Author: Georg v. Zimmermann
Date: 05:04:25 12/07/01
Go up one level in this thread
On December 07, 2001 at 02:57:56, Severi Salminen wrote: >>So for improving the raw performance of your own program, against only itself, >>some of these statistics are good tools then, correct? If my eval function is >>bad, then I should work on improving it, but my program will play better if it's >>searching twice as deep with a bad eval function than if it's searching half as >>deep with that same bad eval function. That's what I was really after. Improving >>a program relative to itself, not to other programs (although I would suspect >>that if you improved your ply depth, all other things remaining the same, that >>the program would perform somewhat better against other programs). > >For developing your own program, all statistics are very useful, including NPS, >search depth, game results... NPS is very useful speed indicator if you know >what has changed in your program. But since we don't know about others' programs >it is useless to compare NPS figures. It is critical to know what changes you >made to your program and then judge if the NPS increase, depth increase or >whatever was worth it. > >Severi I agree with all the above. One big problem is that many improvements will for example get you twice the ply/nodes but a slightly worse performance in positional test suits. What now , thats the question. One of the few good solutions is to eat the pudding. Georg
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.