Author: Uri Blass
Date: 04:22:05 04/14/04
Go up one level in this thread
On April 14, 2004 at 06:12:29, Tony Werten wrote: >On April 14, 2004 at 00:26:34, Eric Oldre wrote: > >>After you find the 1st "good" move don't you narrow the alpha beta window so >>that you don't know how much worse the 2nd move is, only that it is not as good >>as alpha? >> >>Or do you not narrow the window at the root node? that seems like it would >>greatly expand your search tree. >> >>or am i missing something else? > >Before search you check if the last move is a capture. If so you set >maybe_easy_move flag. I think that the best way is simply to calculate exact scores for all moves(I am not doing it today but I believe in that idea) Of course the exact scores will be based on search to reduced depth so it is not going to be expensive. You can start doing it by searching to depth of 1 ply when the remaining depth is 7 plies(the difference may be later even more than 6 plies if you find that 6 plies is too expensive). You should have a special hash table for exact scores based on searching to reduced depth. advantages: 1)If you see mate score against some move then you know never to search that move(when based on normal search you may never find the mate score because you do not reject the move for the right reason). 2)You can also use it to detect easy moves based on the difference between best move and second best move when easy moves do not have to be captures. 3)You may get better order of moves(internal iterative deepening does not give you the best move based on search to reduced depth but only a move that is good enoough to be above beta and here you may have a move based on search to reduced depth to try). Unfortunately it did not work for me in the last time that I tried to implement it but the reason is clearly a bug in the implementation when the result of searching to reduced depth to get exact score caused increasing the tree more than the nodes that were used to search to reduced depth. Uri
This page took 0.01 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.