Author: Uri Blass
Date: 10:17:10 01/15/06
Go up one level in this thread
On January 15, 2006 at 09:15:13, Rolf Tueschen wrote: >On January 15, 2006 at 08:50:20, Uri Blass wrote: > >>On January 15, 2006 at 07:58:12, Marc Lacrosse wrote: >> >>>On January 15, 2006 at 04:56:15, Uri Blass wrote: >>> >>>> >>>>Your experiment show nothing >>>> >>>Mmmm. >>>I agree my method is not perfect but I think it is not without any value. >>>> >>>>My point is that you cannot compare number of solution in x seconds with number >>>>of solutions in y seconds and get conclusions. >>> >>>Your counterexemple is OK but does not prove that this comparison has no sense >>>in any other more usual cases. >>> >>>> >>>>The only logical comparison is comaparison of time to solve x solutions and time >>>>to solve y solutions and you did not do that comparison. >>>> >>>>Uri >>> >>>Why are you always so peremptory ? >>>I do not see anything in your post supporting the fact that your recommended >>>comparison should be the _only_ logical one. >>> >>>Moreover your clear-clut unproven affirmation is easy to refute. >>> >>>Just take this simple example : >>>Two engines A and B >>>100 positions to solve >>>Please plot the following values : >>> >>>solved positions time A time B >>> 0 00 00 >>> 10 10 05 >>> 20 21 13 >>> 30 33 24 >>> 40 46 38 >>> 50 60 55 >>> 60 75 75 >>> 70 91 98 >>> 80 108 124 >>> 90 126 153 >>> 100 145 185 >>> >>>If I take your criterion examining the time needed to solve 30 positions, engine >>>B (24 sec) is a faster searcher than engine A (33sec). >>>But If I take your criterion to see the time needed to solve 90 positions, >>>engine A (126sec) is faster than B (153 sec). >>>Where is the truth ? >>>It completely depends on your arbitrary choice of the time at which you compare >>>the engines. >>> >>>Moreover, you cannot compare engines whose strength is very different with a >>>single set of testpositions using your methodology whereas you can do it with >>>mine as each engine is compared to itself in mine. >>> >>> >>>So I surely would not say that your ideas are false but I feel you could be more >>>prudent when you affirm that yours are the one and only truth... >>> >>>Regards >>> >>>Marc >> >>Based on the data that you give in this post A probably improves more at long >>time control. > >Would you try to explain this conclusion? I know it's false! Note that probably is not definite conclusion but only a conjecture. Take the same type of improvement: improvement from solving 10 positions to solving 100 positions. For A the ratio of times is 145/10=14.5 For B the ratio of times is 185/5=37 It is clear that A improves more with more time in solving test suites. It does not prove that A imroves more than B in real games but it support the conjecture. Of course you can say that the better solver is the better player by this theory but my intuition tells me that it is less correct. The reason is that better result in test suites may be result of extensions that are counter productive in games but I do not expect these extensions to cause the program to be relatively faster in solving very hard problems. 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.