Author: Ed Schröder
Date: 02:13:14 09/14/00
Go up one level in this thread
On September 14, 2000 at 02:57:09, Dann Corbit wrote: >On September 14, 2000 at 02:17:58, Ed Schröder wrote: >[snip] >>IMO every game played in WCCC events is worth at least 10 autoplayer >>games. Authors are present to solve any problem that might occur, no >>book randomness, no learning involved, book preparation should ensure >>that the author's program should play those lines the program likes >>best. >Barring some errant codes sent by Winboard [as is alledged for some >autoplayers], I disagree completely. Then have a look at the last 3-5 WCCC's. If you look at the rankings they don't match with for instance the SSDF list. Especially Shredder comes to mind. Ed >The books used are those created by the >authors. The learning that goes on is the exact same learning that would go on >in normal play. If your program does not learn and the other does, then their >program's edge is one that they have earned. Special books cooked for a >tournament show the ability of the book preparation people and not the ability >of the engines. After a while, killer likes will be debugged by learners and >won't get played anymore by the opposition. > >>The WCCC is playing games under the most optimal conditions for chess >>programs. >> >>Autoplayer tournaments are a whole different world. >> >>Both are valuable but IMO are not comparable. > >Unless bugs are present in the automatic tournament managers, the data is just >as good as any hand run tournaments. Actually, since the errors introduced by >innacuracies of non-automatic move entry will cause the experiment to be hard to >reproduce, if anything such modes of play are inferior, from an experimental >standpoint. If this element of randomness is needed to prevent similar losing >lines from being played repeatedly, then (again) it is a program flaw. > >I have seen no convincing arguments that autoplayer games are inferior except >that invalid command sequences are possibly generated by some autoplayers. I >know of no complaints against Winboard in this regard. > >Furthermore, for Winboard programs (which is what I am testing) they are nearly >always going to be played using a Winboard interface. If played on the net >using an automatic mode (as most seem to do) the results will much more closely >mirror what will be achieved in practice.
This page took 0.01 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.