Author: Ernst A. Heinz
Date: 17:09:09 05/25/00
Go up one level in this thread
Hi Will, >Probably more defensive than aggressive. Different cultural background, I assume. >I'm more interested in your thinking on the control question. >Is a single control using equal depths sufficient? Sufficient for _what_? For the "proof of existence" purpose it is certainly sufficient. >Could the results be skewed by positional/tactical factors >present at 8-ply, which would be different at 9 or 10 ply (for the equal-depth >control)? Yes, they _could_ ... but how likely is the existence of such factors in a program as proven and thoroughly tested in practice at various depths and time controls as "Fritz 6"? The calibration match was mostly meant as a control setting for the _opening book_. Without the underlying assumption that the chess engine behaves "reasonably" with progressing search depths, the whole experimental setup does not work at all. (BTW, all previous self-play research did not even bother to do _any_ control matches of the form "X <=> X".) >I assume you were looking for such feedback? I basically wanted to share my new insights and results. The pros and cons of self-play matches are well-known already. They have been discussed extensively in the literature (please everybody, read all the related work referenced in my paper!) and in personal discussions with others. I briefly refer to the lingering doubts regarding the validity of self-play in the introduction. =Ernst=
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.