Author: Robert Hyatt
Date: 13:35:35 10/04/00
Go up one level in this thread
On October 04, 2000 at 11:48:59, Mike Adams wrote: >I look forward to the day when Pulsar runs on a 2 gigahertz machine or better. >But two points. First i think that two programs that search the same depth >about play each other the one with the better evaluate should win. This is not >always true because even if there at the same speed if one makes better use of >extensions i would imagine it would have the advantage. > Second is that even if hardware improves greatly, in engine vs engine >battles on the same hardware if one is a more effecient searcher or a faster >searcher in nodes per second it should have the advantage all other things being >equale. > Do you have any suggestions for the obvious next thing to work on to get >more depth? I use move ordering in this order: first hash move if available, >second captures and in this i order most valuable capture least valuable piece >and if two captures score the same value by that ordering method i give a slight Here is my suggestion: take crafty and search to a fixed depth on a known position (use sd=n command). Check the nodes searched. Then do the same for your program. If your tree is far bigger, then you know what is wrong. Find out what I am doing order-wise and extension-wise that you are not. >advantage to one if its a recapture, third killer moves i use 2, the last two >moves that produced alpha cutoffs. I imagine i should probably play the hash >move if possible before generateing moves i think i'll try that and see if it >increases speed. I also use R2 Null Move and futility pruning in search and >qsearch. > I've done some web site research the next best thing it seems to me is to >ponder. I could also do a windowed search and history table but I havent heard >that they improve things much. Also i'm trying to improve nodes per second by >making some algorithms faster. See my post about nodes per second above for >some problems i'm having. Normal nodes per second is 15K to 30K but sometimes it >drops as low as 2K for no reason. Maybe hardware problem? Any suggestions are >welcome. I dont expect a detailed analisis of my methods and how to improve them >but if you have a couple of bullet points as to what direction to move in or >what to possibly change it would be appreciated. thanks Mike Adams
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.