Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A proposed WAC replacement for testing

Author: Robert Hyatt

Date: 14:48:28 09/18/01

Go up one level in this thread


On September 18, 2001 at 13:05:08, Gian-Carlo Pascutto wrote:

>Like many others here, I use a standard testset
>for testing improvements and doing regression
>testing to my engine.
>
>So far WAC has been adequate, but as you all
>know it has two annoying problems, namely that
>over 2/3 of the problems are solved by 2 ply
>searches and hence useless, and the level of
>most of the others isn't very hard either with
>nowadays CPU's.
>


Here is what I would like to see in a test suite:

1.  tactical positions.  Not impossibly hard (like nolot positions) so that
tactics can be tested from time to time without taking 2 days to run the
test.  I want these positions to have an 'avoid move' or a 'best move', but
hopefully not several of either.  And I want the best move to be a move
that leads to a _significantly_ better score than any _other_ possible move.
And 'avoid moves' should lead to _significantly_ worse scores than any other
possible moves.  These positions should have some sort of goal, or at least be
constructed such that positional considerations can't lead to the right move
for the wrong reason.  IE winning a queen is one goal that can be verified by
the score.  Winning 1/2 of a pawn is not so obvious.

2.  positional positions.  These should be posions that are solvable without
requiring huge depth.  And they should have one right move that has clear
reasons for being the _only_ reasonable move (and it must not be a tactical
reason.)  IE xx leads to a won ending or yy gives black a weak pawn.  Too many
"positional" test cases have more than one best move which makes them worthless.
Some have tactical shots that also render them useless.

Some nice cases would be positions where (say) black has one move to make to
avoid a wrecked kingside.  Or a weak pawn.  Or else black has one move to do
that to white or else white will defend.

WAC is at least tactical, and I use it a lot because I can do the entire test
in a few minutes.  And all I care about is "did I break something, or did I
make something a bit better/faster?"  Positional test positions are harder to
come by...



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.