Komodo Dragon 2.5 released.

[Wolfgang]

CEGT 40/4 tests:
https://cegt.forumieren.com/t1580-testi ... -mcts#3057

[Jouni]

No wonder, that Dragon is closing gap to Stockfish. Larry: "...we trained primarily against Stockfish for this version... We used some ideas from Stockfish..." .

[AndrewGrant]

Larry: "...we trained primarily against Stockfish for this version... We used some ideas from Stockfish..."

Training against = Using Training Data Generated By (?). Boring.

[dkappe]

Larry: "...we trained primarily against Stockfish for this version... We used some ideas from Stockfish..."

Training against = Using Training Data Generated By (?). Boring.

You should have read the full post instead of this little snippet.

To Answer your question, we trained primarily against Stockfish for this version, but only trained on Komodo's moves and scores, not Stockfish moves or scores. We did not train against Lc0 yet, but you can expect us to do so.

So no training data generated by SF was used. The idea is that the network learns how to beat an opponent through this method of adversarial training.

Some experiments suggest that adversarial training is especially effective against mcts/nn engines like lc0. Time will tell.

[AndrewGrant]

Larry: "...we trained primarily against Stockfish for this version... We used some ideas from Stockfish..."

Training against = Using Training Data Generated By (?). Boring.

You should have read the full post instead of this little snippet.

That would be the point of my "(?)".

[BrendanJNorman]

Larry: "...we trained primarily against Stockfish for this version... We used some ideas from Stockfish..."

Training against = Using Training Data Generated By (?). Boring.

You should have read the full post instead of this little snippet.

That would be the point of my "(?)".

So why not give perhaps the strongest commercial engine ever - one that has a veteran GM actively working - the benefit of the doubt before implicitly accusing them of using Stockfish data and dismissing their work as "Boring"?

Why not use your immense intellectual gift for some positivity instead of always being salty?

How you respond to the rapidly changing computer chess ecosystem is how you'll be remembered.

Think about this. Your IQ seems to be very impressive, but your EQ lets you down.

[pohl4711]

KomodoChess has just released Dragon 2.5 at KomodoChess.com. Normally we would call this 2.1, but it is probably the
largest upgrade we have had between versions in many years, with the exception of the NNUE idea first implemented
in Dragon 1. It could easily be Dragon 3, but it's only been four months since Dragon 2 and we can't use that name
this soon contractually. It has a larger, "smarter", better trained net, and noticeably more search depth due to search
enhancements.

7000 games ratinglist-testrun of KomodoDragon 2.5 finished. Huge progress!

1000 games head-to-head vs. Stockfish 14 is impressive, too: 1000 (+ 27,=872,-101) = 46.3 % (means: -26 Elo performance)

https://www.sp-cc.de

[AndrewGrant]

Larry: "...we trained primarily against Stockfish for this version... We used some ideas from Stockfish..."

Training against = Using Training Data Generated By (?). Boring.

You should have read the full post instead of this little snippet.

That would be the point of my "(?)".

So why not give perhaps the strongest commercial engine ever - one that has a veteran GM actively working - the benefit of the doubt before implicitly accusing them of using Stockfish data and dismissing their work as "Boring"?

Why not use your immense intellectual gift for some positivity instead of always being salty?

How you respond to the rapidly changing computer chess ecosystem is how you'll be remembered.

Think about this. Your IQ seems to be very impressive, but your EQ lets you down.

Its a question. It has a question mark. Someone came and answered my question. Calm down.

[AdminX]

KomodoChess has just released Dragon 2.5 at KomodoChess.com. Normally we would call this 2.1, but it is probably the
largest upgrade we have had between versions in many years, with the exception of the NNUE idea first implemented
in Dragon 1. It could easily be Dragon 3, but it's only been four months since Dragon 2 and we can't use that name
this soon contractually. It has a larger, "smarter", better trained net, and noticeably more search depth due to search
enhancements.

7000 games ratinglist-testrun of KomodoDragon 2.5 finished. Huge progress!

1000 games head-to-head vs. Stockfish 14 is impressive, too: 1000 (+ 27,=872,-101) = 46.3 % (means: -26 Elo performance)

https://www.sp-cc.de

Thanks, I was looking forward to seeing your results.

[pohl4711]

KomodoChess has just released Dragon 2.5 at KomodoChess.com. Normally we would call this 2.1, but it is probably the
largest upgrade we have had between versions in many years, with the exception of the NNUE idea first implemented
in Dragon 1. It could easily be Dragon 3, but it's only been four months since Dragon 2 and we can't use that name
this soon contractually. It has a larger, "smarter", better trained net, and noticeably more search depth due to search
enhancements.

7000 games ratinglist-testrun of KomodoDragon 2.5 finished. Huge progress!

1000 games head-to-head vs. Stockfish 14 is impressive, too: 1000 (+ 27,=872,-101) = 46.3 % (means: -26 Elo performance)

https://www.sp-cc.de

Thanks, I was looking forward to seeing your results.

Result of KomodoDragon 2.5 MCTS will follow tomorrow...

Longtime "SuFi for the poor" testrun of Stockfish 210915 vs. KomodoDragon 2.5 (Because, I believe, that KomodoDragon 2.5 will be the opponent for Stockfish in the next TCEC Superfinal) will need still 5 days (300 games with 7.5min+4.5sec thinking-time, engines using 5.5 of all 6 cores on my 6core Notebook, using J.Noomen TCEC SuperFinal lowdraw openings, around 50 games/day are played).

Stay tuned!