >> >>These 80% may be 90% in cases that the move that you predict is played and less >>than 80% in cases that the move that you predict is not played. >> >>Having less risks in trying to predict the opponent's move is also important >>and if I have to choose between using 60% of my time for the played move in >>every case(option that I do not have) and using 100% of my time for the >>predicted move in 60% of the cases I prefer the first option. >> >>I cannot get the first option but I may get something closer to it. >> >>Uri > >I understand your point. >I think it would be better to do it like this: You spend 100% on the pridicted >move until you start suspecting this wasn't the best move, eg. it fails high. > >Now you go back and do a shallow search on all the possible moves, not including >the move you just tried of course. This shallow search will be more efficient >than the threads because it can use alpha-beta all the way. >After a 5-6 plies search (or however long you want to spend finding the >candidate move) you have a move to ponder and you start pondering that. > >If the best move is the best only for a very deep reason, and it wasn't already >the pridicted move, then you won't be able to find it unless you do a long >search like your opponent. So it seems this will always be a ponder miss no >matter what. We can't make a perfect scheme anyway, so we must be willing to >accept a miss now and then and not go out of our way to fix all the exceptions. > >You probably wanted to use the information from the threaded moves search to >help seeing when the pridicted move was wrong, but I don't see what value lots >of 2-4 ply searches have you when the pridicted move isn't failing high? > >-S. I can learn from a lot of 2-4 ply search about the logical moves. In an extreme case there is only one logical move except the move that I ponder so I can continue to search it and give it almost 10% of the time. Suppose that I see the following after expected move: score -0.2 at depth 1-10 after second best move score 0.1 at depth 1,2 score 0 at depth 3,4 score -0.1 at depth 5,6 score -0.19 at depth 7. It is still not the best move but I can suspect that it is going to be the best move because of the behaviour of the sequence of the scores Uri
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.