Komodo Dragon is Kicking Ass in TCEC

[Patishi]

My 2 cents to this topic, I actually use both Dragon and stockfish 12 (Fritz 17 GUI) and only Dragon can solve the next position in less than a second on my machine (10900k, 32gb 3600mhz ram, samsung 970 evo plus nvme ssd..):



Even the great stockfish takes about 15 seconds to spot the (only) right move. For anyone who doesn't know this, it's black's turn and the only winning move is ..Bh3!!
It manages to see the solution in relatively low depth (30 if I remember correctly) while other engines only see it in much greater depths.

Dragon is the only engine so far that can see the solution right away..on my setup at least.

[Cornfed]

Seriously, does Dragon/Komodo EVER win with Black?

Particular that engine...but others as well. It's largely the openings I know; so maybe 'draws vs losses' often come down the the 'relative ineptitude' of the White engine to take the full point from the Black engine.

Maybe someone could take the ShashChess approach for use ONLY when playing Black...perhaps some half points would not get tossed away?

[jdart]

That's even worse, as it becomes a competition of which opening book is better rather than which engine is better.

SSDF actually does test with opening books enabled, if the engine has one. They always have. I think that goes back to the days when the top engines were pretty much all commercial and had included books. They were a package.

[Uri Blass]

2 loss from 4 games. Disappointing.

If all the programs use their own Opening Books Komodo would have NOT lost those two games with Black since they are too close in ratings specially Stoofvless, LCZero, Stockfish NNUE, and Komodo. It would be more interesting for all programmers to provide their own Opening Books and let them fight it out.

That's even worse, as it becomes a competition of which opening book is better rather than which engine is better.

I guess the empty book is good enough for dragon or stockfish to avoid losing games at long time control in TCEC conditions even if the opponents have some known book.

[Chessqueen]

2 loss from 4 games. Disappointing.

If all the programs use their own Opening Books Komodo would have NOT lost those two games with Black since they are too close in ratings specially Stoofvless, LCZero, Stockfish NNUE, and Komodo. It would be more interesting for all programmers to provide their own Opening Books and let them fight it out.

That's even worse, as it becomes a competition of which opening book is better rather than which engine is better.

I guess the empty book is good enough for dragon or stockfish to avoid losing games at long time control in TCEC conditions even if the opponents have some known book.

That would be an Option to let all other engines use their own Opening Books except LCZero, Stockfish, Allies, or Stoofvlees (No Opening Books) What the result would be, better or worse for those programs without Opening Books ?

[Dann Corbit]

That would be an Option to let all other engines use their own Opening Books except LCZero, Stockfish, Allies, or Stoofvlees (No Opening Books) What the result would be, better or worse for those programs without Opening Books ?

Sedat's book contests can be used to measure the value of a book.

In this contest it was about 15 Elo:
https://sites.google.com/site/computers ... book-cs-47

Note also that a lot of the books were worse than no book.

[Chessqueen]

That would be an Option to let all other engines use their own Opening Books except LCZero, Stockfish, Allies, or Stoofvlees (No Opening Books) What the result would be, better or worse for those programs without Opening Books ?

Sedat's book contests can be used to measure the value of a book.

In this contest it was about 15 Elo:
https://sites.google.com/site/computers ... book-cs-47

Note also that a lot of the books were worse than no book.

I do NOT know how exactly they conducted the test of 300 Blitz games at 3+2", but if somebody can get Stockfish NNUE, Komodo Dragon, LcZero, without Opening Books to compete against, Stoofvlees, Alliestein and Ethereal using their own Opening Book, NOW that would be more interesting to see than using Opening Books that give the advantage to White as in TCEC. What would the Outcome be?

[Cornfed]

That would be an Option to let all other engines use their own Opening Books except LCZero, Stockfish, Allies, or Stoofvlees (No Opening Books) What the result would be, better or worse for those programs without Opening Books ?

Sedat's book contests can be used to measure the value of a book.

In this contest it was about 15 Elo:
https://sites.google.com/site/computers ... book-cs-47

Note also that a lot of the books were worse than no book.

I do NOT know how exactly they conducted the test of 300 Blitz games at 3+2", but if somebody can get Stockfish NNUE, Komodo Dragon, LcZero, without Opening Books to compete against, Stoofvlees, Alliestein and Ethereal using their own Opening Book, NOW that would be more interesting to see than using Opening Books that give the advantage to White as in TCEC. What would the Outcome be?

A lot of...or a lot more draws, don't you think?
A lot of the allure and beauty of chess comes from its structures...I hate to think what this would look like.

[Chessqueen]

We will see what happen in the next two games Stockfish NNUE Vs Komodo Dragon ===> https://tcec-chess.com/live.html
Also Ethereal is doing Great against Stockfisk NNUE https://tcec-chess.com/live.html#div=p&game=3&season=20

The openings in Division Premier are too biased towards white; it is very likely that Komodo has won the games because black made a mistake or a blunder in the opening book line. Komodo as black has lost all of its games so far.

What I do NOT understand is Why TCEC or any other testing site is using Openings created by Humans who are obviously weaker than any of the top 8 engines. Why don't they let Opening that are created by Engines from No Opening Books matches. Or even better any GM can feed the first 4 moves to top engines and let those top engines play whatever variation they calculate as the best, since obviously the Black side of those Openings from TCEC seems to be played by very weak GMs. Is there any TOP engine that can be programmed to create good Openings from any first 2 moves provided by GM.

NOTE: I read somewhere that the difference in using Openings to NOT using it with top engines is only 15 Elo. So the solution is very simple let the top 4 Engines play without Openings and all other engines using whatever chess Openings the programmer chose for its engine.

We Know that Stockfish NNUE will be one of the 2 engines to make it to the Super final, the question is which of these Engines will be joining Stockfish, LCZero, Komodo Dragon, or Stoofvlees https://tcec-chess.com/live.html

[Dann Corbit]

That would be an Option to let all other engines use their own Opening Books except LCZero, Stockfish, Allies, or Stoofvlees (No Opening Books) What the result would be, better or worse for those programs without Opening Books ?

Sedat's book contests can be used to measure the value of a book.

In this contest it was about 15 Elo:
https://sites.google.com/site/computers ... book-cs-47

Note also that a lot of the books were worse than no book.

I do NOT know how exactly they conducted the test of 300 Blitz games at 3+2", but if somebody can get Stockfish NNUE, Komodo Dragon, LcZero, without Opening Books to compete against, Stoofvlees, Alliestein and Ethereal using their own Opening Book, NOW that would be more interesting to see than using Opening Books that give the advantage to White as in TCEC. What would the Outcome be?

There are so many moving targets in that test, I would have no way to interpret the outcome.
In Sedat's test, all the engines were identical, the only thing that changed is the book.
Hence, you can directly infer the effect of books one against another.