Komodo wins TCEC

[shrapnel]

This (as well as the subscription idea) are both under discussion now

I think the subscription idea is really a good one.
Another benefit of this would be that this way there would be less of a mismatch between the latest Stockfish Developmental Version and the last released version of Komodo, which could sometimes get quite out-dated.
It would help the Testers too, to get a more balanced result, so to speak.....witness the lop-sided results of Beram Mourik, for instance.
This way, more people than just Graham Banks, IPman and the like could get hold of the latest Komodo Dev versions, and would be a suitable reward for the loyal followers of Komodo !

[beram]

This (as well as the subscription idea) are both under discussion now

I think the subscription idea is really a good one.
Another benefit of this would be that this way there would be less of a mismatch between the latest Stockfish Developmental Version and the last released version of Komodo, which could sometimes get quite out-dated.
It would help the Testers too, to get a more balanced result, so to speak.....witness the lop-sided results of Beram Mourik, for instance.
This way, more people than just Graham Banks, IPman and the like could get hold of the latest Komodo Dev versions, and would be a suitable reward for the loyal followers of Komodo !

The difference between TCEC Komodo 1331 and Komodo 8 is about 15-20 ELO according to Marl Leffler, as said by him in the TCEC chat. I wouldn't call that quite out-dated.
If Tom Casanovas or me could get a hand on latest Komodo 1339 as now tested by Ray Banks I think our testresult for latest Stockfish 221214 against that Komodo latest dev, would still be about 53-55%, as for now latest SF dev scores about 56-59% against Komodo 8

See for instance latest test result by Tom on this forum where SF 221214MZ scores in 480 games against Komodo 8 (3266) = 59.38% (!)
http://www.talkchess.com/forum/viewtopi ... 26&t=47512

[shrapnel]

Exact ELO difference only Komodo Team knows for sure.
Tom Casanovas was most likely using Default Drawscore -7, which is completely unsuitable against Stockfish.
In TCEC, Drawscore =0 was used.....and Stockfish got WHOMPED

[Sean Evans]

Exact ELO difference only Komodo Team knows for sure.

Bull-crap, anyone can do their own testing!

[Modern Times]

If Tom Casanovas or me could get a hand on latest Komodo 1339 as now tested by Ray Banks I think our testresult for latest Stockfish 221214 against that Komodo latest dev, would still be about 53-55%, as for now latest SF dev scores about 56-59% against Komodo 8

What are your test conditions ? I might have time to run a match here.

[lkaufman]

If Tom Casanovas or me could get a hand on latest Komodo 1339 as now tested by Ray Banks I think our testresult for latest Stockfish 221214 against that Komodo latest dev, would still be about 53-55%, as for now latest SF dev scores about 56-59% against Komodo 8

What are your test conditions ? I might have time to run a match here.

They primarily run blitz games, whereas you run games at reasonably "serious" time controls. Komodo is not (yet) as strong as current dev. Stockfish at blitz, but at longer time controls it's a different story, as the CEGT and CCRL rating lists make clear. The drawscore is also an issue, but in my opinion the longer time control is much more important for Komodo than using zero drawscore.

[shrapnel]

in my opinion the longer time control is much more important for Komodo than using zero drawscore.

Agreed.
But there is another aspect that should be considered, that is, there is quite possibly a relationship between the Drawscore being used and the actual length/type of LTC being played !.
Let me explain with examples.
Let us assume that Drawscore is set to -3.
In games on c2p, a maximum of 120 seconds is allowed for per move per player, so although it may qualify as an LTC game, anyone who uses this drawscore -3 has a good chance of being held to a draw or even being beaten by an opponent on equal or even slightly inferior hardware running Stockfish.
BUT, in Time Controls like the TCEC, Komodo is free to use MUCH MORE than just 2 minutes for a move, if the situation so warrants !
In such a scenario the slightly over-optimistic moves being considered due to the Drawscore -3, are on further thought, self-corrected by the Engine, so to speak !
So, in conclusion, in TCEC-like conditions, one might get away with a Drawscore of -3 and might even benefit ; but in slightly shorter LTCs a Drawscore of -3 may backfire.
This is the relationship I was talking about.

[beram]

If Tom Casanovas or me could get a hand on latest Komodo 1339 as now tested by Ray Banks I think our testresult for latest Stockfish 221214 against that Komodo latest dev, would still be about 53-55%, as for now latest SF dev scores about 56-59% against Komodo 8

What are your test conditions ? I might have time to run a match here.

Hi Ray,

you can find results and match conditions for my tests in topic links beneath
I can find no better results for K8 against SF dev, when going from TC3m2s to 15m10s to 30m20s
on both my systems I5 4200M 2cpu nor on AMD 1090T 4cpu
Perhaps you would like to do some same tests from bullet to rapid with Komodo 1339 against latest SFdev on your 8core AMD ?

http://www.talkchess.com/forum/viewtopi ... 33&t=53844

http://www.talkchess.com/forum/viewtopic.php?t=54632

[Modern Times]

Thanks I'll look. If I run anything it will be 6 cores on my AMD 1090T.

[vittyvirus]

Heck, I expected SF to win this time...

Anyway, congrats to the Komodo team!