TCEC Season 12 Update: A Tight Race at the Top
The penultimate stage prior to the super-final is getting close to the end, and all three of the top engines are still in the running. Stockfish leads with 27/42, Komodo has 26.5, and Houdini has 25.5. (In fourth place is Fire 7 with just 19 points.) Just counting the head-to-head-to-head of the big three, Stockfish leads with +1 (a win over Komodo and a bunch of draws); Houdini is equal (a win and a loss to Komodo) and Komodo is therefore obviously at -1.
All three of the top engines are pounding the snot out of Andscacs, Ginkgo, and Jonny, but surprisingly Houdini is a winless -1 against Fire 7. (Komodo is also struggling against Fire 7, relatively speaking, with just one win among a string of draws. Only Stockfish has a comfortable plus score against Fire: +3=4.)
For more, here's the site.
Reader Comments (3)
As you said Dennis the Big Three are killing the wannabes, but Stockfish and Houdini are losing a lot of points by not winning enough. Komodo 12 on the other hand has pick up over 50 points by overwhelming last two engines. I am hoping the championship allows all three to play.
[DM: They are losing some points, but I think the +/- number gives the difference between their rating and their performance rating. But I might be wrong. On your hope (and mine, too): no, they haven't changed the rules. Maybe next time?]
What is interesting is that Stockfish (scoring +8) is not performing as well against the bottom three as Komodo (+11) and Houdini (+10). Is it a matter of the contempt settings? Does Stockfish take fewer risks?
[DM: On the contrary, Stockfish has a higher contempt setting, and to that extent takes more risks.]
It would be great to see a three engine championship between S, K and H to decide the ultimate champion!
[DM: Agreed!]
So if Stocky takes more risks then it doesn't appear to be paying off.
[DM: I was careful in my reply to your last comment: I didn't say that Stockfish took more risks than the other engines, but only insofar as it had a higher contempt setting. The contempt setting has to do with an engine's willingness to take a draw, but isn't what we might anthropomorphize as a stylistic issue. Let's pretend that Anatoly Karpov and (the young) Mikhail Tal are chess engines. Karpov plays relatively low-risk chess, typically trying to grind out games by positional means; while Tal sacrifices, attacks, and plays very speculative chess. If we give Karpov a high contempt setting and Tal a low one it won't change their styles; it just means that Karpov will do his best to avoid draws by repetition and tablebase draws in his generally low-risk way, while Tal might sacrifice a piece and then accept a draw offer or head for a perpetual check.]
The strange thing is if Fire wasn't playing then Stocky might not have made the grand final.
[DM: I guess...but let's say Houdini had a better event and then Komodo qualified just ahead of Stockfish because it obliterated last-place Jonny with a +6 score compared to Stockfish's +3. Assuming current trends hold out for the last two rounds, I don't see any reason to think Stockfish's place in the final is the result of undue luck, as it has gone undefeated and enjoys a plus score against every engine but Houdini. But as I've said many times, it would be much, much more interesting if TCEC had a three-engine final. The top three have a big enough gap between them and their nearest rivals that all the preliminary rounds are pointless from a purely competitive standpoint. (It doesn't mean they aren't entertaining or useful to the programmers and spectators.)]
I think you are right on the rating calculation. The calculation is based on the current rating and not the live rating so each time Stockfish draws with Komodo the +/-ELO number goes down 4 points. I know this because I watched the number go down after S and Ks last draw. I need to get out more...