Author: Robert Hyatt
Date: 10:50:06 05/04/04
Go up one level in this thread
BTW there is +plenty+ of things in the DTS paper that could be criticized. I expected that. It is easier to test on a set of random positions to a fixed depth, as opposed to trying to answer the question I tried to answer in that paper "I've seen your speedup numbers on positions like the kopec positions, the Nolot positions, etc. But what is your speedup in a real game?" That is _tough_ to answer. And it produces a problem in that others can't compare directly to it since the positions tested were positions my program "liked" since it actually played the moves in the real game. I expected that as the referees had comments about it. But flawed or not, I could not think of any _other_ way to answer the question. But "fabrication" is out. The speedup numbers were computed exactly the same way any other speedup numbers you will ever see from me: speedup = 1cpu-time / Ncpu-time nothing more or less will do... Whether they should be reported as xx.x or xx.xx or xx.xxx is another topic. I chose xx.x....
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.