Author: Steven Edwards
Date: 02:54:31 08/24/03
Go up one level in this thread
On August 24, 2003 at 05:01:49, Renze Steenhuisen wrote:
>I want to know why the best moves in test suites are said to be THE best moves.
>We know that the game of chess isn't solved yet, but it seems that the positions
>in these test suits already know what is best! That's weird, so how did these
>best moves become THE best moves?
>
>[test suites: I mean the wel known "Bratko/Kopec", "Win at Chess", etc]
I'll take part of the blame for this.
When I was designing EPD, I assigned the EPD opcode mnemonic set using
relatively short identifiers in the style of an assembler for a long word
microcoded architecture. And I wanted some degree of consistency; see:
pm: predicted move
pv: predicted variation
sm: supplied move
sv: supplied variation
Et cetera. Using EPD for test suites meant having some way to define the
solution set, so I picked "bm" for "best move(s)" and "am" for "avoid move(s)"
as easy to remember names. So, when you see the "bm" mnemonic in an EPD record,
it really means
"The set of one or more moves presented in Standard Algebraic Notation
sorted in ASCII order that are legal in the specified position as solution moves
by the player or the composer of the position."
The convention used for testing a chess playing program is that for every EPD
record processed, the program gets one point if its predicted move (pm) matches
any of the best moves (bm) while the program loses one point if its predicted
move (pm) matches any of the avoid moves (am) and gets zero points for any other
response.
Since the vast majority of chess positions are unsolved, test suites may have
theoretically incorrect values for "bm" and "am". When it becomes obvious that
a particular EPD record in a test suite is flawed in this way, that record in
edited (correctly, one hopes) and the so the suite is revised.
What may be needed are persons with appropriate skills and resources to take
upon themselves the responsibilites to be long term "Keepers of the Test Suites"
and provide periodic authoritative review and revisions.
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.