Author: Pete Galati
Date: 13:27:26 01/17/01
Go up one level in this thread
On January 17, 2001 at 15:54:35, Dann Corbit wrote: >On January 17, 2001 at 15:41:53, Pete Galati wrote: >>On January 17, 2001 at 13:22:02, Dann Corbit wrote: >[snip] >>>Better yet, analayze the entire trace of: >>>1. The games actually played >>>2. The suggested alternatives >>>3. The moves the engines suggest >> >>I don't completely understand. Are you talking about analyzing games as in >>Crafty's "annotate" command? If so, then what would you do with the >>annotations? > >No. Using OCD to translate the PGN into EPD. What is OCD? I know that SAN can spit out EPDs form a PGN file, but I don't know what OCD is. >Batch process the EPD at 12 minutes per record. I think Crafty and Comet can both test positions with a bat file, but I don't know who else can. I didn't think the comercial programs could do that. Actually, now that I think of it, I think there's actually a version of Nero that's specifically made for that (not that I'm expecting you to use it) >For the supposed traces from a given position, use Thomas Mooney III's >EpdProcessor to generate EPD for the suggested PV. >Batch process the EPD at 12 minutes per record. That sounds like something I've seen on your ftp, but I know nothing about it. > >If suggested moves do NOT follow the suggested PV or the game, then generate EPD >and analyse the suggested trace until termination at 12 minutes per position. > >>Also, if you have several different programs crunching numbers on the games and >>positions, how do you decide which one is correct? > >Look at the final outcome and look at how things turned out during the search. >For each suggestion by each program, follow the exact same steps as outlined >above. If it turns out that there are several winning trajectories beyond >refutation, then all of them are equally good. > >>>It would take about a month of effort with several computers to complete. But >>>when finished, the results would be a worthy test suite. >> >>If possible, it would be great if somebody here at CCC knew a highly rated GM >>that would be willing to sit down and provide opinions, partly because computer >>Chess programs can't be incorporating as much human instinct for the game as >>they'd like to. > >That would be ideal. Then computers could double check for tactical traps that >the GM's missed. Even without a GM's input, it sounds like there's a lot of potential for putting together a killer epd test suite. Pete > >>Anyhow..., this below may or may not illustrate what Uri is trying to say. From >>that test suite, pos 8, Comet in Chessbase instead of the Dos version, agreed >>with b4 for a while, and then said "screw that" and went and picked the lowest >>score Qc1! How could this be? > >Tactical snacking is what drives computers. Positional moves are the hardest >thing for computers. Of course, it may also be the case that there is a hidden >tactical flaw in the positional choice. Or the positional choice may require a >sacrifice at some point in the future that the computer really does not >understand. There are a thousand additional reasons the computer could be wrong >and the same number that it might be right and the GM wrong. Depth of analysis >is not sufficient unless a checkmate is proven.
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.