Author: Dann Corbit
Date: 18:30:22 02/16/06
Go up one level in this thread
On February 16, 2006 at 20:47:25, vladan wrote: >After some hot messages about Rybka’s NPS, I must say something (I was told it >also to my friend Djordje Vidanovic) two months ago. > >It is obvious that Rybka’s NPS is not programs real NPS. > >It is impossible that program with lowest NPS compared to other engines achieves >depths sand playing strengths above 2850 (or 2900) ELO points, the best of all! > >But, I am not sure that Vasik did purposely tune down real NPS. The only >explanation for that will be that he tried to hide Rybka’s origin in some other >public source engines. But Rybka’s plays much different and superior chess, >which implicates me that Vas however did original and good job with his engine. >Of course, as he said, he used ideas, maybe some source portions or solutions >from Fruit, but it is irrelevant if he added (and that is the case) his original >code somewhere, which amplifies playing strings for a major margin. It is not >cloning. If it is, all of us are monkey clones (with 95% same DNK with our >primate relatives :) > >Maybe, Vasik counts not all generated nodes, not only terminal (leaf) nodes >without quiescent generated ones. Each terminal node expands to 2-20 q-search >nodes, depended on q-search heuristics (if program has only captures, or checks, >promotions etc.). This presumption explains why Rybka’s NPS varieties and drops >significantly in positions where exsist much checks, or recaptures, and stay >stable at high level in strategically (quit) positions. It's a tempest in a teapot, anyway. The proof of the pudding is in the eating. Whether he processes 1 NPS or 1e9 NPS, the program plays like some child of Kasparov and Polgar.
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.