Author: blass uri
Date: 06:43:21 08/22/98
Go up one level in this thread
On August 22, 1998 at 07:07:33, fca wrote: >fca hereby defines the > >"Inherent Strength Of A Chess Engine" > >as being the ELO rating of that engine computed as normal when it is >operating in the following conditions, all of which must apply: > >(a) Real games (not problem sets!) at Time control 40/2 (however little >may be actually used by the engine); and >(b) No opening book or endgame tablebases; and >(c) Only ply ZERO evaluation allowed i.e. no searching or look-ahead OR >deeper tree-searching WHATSOEVER; just static position evaluation >(including pattern recognition, if any) of the root position. The >engine is permitted after the first pass to spend more time on any root >positions it wishes - but no looking further than the root. > >Why do it? > >It just may be a useful measure. The knowledge-based systems would be >expected to do better... do they?? This would tell. And more too. It >might indicate ways forward. I have many thoughts here. > >Some pretty obvious points: > >1. It does not matter what hardware is used - even emulators (!) - as >the computer will take much less than 3 minutes a move (maybe 3 >milliseconds per move for some combinations!). >2. Probably some programmers use this technique to judge improvements or >disadvantages between engine versions. >3. Of course the engine was written as it was to be a part of a whole >i.e. including an alpha-beta based look-ahead system, and it is pretty >artificial and anti-synergistic to pluck one part out of this >environment. That does not mean no useful conclusions can be drawn from >this. >4. Search depth = 1 (or 0) ply and Permanent brain, Book and Tablebase >off may produce the "hobbled" environment needed in most programs. > >What "Inherent Strengths" are likely to emerge? I did some tests vs a >fully enabled program, but the grading difference was too great. I'll >dig out my 386 and run some old programs on it, vs a hobbled engine. My >guess is 1500 ELO for Fritz5. Looking subjectively at the quality of >the moves played, Chess Genius seems to do *very* well in comparison >with other programs... (yes, some correlation with its super-fast blitz >performance). > >Views and results? 1)Did you use selective search 0 in chess Genius? 2)I think it is not important to do it because you cannot know if the evaluation you call evaluation without search is not based on some search and you do not know if the program use the same evaluation function in the root and in the leaves of the tree(I am sure It is not the same for fritz5 and many programs). Uri > >Kind regards > >fca
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.