Author: Uri Blass
Date: 02:21:31 03/12/04
Go up one level in this thread
On March 12, 2004 at 05:02:28, Fabien Letouzey wrote: >On March 12, 2004 at 02:04:31, Uri Blass wrote: > >>I understand but I try to explain what is the reason that fruit get good >>results inspite of having almost no evaluation(it has only piece square table >>and few boolean features and has for example no king safety evaluation and no >>endgame knowledge) > >Maybe we should wait for more time before claiming any "good results". >They are not apparent at all to me. >Any pointer? > >>The fact that based on the results that I read if you add to fruit book it >seems >>to be only something like 100 elo weaker than Crafty is surprising and suggest >>that something should be improved in the search of Crafty. > >Where does this estimate come from? There was one result that can give even better estimate for fruit than 100 elo difference. see http://www.talkchess.com/forums/1/message.html?353818 I know there were not enough games and there were also worse result but my impression based on the results that I read was that it is near 100 elo weaker than Crafty. >I haven't tested matches against Crafty because I considered it a loss of time >(on my computer); to me Crafty is obviously at the very least 200 elo points >stronger. > >>Maybe you need to do other changes in the search before ETC works for you(for >>examples checks in the first ply of the qsearch and changing that are done by >>fruit and changing R=2/3 to R=3 after having checks in the first ply of the >>qsearch). > >Uri, > >Read Crafty's main.c (yes you have done so, so read it again). >Bob has tried all that, and discarded them after testing. > I remember that Bob tried checks in all plies of the qsearch and rejected it. I remember that in a discussion (about year ago) he admitted that he did not try checks only in the first ply of the qsearch. Note also that it is possible to reject an idea not because the idea is bad but because some bug in the implementation so the fact that he rejected them does not prove that they cannot help crafty. 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.