Author: Uri Blass
Date: 10:59:36 12/07/01
Go up one level in this thread
On December 07, 2001 at 13:33:34, Severi Salminen wrote: >>both make unmake and gen are important > >Of course, but one should sacrifice the speed of make/unmake to make gen faster. > >>At this time make unmake are relatively slow and making a move takes clearly >>more time then generating a move. > >When you have programmed searching functions also, you can test it more. The >better the ordering, the faster the gen() should be (and usually this will bring >the speed of make()/unmake() greatly down, and also the perft speed). Obviously >if you have to make()/unmake() every move this kind of optimization will hurt. I do not understand. why do you think that better ordering of moves is going to do my gen faster. The opposite it may do my gen slower if I need there to calculate the order of moves > >>I do not think that calculating perft emphasize making moves. > >It ephasizes if you do it like it is done in Crafty. But since you don't >make/unmake() the last ply it emphasizes gen() more in your program. The problem >with your approach is that it doesn't reflect anything about the speed of a >normal search. In normal search if you visit N nodes you also have to >make()/unmake() the last moves - perft() is like a full width search to a >certain depth, we just don't evaluate anything. And you don't even reach those >nodes, you just know how to get there. Yes, you will get the exact node count as >Crafty but you can't compare the time. I get a lot of knowledge by my way about attack arrays. If I decide that I do not want this knowledge for the evaluation function then I can decide to make the last ply by a different function that does not generate knowledge that I do not use in the last ply. I have to know that a ply is the last ply and I understand that there may be a problem here because I am not sure if programs know that the ply is the last ply before generating the move but I still can save time by using another function when I suspect that I do the last ply and if I discover that my suspect was wrong I can complete the generation of the missing information. 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.