Author: Sandro Necchi
Date: 06:17:39 03/26/05
Go up one level in this thread
On March 25, 2005 at 12:17:47, William Penn wrote: >On March 24, 2005 at 17:09:22, Sandro Necchi wrote: > >>On March 24, 2005 at 16:40:19, Rajen Gupta wrote: >> >>>somebody, i think it was Sandro Necchi who had posted earlier that although the >>>engine is identical, the CB an UCI versions have different learn functions. >>> >>>could they elaborate; in particular which of these has a better learning >>>function >> >>I believe UCI has a better learning function than CB. >> >>Sandro > >In my many tests with the Shredder9 UCI version which uses a new .pl2 file for >learning, I can't find any evidence of reliable learning! It just doesn't work, >usually. > >For example, analyze a position for 1 hour in infinite mode, which should give >it quite a bit of learning. Right? > >Then exit the software, reset the computer, reload the software, and bring the >position back up for infinite analysis. It should now benefit from the prior >analysis. Right? No, the learning is linked also with the game score, so it is based only on played games. Sandro > >In only about 1 test position in 5 (20% of the time), it remembers the prior >long analysis OK. In about 4 test positions in 5 (80%), it remembers absolutely >nothing, zero, zilch. >WP
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.