Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Proposal: New testing methods for SSDF (1)

Author: Enrique Irazoqui

Date: 10:26:06 04/13/98

Go up one level in this thread


I agree about testing programs on equal platforms. Instead, I don't
think the rest is necessary. Programs should be considered as a whole:
books, engine, table bases. Learners, even better if in combination with
wide books, take care of cooks and double games. In my opinion this is
not an issue any more.

As for autoplayers, I have seen enough erratic behavior from auto232
games to consider it flawless and to accept it as the mandatory
standard.
- Programs overruling the autoplayer by playing longer games than
allowed by the /m parameter.
- Programs that terminate the game for no apparent reason.
- Programs that terminate the game and count it as a win when in their
evaluation the opponent is at - 5, forcing the opponent to resign.
- Programs that terminate a game when arbitrarily decide it's a double
game.

Tricks are possible even through a standard auto232 device. If programs
learned to take care of themselves regarding cooked lines, they should
do the same about possible autoplaying tricks.

In the end, and thanks to learners, engines decide the outcome of games
more now than before, because cooks and doubles in games played by new
programs are infrequent enough to be of statistic significance. So I
don't see the use of complicating matters in this area. Instead, I think
learners are a genuine improvement.

Enrique



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.