Discussion of anything and everything relating to chess playing software and machines.
Moderators: hgm , Rebel , chrisw
Rebel
Posts: 6997 Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder
Post
by Rebel » Sat Sep 26, 2015 9:53 am
Code: Select all
ccrl 40/4 40/40 elo
Komodo 9.2 64-bit 4CPU 3406 3353 -53
Komodo 9.1 64-bit 4CPU 3375 3326 -49
Stockfish 6 64-bit 4CPU 3386 3311 -75
Stockfish 5 64-bit 4CPU 3362 3283 -79
Gull 3 64-bit 4CPU 3263 3198 -65
Gull 2.8b 64-bit 4CPU 3252 3194 -58
Based on this shorty, can you conclude Komodo scales best and that Stockfish from the 3 scales worst?
BBauer
Posts: 658 Joined: Wed Mar 08, 2006 8:58 pm
Post
by BBauer » Sat Sep 26, 2015 12:36 pm
Based on this shorty, can you conclude that shorter time control lead to higher rating?
kind regards
Bernhard
Rebel
Posts: 6997 Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder
Post
by Rebel » Sat Sep 26, 2015 12:41 pm
BBauer wrote: Based on this shorty, can you conclude that shorter time control lead to higher rating?
kind regards
Bernhard
That's for the CCRL guys to explain.
Rebel
Posts: 6997 Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder
Post
by Rebel » Sat Sep 26, 2015 12:43 pm
SzG wrote: Rebel wrote: Code: Select all
ccrl 40/4 40/40 elo
Komodo 9.2 64-bit 4CPU 3406 3353 -53
Komodo 9.1 64-bit 4CPU 3375 3326 -49
Stockfish 6 64-bit 4CPU 3386 3311 -75
Stockfish 5 64-bit 4CPU 3362 3283 -79
Gull 3 64-bit 4CPU 3263 3198 -65
Gull 2.8b 64-bit 4CPU 3252 3194 -58
Based on this shorty, can you conclude Komodo scales best and that Stockfish from the 3 scales worst?
I don't understand what you mean. Scaling is within one version, not between two, isn't it?
Komodo loss in elo -53 | -49
Gull -65 | -58
SF -75 | - 79
Gurcan Uckardes
Posts: 196 Joined: Wed Oct 29, 2014 12:42 am
Post
by Gurcan Uckardes » Sat Sep 26, 2015 11:14 pm
It's known Komodo scales better than SF. Ed mentioned the scaling vs time. There's also an SMP efficiency issue which favors Komodo.
But SF team looks like finding a cure soon. Check here:
http://www.fastgm.de/threads5.html
BubbaTough
Posts: 1154 Joined: Fri Jun 23, 2006 5:18 am
Post
by BubbaTough » Sat Sep 26, 2015 11:28 pm
Rebel wrote: Code: Select all
ccrl 40/4 40/40 elo
Komodo 9.2 64-bit 4CPU 3406 3353 -53
Komodo 9.1 64-bit 4CPU 3375 3326 -49
Stockfish 6 64-bit 4CPU 3386 3311 -75
Stockfish 5 64-bit 4CPU 3362 3283 -79
Gull 3 64-bit 4CPU 3263 3198 -65
Gull 2.8b 64-bit 4CPU 3252 3194 -58
Based on this shorty, can you conclude Komodo scales best and that Stockfish from the 3 scales worst?
Code: Select all
Hannibal 1.5 64-bit 4CPU 3153 3113 -40
Hannibal 1.4b 64-bit 4CPU 3113 3092 -21
Rebel
Posts: 6997 Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder
Post
by Rebel » Sat Sep 26, 2015 11:41 pm
BubbaTough wrote: Rebel wrote: Code: Select all
ccrl 40/4 40/40 elo
Komodo 9.2 64-bit 4CPU 3406 3353 -53
Komodo 9.1 64-bit 4CPU 3375 3326 -49
Stockfish 6 64-bit 4CPU 3386 3311 -75
Stockfish 5 64-bit 4CPU 3362 3283 -79
Gull 3 64-bit 4CPU 3263 3198 -65
Gull 2.8b 64-bit 4CPU 3252 3194 -58
Based on this shorty, can you conclude Komodo scales best and that Stockfish from the 3 scales worst?
Code: Select all
Hannibal 1.5 64-bit 4CPU 3153 3113 -40
Hannibal 1.4b 64-bit 4CPU 3113 3092 -21
Cool numbers.
Mine are in the 80's, the good old days so to say
I am currently trying improve here.
BubbaTough
Posts: 1154 Joined: Fri Jun 23, 2006 5:18 am
Post
by BubbaTough » Sat Sep 26, 2015 11:57 pm
Rebel wrote:
Cool numbers.
Mine are in the 80's, the good old days so to say
I am currently trying improve here.
I was chatting with Fabien about this kind of thing when Senpai came out. He noticed Senpai was better than Hannibal at fast games, but in long games Hannibal was better. Early on I found some search related stuff that hurt performance at most reasonable test time controls, but helped at longer games (like 5 minutes or over). This was very painful to confirm with any statistical significance, but I eventually did. I believe some categories of endgame knowledge can be like that as well. There is a lot to be said for the modern trend of fast test time controls, but I think there is still fertile ground to be explored in techniques whose benefits are hard to measure except in long games.