Author: Albert Silver
Date: 10:55:39 09/22/04
This is old news to the engine programmers here, and I knew it in theory as
well, but it's a funny thing having to sacrifice some tactical acuity to get....
better play. In my tinkering with the manifold switches Ed has put at the Pro
Deo users' disposition, I have managed to get a group of settings that clobbered
Manfred Meiler's WeltMeister Test in the King Attack part. The test consists of
100 positions covering King Attacks, Positional play, and Endings. The King
Attack part consists of 38 easy-to-tough positions and there is a large Excel
file with the results for a number of top amateurs and pretty much all of the
commercial programs. The best result, all on an Athlon 1400, has been 32
positions found. Pro Deo had 26. And even though test suites are obviously not
the end-all of end-alls, they do help provide a minimum of guidance.
Well, I first managed to configure the settings for Pro Deo to solve as many as
32 positions, and was quite pleased.... until I saw it play. It played like a
sheer lunatic, and was losing badly. Lesson learned: brilliant test suites do
not mean brilliant play. not even decent play in fact. I then did some more
tinkering, being very careful to keep the parameters on relatively normal
values, and managed a record 34 positions solved (on an Athlon 2400XP+ BTW), and
within an inch (1 ply) of the remaining 4. I still kept my cool and watched it
play some blitzes. It was doing very well, but played some strange positional
moves at times, and then I saw this one:
[D]r2bk2r/2p1npp1/pp2b2p/P1p1P3/5BP1/2N2N1P/1PP2PK1/R2R4 b - - 0 20
My settings played bxa5, and I thought "Oh yuck!". I checked and it played this
not only at 11 plies deep, but even as far as 13. It actually won that game
thanks to its tactics, but still...
Anyhow, jacking up the value of Weak Pawns from 115 to 117, solved this. Some
quick tests say that this minute change also hurt its results in some positions
of the test (and improving a couple too), but the games it played after are far
better. That's ok. I can deal with that far easier than watching it play things
like bxa5... :-)
Albert
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.