First knight odds match with IM Saturday

[lkaufman]

Tried one against SF:

[pgn]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]

1.Nf3 d5 2.d4 Nf6 3.c3 e6 4.Bf4 Bd6 5.e3 O-O 6.Be2 c5 7.dxc5 Bxc5 8.O-O Nc6 9.Rc1 Bd6 10.Bxd6 Qxd6 11.c4 Rd8 12.Qc2 dxc4 13.Rfd1 Qe7 14.Rxd8+ Qxd8 15.Bxc4 Bd7 16.a3 Qe7 17.Bf1 h6 18.Qb3 Be8 19.Be2 Rd8 20.Qc2 e5 21.Nd2 Bd7 22.b4 a6 23.h3 Rc8 24.Ne4 Nxe4 25.Qxe4 Nd8 26.Rxc8 Bxc8 27.b5 axb5 28.Bxb5 Qxa3 29.Qxe5 Be6 30.Qb8 Qe7 31.Kh1 g6 32.Bf1 Kg7 33.Qa8 Bd5 34.Qa5 Bc6 35.Qb6 Ne6 36.Kg1 Qc5 37.Qxc5 Nxc5 38.f3 Kf6 39.Kf2 g5 40.Ke1 Ke5 41.Kd2 f5 42.Kc3 Ne6 43.Bc4 f4 44.e4 Nd4 45.Bf1 Bb5 46.Bd3 Bxd3 47.Kxd3 b5 48.Kd2 Nc6 49.Kc3 b4+ 50.Kd3 Na5 51.Ke2 Kd4 52.e5 Kxe5 53.Kd3 Kd5 54.Kd2 Kd4 55.Kc2 Ke3 56.Kb1 Kf2 57.Kb2 Kxg2 58.Kc2 Kxf3 59.Kd3 h5 60.Kd4 g4 61.hxg4 hxg4 62.Ke5 b3 63.Kd4 b2 64.Kc5 b1=Q 65.Kd6 g3 66.Kc5 g2 67.Kd6 g1=Q 68.Ke7 Qe4+ 69.Kf8 Qc5+ 70.Kg7 Qee7+ 71.Kh8 Qc8# 0-1

[/pgn]

It's always similar story. I think one should stick to the following:
-play as classical openings as possible, put pawns in the center (e4 e5, d4 d5 etc.)
-don't fianchetto anything
-don't gambit anything (I didn't play 3...c5 cause I wasn't sure white can't defend the pawn without Nb1)

I would love to see an engine that can win against me when I do that

Did you set SF Contempt to its max value (100)? That is important, although in my experience even doing so I have a much easier time at knight odds with SF than with Komodo. Against engines regular Komodo is a bit stronger than MCTS at giving knight odds, but so far against humans we have had better results with MCTS (both with and without nn). I think SF just isn't as strongly motivated to avoid even exchanges at knight odds as Komodo is. Aside from Contempt = 150, and setting threads to match your core count, the values you would use in an engine match at 15' + 10" are probably suitable for this purpose. The default MCTS hash value is probably fine, but I would double it to be safe, same with table memory. Hash size depends on your hardware, maybe 4 GB for a typical i7 at this tc might be reasonable, 2 is fine if memory is an issue.

[Jhoravi]

Does this Komodo uses Neural Network evaluation derived from training or just a standard Eval?

[lkaufman]

Does this Komodo uses Neural Network evaluation derived from training or just a standard Eval?

It combines info from a standard Lc0 net with Komodo MCTS analysis in a proprietary way.

[OneTrickPony]

Hi Larry, thank you for answering. I don't have a good i7 as of now (will be purchasing big ThreadRipper soon though). I do have a very good GPU though.
My ELO rating is low 23xx but I stopped playing many years ago during big raise. I am between 2500 and 2600 on Lichess blitz which is close to GM level (on a weaker side). I have some experience playing engines as well (I was addicted to playing Houdini back in the day and was able to draw vs it in even games once in a blue moon with both black and white although that was because I came up with openings it was very bad against (some subvariation of London with white and Nf6 Caro with black).
I will try some games at home once I get a better CPU before I waste any more of your time as your estimated ELO difference for hardware is quite big.

