Houdini 2.0b T4 settings

[Tennison]

Here is a partial result for T4 version :

Code: Select all

Games Completed = 602 of 1000 (Avg game length = 30.754 sec)					
Settings = Gauntlet/64MB/500ms+200ms/M 500cp for 12 moves, D 150 moves/EPD:openings.epd(4000)					
Time = 18938 sec elapsed, 12520 sec remaining					
 1.   Houdini 2.0b x64        	283.5/602	181-216-205  	(L: m=6 t=13 i=0 a=197)	(D: r=38 i=57 f=49 s=2 a=59)	(tpm=189.8 d=14.0 nps=2598893)
 2.   Houdini 2.0b z x64      	149.5/301	102-104-95  	(L: m=4 t=2 i=0 a=98)	(D: r=19 i=27 f=18 s=0 a=31)	(tpm=188.9 d=13.7 nps=2581033)
 3.   Houdini 2.0b T4 x64      	169.0/301	114-77-110  	(L: m=3 t=0 i=0 a=74)	(D: r=19 i=30 f=31 s=2 a=28)	(tpm=189.6 d=14.1 nps=2610196)

Houdini 2.0b z : 49,67% against Houdini 2.0b STD

Houdini 2.0b T4 : 56,15% against Houdini 2.0b STD

But it's on very fast time control.

[Tennison]

If somebody can make other tests to confirm and help me to fix the values, it would be nice.

Maybe I have to slightly tuned the values depending the results ...

Houdini 2.0b T4 settings :

Pawn Shield Outer File = 36
Pawn Shield Inner File = 45
Pawn Shield Middle File = 54
Pawn Shield Diagonal = 32

Pawn Storm Outer File = 40
Pawn Storm Inner File = 50
Pawn Storm Middle File = 60

No changes on pieces' values.

Key value = 5D717974702A2020202020202020202020202020 20202020202020202020242D362028323C

[jpqy]

Sure Ben..i will try them out..Thanks!

JP.

[Houdini]

But it's on very fast time control.

More importantly, it's not enough games. With 300 or 500 games you're just observing random variations.
If you really want to "tune" anything, you should play 8000 or 16000 games.

[Tennison]

Hello Robert,

Happy to read you.

I agree completely ...

But it need lot of time and it's why I ask if somebody wants to help me testing ...

It's too why I said it's partial.

But I think it's a pretty good beginning. Better than if it scores only 40%

I don't want to say I understand anything in your job. I just try sometimes different values just for fun.

[Tennison]

Code: Select all

Games Completed = 1000 of 1000 (Avg game length = 30.699 sec)
Settings = Gauntlet/64MB/500ms+200ms/M 500cp for 12 moves, D 150 moves/EPD:openings.epd(4000)
Time = 33437 sec elapsed, 0 sec remaining
 1.   Houdini 2.0b x64        	479.0/1000	301-343-356  	(L: m=8 t=25 i=0 a=310)	(D: r=67 i=96 f=88 s=2 a=103)	(tpm=190.2 d=14.1 nps=2623789)
 2.   Houdini 2.0b z x64      	252.0/500	166-162-172  	(L: m=6 t=6 i=0 a=150)	(D: r=38 i=50 f=37 s=0 a=47)	(tpm=189.8 d=13.9 nps=2608280)
 3.   Houdini 2.0b T4 x64      	269.0/500	177-139-184  	(L: m=5 t=6 i=0 a=128)	(D: r=29 i=46 f=51 s=2 a=56)	(tpm=189.8 d=14.1 nps=2633691)

Final results of this first test with T4 settings :

Houdini 2.0b (z values) scores 50,4 %

Houdini 2.0b T4 scores 53,8%

I need again lot of games at different time controls to conclude anything.

Now I will start a 10" match for test 2.

[Houdini]

But I think it's a pretty good beginning. Better than if it scores only 40%

Ben, some good improvements start off with 45% after 300 games. Some bad ideas score 55% after 300 games. The whole point being that with so few games the random variations are 5 to 10 times larger than the expected strength difference between the two versions.

Playing 8000 games is not beyond reach of standard hardware. At 30 seconds/game you can play over 10,000 games/day using 4 cores.

[Tennison]

The games

[Tennison]

At 30 seconds/game you can play over 10,000 games/day using 4 cores.

Not on my computer ...

I can play +/- 2000 games per day (if I don't need my computer that day otherwise I have to stop the test and restart after).

It's why some testers should be nice ... lol

[Sedat Canbaz]

But it's on very fast time control.

More importantly, it's not enough games. With 300 or 500 games you're just observing random variations.
If you really want to "tune" anything, you should play 8000 or 16000 games.

Hello dear Robert,

Agreed that more games is better

But i hardly doubt that current SCCT Rating List's standings will be changed a lot ...e.g after 10.000 games per player
Btw,Robert Hyatt says 30.000 games per player is needed,but unfortunately i could not find a such rating so far

Why i believe that the SCCT standings will be not changed a lot,due to all engines use same opening book
In case of such testing (10.000 games per player or more),there is BIG possibility to see many double or almost same/similar games
I think those double games will appear, due to Perfect 2012b does not include thousands of lines for playing

Of course if we will use thousands (e.g 1000-3000 positions or more) of variety openings,then we need at least 10.000 games per player
Because such variety openings include a lot of holes,many engines will loose due to weak openings...

One thing more,we should concentrate more on most popular time controls (e.g 3min;5min;3m+2s;4m+2s;5m+3s;15m+10s;25m+10s;60m+10s;40/10;40/40;40/120...)



Best Wishes,
Sedat