Author: Omid David
Date: 14:09:49 10/04/02
Go up one level in this thread
On October 04, 2002 at 16:49:56, Vincent Diepeveen wrote: >On October 04, 2002 at 14:54:44, Omid David wrote: > >>On October 04, 2002 at 14:21:16, Vincent Diepeveen wrote: >> >>>On October 04, 2002 at 11:48:06, Omid David wrote: >>> >>>How many times must i say the same thing? >>> >>>Suppose i have 200 patterns to adjust my score. >>> >>>Now 1 pattern can undo all those 200 patterns. >>> >>>I'm then either an idiot then to have those 200 patterns, >>>or to use that one pattern, if it is in 50% of the cases wrong. >>>Even if it is in 80% of the cases right :) >> >>But what of if in 100% of the cases it returns a correct result: > >Only EGTBs are 100% correct. Any other claim is a lie in itself. > >>either: >> >>- Yes, the game is draw => return 0; > >Only egtbs can do this or repetition detection. No. If a blockaded position is sent to the eval function, and the eval function (using certain heuristics) determines that it's undoubtedly a draw, it can safely return 0. > >Not blocked position code. the risk is too big in blocked positions with >material behind to evaluate it as a draw. Your thing will do insane efforts >to get from a little bit worse position into a position it considers 'blocked'. > >of course if all you want to is a 50% score at most against opponents there >are other ways working for you too. > >If this works for you, then multiplying your king safety with 10 works too >for the same goal. > >>- I don't know => do your regular search using your 200 patterns, etc. > >this isn't english, how must i parse the line? If your blockade detection doesn't detect a draw, just continue with your regular search, evalution, etc. > >>How can it possibly cause a problem? > >> >>> >>>>On October 04, 2002 at 11:34:39, Vincent Diepeveen wrote: >>>> >>>>>On October 04, 2002 at 11:26:51, Omid David wrote: >>>>> >>>>>slowing down the search is not the problem. the problem i >>>>>explained in email to sargon >>>> >>>>So I assume the problem is that it's not 100% reliable, and may return >>>>inaccurate estimates. Right? >>>> >>>> >>>>> >>>>>>On October 04, 2002 at 11:22:05, Vincent Diepeveen wrote: >>>>>> >>>>>>>On October 04, 2002 at 10:34:43, Omid David wrote: >>>>>>> >>>>>>>had fritz implemented such a scheme it would play 300 points weaker >>>>>>>anyway. >>>>>> >>>>>>It depends on how you implement it without slowing down the search >>>>>>significantly, and that's what I'm currently working on... >>>>>> >>>>>>It's obvious that applying the whole heuristic on each and every node is too >>>>>>expensive. But a wise application of it, could detect many draws without any >>>>>>significant impact on the total speed. >>>>>> >>>>>>Omid. >>>>>> >>>>>> >>>>>>> >>>>>>>>Just today we had a discussion regarding the importance of blockade draw >>>>>>>>detections http://www.talkchess.com/forums/1/message.html?255950 >>>>>>>> >>>>>>>>Had Fritz implemented such a scheme, it wouldn't have stepped into the drawing >>>>>>>>variation. >>>>>>>> >>>>>>>>My engine instantly announced draw based on its blocked-draw detection >>>>>>>>heuristic. While Fritz never realized the draw... >>>>>>>> >>>>>>>>Omid.
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.