Post
by lkaufman » Tue Aug 20, 2019 1:15 am
Okay, we settled on the details for this Wednesday's match on chess.com at 5 p.m. Eastern time. It will be Komodo 13.1, standard version this time, vs. my son IM Raymond Kaufman. Komodo will run on my fast (4.1 GHz) 16 core i9 machine. Time limit 10 minutes plus five seconds increment, which is considered rapid (not blitz) on chess.com. Six games, so it should not run past 8 p.m. Eastern Time unless there are some very long games. Here are the handicaps: Two games at knight odds (one game b1, one game g1 odds). Two games at knight odds from chess960 positions, but only using positions with the kings and rooks on their normal squares so the castling rules are normal and the games don't require using chess960 software. One game at pawn (f7) and three moves handicap (White starts with 1e4 and 2d4 already played, and chooses his third move at the board, with f7 removed). One game at f7 + g7 handicap.
Comments: The time limit is the same as in Nakamura's last blitz/rapid match with Komodo, and just slightly faster than the 15' + 2" used in the MVL vs Komodo handicap rapid match. Based on those matches, it seems that the fair handicap for the world's top three rapid/blitz players at this time control is about a +2.00 eval based on Komodo 13.1 MCTS. Of course Ray is about five hundred elo weaker than those guys in rapid/blitz play, so he certainly needs more than a 2.00 handicap, but we know from experience that Ray should be favored at knight odds (about 3.00 per Komodo MCTS) at this time control, and at longer time limits (45' + 15") even FMs rated a couple hundred elo below Ray have won knight odds matches decisively. So probably this means the fair handicap is somewhere around 2.6 or 2.7. The other four handicaps all have evals in or near that range. It is certainly harder for the human to win at knight odds in chess960 compared to normal chess, due to unfamiliarity with typical development schemes, and in the one match we held that way FM John Meyer lost all four games to Komodo. But requiring the kings and rooks to be on their normal squares, making normal castling possible, should reduce the "weirdness" factor. Probably Komodo is favored a bit in these two games, but it's never been tried. The f7 + g7 handicap looks like a joke, as after White opens with 1.e4 he already threatens mate, and the natural defense 1...Nf6? loses the knight to 2.e5. It is clearly much more than any other two pawn handicap, but it is also clearly less than knight odds. In the only game on record with this handicap, Komodo defeated FM Larry Gilden (about 2100 FIDE at the time). As for the f7 and three move handicap, that was typically considered as the final step before knight odds back in the Morphy era. Komodo has given this handicap twice in official games, losing to GM Simon Williams but defeating IM Danny Rensch, who (in his defense) seemed to be spending more time talking to the viewers than thinking during his game. No one has ever beaten Komodo at f7 and two move handicap in an official game (lots of draws by GMs, but also lots of losses), but the third move makes quite a large difference.
So I rate Ray as the favorite in two games, Komodo as the favorite in four, but all should be close and all three results are plausible in each game. Perhaps in the future we'll repeat the match using Komodo MCTS so as to be able to compare the two matches.
Komodo rules!