[lkaufman]

Hi Larry, thank you for answering. I don't have a good i7 as of now (will be purchasing big ThreadRipper soon though). I do have a very good GPU though.
My ELO rating is low 23xx but I stopped playing many years ago during big raise. I am between 2500 and 2600 on Lichess blitz which is close to GM level (on a weaker side). I have some experience playing engines as well (I was addicted to playing Houdini back in the day and was able to draw vs it in even games once in a blue moon with both black and white although that was because I came up with openings it was very bad against (some subvariation of London with white and Nf6 Caro with black).
I will try some games at home once I get a better CPU before I waste any more of your time as your estimated ELO difference for hardware is quite big.

Sounds like you are a strong enough player to have a good chance to beat "playkomodo" at knight odds. If you want to try some games on chess.com just let me know. If you don't want your name known I would just use it to look up your rating and refer to you as "FM 2315" or whatever. I assume you are an FM if you have 23xx rating, if that is FIDE rating? Also, do you know how Lichess blitz ratings compare with chess.com blitz ratings for the same players?

[carldaman]

Does this Komodo uses Neural Network evaluation derived from training or just a standard Eval?

It combines info from a standard Lc0 net with Komodo MCTS analysis in a proprietary way.

Does this combination leave out Komodo's legendary evaluation function, or is that included in the mix as well?

[lkaufman]

Does this Komodo uses Neural Network evaluation derived from training or just a standard Eval?

It combines info from a standard Lc0 net with Komodo MCTS analysis in a proprietary way.

Does this combination leave out Komodo's legendary evaluation function, or is that included in the mix as well?

It most definitely does include Komodo's eval, I would say it still dominates.

[JJJ]

Hi Larry, thank you for answering. I don't have a good i7 as of now (will be purchasing big ThreadRipper soon though). I do have a very good GPU though.
My ELO rating is low 23xx but I stopped playing many years ago during big raise. I am between 2500 and 2600 on Lichess blitz which is close to GM level (on a weaker side). I have some experience playing engines as well (I was addicted to playing Houdini back in the day and was able to draw vs it in even games once in a blue moon with both black and white although that was because I came up with openings it was very bad against (some subvariation of London with white and Nf6 Caro with black).
I will try some games at home once I get a better CPU before I waste any more of your time as your estimated ELO difference for hardware is quite big.

Sounds like you are a strong enough player to have a good chance to beat "playkomodo" at knight odds. If you want to try some games on chess.com just let me know. If you don't want your name known I would just use it to look up your rating and refer to you as "FM 2315" or whatever. I assume you are an FM if you have 23xx rating, if that is FIDE rating? Also, do you know how Lichess blitz ratings compare with chess.com blitz ratings for the same players?

As a bullet player I m around 1700 on chess.com and 2000 on lichess.org. But the difference is slighty less for higher player, maybe around 150-200 elo.

[lkaufman]

Hi Larry, thank you for answering. I don't have a good i7 as of now (will be purchasing big ThreadRipper soon though). I do have a very good GPU though.
My ELO rating is low 23xx but I stopped playing many years ago during big raise. I am between 2500 and 2600 on Lichess blitz which is close to GM level (on a weaker side). I have some experience playing engines as well (I was addicted to playing Houdini back in the day and was able to draw vs it in even games once in a blue moon with both black and white although that was because I came up with openings it was very bad against (some subvariation of London with white and Nf6 Caro with black).
I will try some games at home once I get a better CPU before I waste any more of your time as your estimated ELO difference for hardware is quite big.

Sounds like you are a strong enough player to have a good chance to beat "playkomodo" at knight odds. If you want to try some games on chess.com just let me know. If you don't want your name known I would just use it to look up your rating and refer to you as "FM 2315" or whatever. I assume you are an FM if you have 23xx rating, if that is FIDE rating? Also, do you know how Lichess blitz ratings compare with chess.com blitz ratings for the same players?

As a bullet player I m around 1700 on chess.com and 2000 on lichess.org. But the difference is slighty less for higher player, maybe around 150-200 elo.

Thanks; would the same be true for blitz ratings, which are more relevant for our purposes?

[JJJ]

Not easy to say...
Zhigalko_Sergei is 2800 on lichess and 3000 on chess.com
Alexander_Zubov is 2900 on lichess and 3000 on chess.com