Author: Alvaro Polo
Date: 13:49:27 07/25/00
Go up one level in this thread
On July 25, 2000 at 16:22:08, blass uri wrote: >On July 25, 2000 at 15:24:04, Alvaro Polo wrote: > >>On July 25, 2000 at 06:23:51, Ralf Elvsén wrote: >> >>>On July 25, 2000 at 05:05:44, blass uri wrote: >>> >>>>> >>>>>Alvaro >>>> >>>>Your calculation is wrong because of diminishing return from speed. >>>> >>>>Uri >>> >>>Right or wrong belongs to pure mathematics. Here we need an estimation >>>of the uncertainty. If a result is in the right neighbourhood >>>it's usable. >>> >>>Ralf >> >>I am going to modify my "I am wrong" assessment. DBJ was making 750,000 nodes >>per search, and CT 375,000 nodes per search, but DBJ was using only 1 second and >>CT 22 secs per search. This difference compensates the weak CPU being used by >>CT. I hence believe that this is equivalent to DBJ against CT (under a powerful >>P3) if both were using the same time per search (DBJ using equal time >>compensates the P3-Pentium 150Mhz difference). Then the full DB, at 200Mnps >>rather than 750Knps would be about 560 Elo higher than CT on a modern machine, >>assuming that diminishing returns don't affect comp-comp matches, something, on >>the other hand, that has never been proven wrong. > >I learned that Diminishing return was proved in Fritz6-Fritz6 games. >I learned it from Ernst Heinz's post some weeks or some monthes ago. > >I also believe that other programs may earn more from time. > >possible reasons: >1)Deeper blue had bugs(see the first game when they lost against kasparov and >the game when they lost against Fritz3 thaks to a tactical error). > >2) Other programs have better evaluation function than Deeeper blue. > >3)Deeper blue used singular extensions and it may be counter productive when you >search very deep because one ply more in quiet positions may be more important. > >I also doubt if the 750,000 nodes per search is correct and I suspect that they >searched more nodes per search. > I also doubt this but I was using the number quoted by Christophe. Assuming the number of NPS is 1,500,000 (which appears closer to the truth) the Elo increment would be 490 Elo points (less diminishing returns, if there are any). >I believe that something that is 560 elo better than tiger in the ssdf rating >could win kasparov more convincingly. > This involves a comparaison between comp-comp and comp-human, an obscure matter. It could be possible that something 560 points stronger than Tiger, based on a deeper search, wouldn't be able to beat Kasparov more convincingly. At this moment we cannot know. >Deep Junior that is clearly weaker than 560 elo better got more than 2700 >performance against prepared players. > >The performance of Deep Junior against prepared players suggest that it is in >the same level of Deeper blue(the performance against kasparov was better but >being able to train against the program is a big advantage). > I agree with that. It would be interesting to know the SSDF rating of Deep Junior. But this connects with my last point. Deep Junior has a comparable performance against humans even though it has a shallower search than Deep Blue, because search depth is less determinant against humans than against computers. Alvaro > >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.