Page 3 of 4

Re: SF130118

Posted: Tue Jan 23, 2018 8:07 pm
by jhellis3
It is difficult to say exactly how it will scale with regard to opposing engine strength. On the one hand, avoiding draws is more likely to pay off (as opposing engine strength decreases)... but on the other hand, you are already going to be winning most of those games anyway, so there are fewer opportunities for a given number of games for it to come into play.

Re: SF130118

Posted: Wed Jan 24, 2018 5:01 am
by JJJ
Yep, regression test showing a nice elo jump.

Re: SF130118

Posted: Wed Jan 24, 2018 5:47 pm
by IWB
I am running the same test again with 230118:

http://www.inwoba.de/C20.html

You can find the full rating for 130118 here in the thread. 130118 got 81%. I am quite sure it will be more with the contempt but the interesting question is if there is a performance boost against Houdin and probably Komodo and if not if that is pushing the contenders upward ...
130118 was only 3 Elo ahead of Houdini at the end ... I am courious to see the final ORDO calculation!

Ingo

Re: SF130118

Posted: Wed Jan 24, 2018 6:11 pm
by pohl4711
IWB wrote:I am running the same test again with 230118:

http://www.inwoba.de/C20.html


Ingo
The link leads only to an empty page ?!?!

Re: SF130118

Posted: Wed Jan 24, 2018 6:13 pm
by IWB
pohl4711 wrote:
IWB wrote:I am running the same test again with 230118:

http://www.inwoba.de/C20.html


Ingo
The link leads only to an empty page ?!?!
Works for me ...

Re: SF130118

Posted: Wed Jan 24, 2018 6:14 pm
by pohl4711
Yes. Now it works.

Re: SF130118

Posted: Wed Jan 24, 2018 6:16 pm
by IWB
The "old" 130118 had 56% vs Houdini and 60% vs komodo ... but that were only 220 games ... anyhow, it might look good in the rating lists but C20 might fail on a server ...

the old individual result is still online:
http://www.inwoba.de/SF.html

Re: SF130118

Posted: Wed Jan 24, 2018 7:18 pm
by JJJ
Too soon to tell, but it seems to hurt only vs Houdini so far.

Re: SF130118

Posted: Wed Jan 24, 2018 9:22 pm
by Vinvin
IWB wrote:I am running the same test again with 230118:

http://www.inwoba.de/C20.html

You can find the full rating for 130118 here in the thread. 130118 got 81%. I am quite sure it will be more with the contempt but the interesting question is if there is a performance boost against Houdin and probably Komodo and if not if that is pushing the contenders upward ...
130118 was only 3 Elo ahead of Houdini at the end ... I am courious to see the final ORDO calculation!

Ingo
Thanks for this test !
Very nice start for C=20 ! Currently, it's around 21 points (+3%) above Houdini !

Code: Select all

Stockfish 230118 64 POPCNT C20 : 447.5	-	82.5		84.43%

Stockfish 130118 64 POPCNT    :   2672.0	-	628.0	80.97%
Houdini 6.02        :      81.4%

Re: SF130118

Posted: Wed Jan 24, 2018 10:59 pm
by Vizvezdenec
At fishtest C=20 had shown 0 measurable regression, so it probably should not result in any meaningful regression vs houdini or komodo (also it didn't in some 1000-games samples of other testers).
Right now it scores slightly more vs houdini and slightly less vs komodo but I guess it's just a noise. But it for sure helps vs weaker engines a lot.