Author: martin fierz
Date: 08:33:24 03/30/04
Go up one level in this thread
On March 30, 2004 at 01:11:41, Dave Gomboc wrote: >On March 29, 2004 at 06:25:28, martin fierz wrote: > >>without knowing too well: i don't think phoenix is a very good chess program. i >>would take schaeffer's statement on this with a grain of salt. many things that >>are true for one program are not true for other programs. i would tend to be >>more interested in experiments done by authors of really strong programs. > >Phoenix tied for first in the 1986 WCCC, and of the four tied entrants, Phoenix >had the weakest hardware. By 2004 standards, of course Phoenix isn't a good >chess program, but back when it was competing it was quite reasonable. > >It's certainly possible to argue that with today's hardware and software >improvements, the history heuristic is no longer a win. This can be done >without disparaging the strength of Phoenix relative to its competitors, though. i don't think i was disparaging. IIRC schaeffer writes about phoenix in "one jump ahead" that it was not a top program despite huge efforts on his part - and about his frustrations with it. this was one of the reasons he moved to checkers. that was after 1986 of course, because chinook started '89 i think. > IIRC, the history heuristic was an important contribution within Schaeffer's >(1986) Ph.D. thesis -- which would not have been the case if the technique >wasn't an improvement on what was known at the time. the history heuristic is a great idea: relatively simple, very useful, and AFAIK domain independent. nothing wrong with that... >I think many people would be interested in experiments done by authors of very >strong current programs, but those seem to be in short supply... i know. all i'm saying is that people should maybe not take "ancient" papers to be "the truth". in this sense, computer chess is a very rapidly changing field; things that might have been true 10 years ago might not be true any more today, whereas in other fields things are more stable. as search depth increases over time simply due to hardware improvements, things which once worked no longer work and vice versa (de koning just posted something that he's been removing extensions over the years in the king, as an example of this - and as an example of a top engine doing such things). i would consider papers published in pre-nullmove days to be suspect too. at the very least, papers reporting findings in weaker engines or older papers should be verified before relying on them. i didn't mean to target schaeffer in particular here - it's just that his came up. all papers older than N years for example would also qualify (N~10). or, for another example, i also wouldn't just believe what omid wrote about verified null-move pruning, because he implemented that in a weak engine (genesis). such a paper would be considerably more valuable if verified null-move pruning had been implemented in falcon for instance (it probably is). cheers martin
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.