Author: KarinsDad
Date: 08:14:30 05/05/99
Go up one level in this thread
On May 04, 1999 at 18:35:50, William H Rogers wrote: >I read your post and I don't think that I understand what's wrong. >I am using a Pent 120 and QBasic (not quick basic) and set maximun depth to >4 plys. >From the starting position, white moves e2e4, my program calculated 45,614 >positions in 40.60156 seconds for a nps of 1140. >Making a second move of f1c4, the program calculates 60,029 positions in >54.60938 seconds for a nps of 1111. >I increment my node counter with every move that the program calculates and >evauluates. Maybe this is wrong, but I think that one of us is doing something >not quite correct. Lets try to find out who, what or why. >Bill I calculate my nps at the moment with every call to GenMove. GenMove should really be called GenPosition since it grabs a chunk of memory and stuffs in all of the valid position information (including position, flags, BitBoards, and PieceAttack tables). I do not calculate a nps for every move from the legal move engine since a lot of them are not used (i.e. if I get a cutoff). However, if I did that, it would probably increase (at the moment until I get better evaluation code) my nps by 30%, but I do not think that is right. I do not calculate my nps from the evaluator since I often skip nodes (for example with an extension) where I do not call the evalulator for the parent node, but only for the leaves of the checked nodes below it (i.e. the evaluator is not called for any of the branches below it, just the leaves). I still count the branch nodes in my nps since they acquire their score from their children, just not from the evaluator directly. I feel that since I created the position for the branch and it has a score (probably more accurate than the leaves anyway), that it is a legitimate node. Do other people do it this way? KarinsDad :)
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.