Computer Chess Club Archives


Search

Terms

Messages

Subject: Puzzled about testsuites

Author: Michel Langeveld

Date: 08:28:16 03/09/04


Last week I worked hard on Nullmover and the ecm-gcp testsuite in particulair.
I thought this testsuite is a fast way to the tune middlegame knowledge, kick
things out and get things back in and do multiple testruns.

I did the following:
*1 kicked my kingsafety totally and use only pawnshelter
*2 kicked my mobility totally out
*3 added more information in the pawn struct

*1 gives me +- 10% performance back
*2 gives me +- 10% performance back
*3 gives me +- 5% extra performance.

I know that nps is not important but on a P3-800Mhz most I get most of the time
around 450-500knps. On a PIV 2.4Ghz machine I have most of the time
1000-1100knps.

I tested:
- testsuite: ecm-gcp.epd
- timecontrol: 5 seconds
- hardware: P3 800Mhz machine
- hashtable: 50Mb

P3 800: The old nullmover (v0.25) scored 58 positions right in 5 seconds.
P3 800: The new nullmover (v0.25c) scores 75!! positions right in 5 seconds.

I also tested on 10 seconds a move and there it is also an improvement:

P3 800: The old Nullmover (v0.25) scores 78 positions right in 5 seconds.
P3 800: The new Nullmover (v0.25c) scores 95!! positions right in 10 seconds.

I made also another testing mechanism that emulates a game. It skips the opening
and I use this to try the hashing algorithm:

HASHTEST 10
Test01:  3080962 time= 6.31 FH= 96.6% hhit%:  6.1% a4xb6=27
Test02:  4408305 time= 8.92 FH= 94.4% hhit%:  7.1% f4xe5=23
Test03: 11364980 time=21.20 FH= 92.5% hhit%:  5.8% o-o=-4
Test04:  5378200 time= 9.73 FH= 93.4% hhit%:  6.3% h2-h3=25
Test05:  2149098 time= 4.02 FH= 95.3% hhit%:  7.2% d5-b3=26
Test06:  9823962 time=17.91 FH= 88.9% hhit%:  5.8% h2-h3=26
Test07:  2493285 time= 4.39 FH= 94.0% hhit%:  6.7% c1-g5=24
Test08:  2344532 time= 4.05 FH= 96.4% hhit%:  6.4% g5xf6=49
Test09:  5186665 time= 9.06 FH= 90.6% hhit%:  6.8% h2-h3=50
Test10:  2473415 time= 4.36 FH= 92.0% hhit%:  8.6% h2-h3=50
Test11:  1252083 time= 2.50 FH= 88.5% hhit%:  8.2% e2xf3=52
Test12:  4202884 time= 7.72 FH= 87.8% hhit%:  8.2% f3-h5=58
Test13:  4188662 time= 8.89 FH= 84.5% hhit%:  8.7% f3-h5=62
Test14:  2227369 time= 4.19 FH= 86.3% hhit%:  8.7% f2-c2=52
Test15:  3483631 time= 6.33 FH= 87.6% hhit%: 10.8% a2-a4=38
Test16:  2648911 time= 4.84 FH= 87.5% hhit%: 10.3% a2-a4=21
Test17:  2582746 time= 4.61 FH= 87.4% hhit%: 10.7% a2-a4=25
Test18:  1975938 time= 3.58 FH= 90.0% hhit%: 11.4% f2-e3=24
Test19:  2570348 time= 4.53 FH= 89.5% hhit%: 12.2% a2-a4=13
Test20:  2786429 time= 5.20 FH= 91.6% hhit%: 14.7% g3-e3=7
Test21:  1312456 time= 2.61 FH= 88.8% hhit%: 14.7% f2-h4=-1
Test22:  2294598 time= 4.26 FH= 87.2% hhit%: 13.5% a2-a4=4
Test23:  3625229 time= 6.95 FH= 87.3% hhit%:  9.7% e1-e2=5
Test24:  1322411 time= 2.72 FH= 88.5% hhit%: 10.5% e1-e3=-1
Test25:   543736 time= 1.33 FH= 93.3% hhit%:  9.0% e3-f3=-5
Test26:  2453827 time= 4.95 FH= 89.9% hhit%:  8.2% g1-h1=5
Test27:  2146519 time= 4.33 FH= 91.8% hhit%:  8.4% d5-a2=5
Test28:  1233153 time= 2.56 FH= 91.3% hhit%: 10.2% h2-f2=24
Test29:   560284 time= 1.25 FH= 89.7% hhit%: 13.1% g1xf2=22
Test30:  1673102 time= 3.27 FH= 85.9% hhit%: 11.1% d1-b1=37
Test31:   651264 time= 1.45 FH= 83.0% hhit%: 14.0% a2-d5=48
Test32:   491891 time= 1.16 FH= 84.7% hhit%: 13.3% g1-g2=51
Test33:  1812927 time= 3.36 FH= 87.0% hhit%: 11.8% g1-f2=53
Test34:   535576 time= 1.17 FH= 87.8% hhit%: 10.9% c3xd4=44
Test35:   783155 time= 1.83 FH= 91.2% hhit%: 15.0% e1-c1=70
Test36:   610921 time= 1.41 FH= 95.5% hhit%: 10.8% h3-h2=75
Test37:   618524 time= 1.38 FH= 92.3% hhit%: 10.6% h2-c2=80
Test38:   699858 time= 1.55 FH= 94.1% hhit%: 13.1% c1-d1=92
Test39:   488832 time= 1.11 FH= 96.7% hhit%:  9.5% c2xc3=324
Test40:   118758 time= 0.42 FH= 97.0% hhit%: 12.2% c1xc3=408
Test41:   180520 time= 0.56 FH= 94.9% hhit%: 14.7% c3-c7=409
Test42:   189391 time= 0.58 FH= 96.0% hhit%: 12.4% c7xf7=535
Test43:   102407 time= 0.41 FH= 93.1% hhit%: 13.1% f7xf6=591
Test44:   101539 time= 0.44 FH= 90.6% hhit%: 15.6% f6xd6=611
TOTAAL: 101173283 time=193.39 nps=523157

I expected a big elo improvement with this new version when playing comp vs comp
games.

The opposite happened. :-( The new version scores 122 elo!! worse. I tested this
on the same hardware (P3 800Mhz) with a timecontrol 40 moves / 5 minutes games
as the older version.

26 Nullmover 0.25 : 2019   58  50   148    50.3 %   2017   14.2 %
65 Nullmover 0.25c: 1897  115  85    44    30.7 %   2039   20.5 %

I should play more games to have more more reliable elo rating but this is not
what I expected. And it looks already so worse that futher testing doesn't make
sense.

I am very puzzled. Somewhere I should have made a mistake. Maybe my approach for
tuning on ecmgcp.epd was wrong.

My questions:
* Is 5 seconds a move a good timecontrol to test ecmgcp.epd ?
* Does it makes sense to use a reall long timecontrol?
* I learnt that ecmgcp.epd is much more tactical as strategical. Does someone
has other/ better positional/eval-tweaking advice? No testsuites? Other
testsuite, like lct2.epd, the comeplete ecm.epd ?



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.