Author: Jeremiah Penery
Date: 13:57:05 01/06/00
Go up one level in this thread
On January 06, 2000 at 16:04:04, Dann Corbit wrote: >On January 06, 2000 at 15:43:25, Jeremiah Penery wrote: > >>On January 06, 2000 at 14:25:10, Dann Corbit wrote: >> >>>The data can be retrieved in a side-by-side format from: >>>ftp://38.168.214.175/pub/both.epd >>> >>>This is a comparison of data produced by PC's at very long time controls with >>>data produced by a supercomputer. >>> >>>This is a very significant and important step for C.A.P. (and I think it will >>>also be for chess theory). >> >>I'm a bit unclear on this...What exactly did you mean by "see how well it meshes >>with theory"? Do you mean chess theory or statistical theory? >Chess theory. Sometimes the C.A.P. data suggests a move that is known to be >inferior, especially in the opening section of games. Ah. >>And what is the >>current theory? >Good question. I hope that owners of the most recent ECO and related volumes >can look it up. The obvious problem is that these books often have plenty errors of their own. >>One thing that makes it difficult to compare the two sets is that the formatting >>of the output is so different. I can hardly read the second result of each >>pair, because of the weird move formatting. >It is non standard, but pretty obvious. The first letter of a move is the >piece, then the source square, then the target square. On many of the moves, there is another letter at the end of the move, like nc5e7P. I think this means a pawn was captured, but I haven't really looked through this yet to be sure. >>Also, it has no given search depth >>and is missing any other indicator of how deeply it was searched. >It was searched much more deeply than a PC can hope even to imagine. Oh, my. *boggle* *gasp* >>One other question I have is what program(s) the new data came from? >I am not at liberty to say. Can you say what type of computers were doing the searching? :) >>It appears that the supercomputer data gives a higher score for almost every >>position, which is strange. >There are both higher and lower evals. If you look at the side-by-side comparison from your FTP site, you will see that in almost every case, the supercomputer eval has a higher ce value. >>Is this just some subset of the total positions >>where this occurs, or is this indicative of the entire set? >It appears to me (at first blush) that the supercomputer evals are better. I hope so. I really hope I have the time to more closely examine this stuff soon, as I find it really interesting. >>Sorry for all the questions. :) Hopefully you can provide me with some >>enlightenment. :) Thanks for the answers. :)
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.