Author: Don Dailey
Date: 11:36:53 01/27/99
Go up one level in this thread
On January 26, 1999 at 18:00:04, Bruce Moreland wrote: > >On January 26, 1999 at 16:29:15, Dann Corbit wrote: > >>Darn, left out one more key question: >> >>How *repeatable* are these tests? In other words, with 20 trials do you get the >>same answers each time? >> >>And yet another: >>What opening books are being used? Is B.I.'s opening book identical to that >>used during the contest? Is the version of B.I. used for these tests identical >>to that used in the contest (or are you simply using moves from that contest)? >> >>Do we know when B.I. went out of book? It can probably be deduced from time >>spent, if we have exact timings for the moves (which I doubt is available). >> >>I will again assert also, that similarity of play is a doubtful measure of the >>number of lines of code that are original. Suppose that we had 100% agreement >>on each and every move. Would that prove that the code was the same? Suppose >>that we had only 30% agreement. How would we know whether or not they simply >>changed the value of a knight by .3 and a pawn by .05 in eval? >> >>I think the only way you can really know is to see the code. > >I'm starting from move 20 in each game, which may be in book in some cases. >This test isn't perfect. If someone wants to devise a perfect test, I'd be >happy to see the FENs and report results to someone who wants to compile the >data. > >I haven't tried to figure out what each hypothetical outcome would imply. > >bruce Ok let's do the test according to your specifications. They main thing is that we are organized and accurately report the conditions so this is certainly a good start. - Don
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.