A Public Service Announcement

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
leavenfish
Posts: 282
Joined: Mon Sep 02, 2013 6:23 am

Re: A Public Service Announcement

Post by leavenfish » Sat Feb 17, 2018 4:50 pm

syzygy wrote:
Dann Corbit wrote:
syzygy wrote:
Dann Corbit wrote:Public service announcement number 2 (Look at the top graph):

http://www.sp-cc.de/


If you want the fastest and strongest Stockfish, it is almost certainly the Cfish variant.

Something bad happened to AsmFish. Or so it appears.
No, it's just that Stockfish has improved since November.
It has. But AsmFish used to maintain a 30 Elo lead over SF (for about a year or so). None of the recent tests show that. Not only Pohl's tests, but also CCRL tests, show that SF is now stronger than AsmFish. This is strange because the AF team moves all the improvements to SF forward to the new AsmFish versions. Why should the 30 Elo gap narrow to 0?
As I said, Stockfish has improved since November.

Did you look at the asmFish "version numbers"? 171107 means 7 November 2017, 171119 means 19 November 2017. Those versions match SF of a couple of weeks earlier.

The Stockfish/Brainfish versions it is tested against are from January and February 2018. They contain improvements that are not in asmFish.

If you run any version of asmFish against a corresponding version of Stockfish. the difference in strength will be explained exactly by the difference in speed. Because there is no other difference.
Ah but...when you say SF has 'improved' since November....do you perhaps mean that the Contempt is set to +20 now? That is to say, for engine vs engine play. If contempt was set back to zero, would there be any real improvement?

Damir
Posts: 2471
Joined: Mon Feb 11, 2008 2:53 pm
Location: Denmark
Full name: Damir Desevac

Re: A Public Service Announcement

Post by Damir » Sat Feb 17, 2018 4:59 pm

With Contempt +20, engine will take unnecessary high chances to try and win the game.... :) :) :D

syzygy
Posts: 4971
Joined: Tue Feb 28, 2012 10:56 pm

Re: A Public Service Announcement

Post by syzygy » Sat Feb 17, 2018 5:18 pm

leavenfish wrote:
syzygy wrote:
Dann Corbit wrote:
syzygy wrote:
Dann Corbit wrote:Public service announcement number 2 (Look at the top graph):

http://www.sp-cc.de/


If you want the fastest and strongest Stockfish, it is almost certainly the Cfish variant.

Something bad happened to AsmFish. Or so it appears.
No, it's just that Stockfish has improved since November.
It has. But AsmFish used to maintain a 30 Elo lead over SF (for about a year or so). None of the recent tests show that. Not only Pohl's tests, but also CCRL tests, show that SF is now stronger than AsmFish. This is strange because the AF team moves all the improvements to SF forward to the new AsmFish versions. Why should the 30 Elo gap narrow to 0?
As I said, Stockfish has improved since November.

Did you look at the asmFish "version numbers"? 171107 means 7 November 2017, 171119 means 19 November 2017. Those versions match SF of a couple of weeks earlier.

The Stockfish/Brainfish versions it is tested against are from January and February 2018. They contain improvements that are not in asmFish.

If you run any version of asmFish against a corresponding version of Stockfish. the difference in strength will be explained exactly by the difference in speed. Because there is no other difference.
Ah but...when you say SF has 'improved' since November....do you perhaps mean that the Contempt is set to +20 now?
No, I was not thinking of any particular change in particular. My point was just that each asmFish version is functionally identical to some Stockfish version, only faster. Any Elo advantage of asmFish is explained by the increased speed and nothing else. Any disappearance of the Elo advantage compared to newer versions of Stockfish is explained by improvements in Stockfish which have not yet made it into asmFish.

However, you have a very good point. asmFish does not yet include the new contempt implementation and the non-zero default contempt setting of Stockfish. If the engine is tested against weaker players, then this indeed makes a significant difference. Against an opponent roughly equal in strength, the non-zero contempt setting will help, at best, only a little.
That is to say, for engine vs engine play. If contempt was set back to zero, would there be any real improvement?
There were some patches that gained Elo since Novermber, so yes there is also real improvement.

Post Reply