Any news of a Komodo update in sight?

[Nordlandia]

Looks like last abrok patch backfires.

Code: Select all

Remove QueenMinorsImbalance array #1340 

Remove QMI array and adjust bishop, knight and queen coefficients 
in QuadraticOurs and QuadraticThei

[pgn][Event "EXPERT-PC, Rapid 15m+30s"]
[Site "EXPERT-PC"]
[Date "2017.12.17"]
[Round "1"]
[White "Stockfish 171217 64 BMI2"]
[Black "Komodo 11.2.2 64-bit"]
[Result "0-1"]
[Annotator "1.05;0.43"]
[SetUp "1"]
[FEN "3r2k1/pbpp2p1/1p2p2p/2b1Pr2/2P2PnP/P7/1P2Q1P1/R1B1K2R b KQ - 0 1"]
[PlyCount "49"]
[TimeControl "900+30"]

1... h5 {0.43/30 38 Both last book move} 2. Rf1 {1.05/30 24} Bd4 {0.26/27 23
(Nh2)} 3. Kd1 {0.83/33 163 (a4)} Rdf8 {0.00/35 38} 4. g3 {0.74/34 47} Nxe5 {
0.00/36 46} 5. Rb1 {0.83/33 47 (Be3)} Ng4 {0.00/35 57} 6. Bd2 {0.84/34 33} e5 {
0.00/36 42} 7. Kc2 {0.73/34 42} c5 {0.00/38 53 (Nf6)} 8. Rbc1 {0.82/31 75
(fxe5)} e4 {-0.37/32 64 (Ba6)} 9. Kb1 {0.00/33 88 (Bc3)} e3 {-0.74/33 22 (d5)}
10. Bc3 {0.00/29 24} Be4+ {-0.67/35 47} 11. Ka2 {-0.35/36 53 (Ka1)} R8f6 {
-1.10/27 34 (R5f6)} 12. a4 {-0.40/34 141 (b4)} Rd6 {-1.27/28 25} 13. Rcd1 {
-0.64/32 49 (a5)} g6 {-1.44/27 21 (Rff6)} 14. Kb3 {-0.50/38 37} Rff6 {-1.81/28
29} 15. f5 {-0.58/39 42} gxf5 {-1.96/29 23 (Bxf5)} 16. Be1 {-1.75/34 280 (Rf4)}
Kf7 {-2.03/32 90 (Kf8)} 17. Ka2 {-1.82/32 74} Rfe6 {-2.44/30 90 (Ke7)} 18. Rf4
{-1.67/30 24 (Rd3)} Ke7 {-2.70/29 38} 19. b3 {-1.62/33 70 (Rc1)} a5 {-3.71/33
31 (Be5)} 20. Ka3 {-4.87/38 104 (Rc1)} Kd8 {-3.87/37 52 (Be5)} 21. Rc1 {
-4.82/37 68 (Ka2)} Ne5 {-5.19/34 56 (Be5)} 22. Rd1 {-4.87/36 35} Nd3 {-6.20/34
35 (Ng4)} 23. Rf1 {-6.55/35 39} Bb2+ {-6.73/35 22} 24. Ka2 {-6.97/37 30} Nc1+ {
-6.97/36 30} 25. Rxc1 {-6.95/34 9} Bxc1 {-7.33/36 29} 0-1

[/pgn]Lichess analysis - https://lichess.org/GRcGhtiF

[h1a8]

Are you serious ?
Really who cares anymore whether Komodo or any other Engine gets updated anymore unless they incorporate MCTC algorithm into it, instead of the now obsolete Alpha-Beta Search.

This logic is pretty flawed. Alpha Zero won on much more powerful hardware. There won't be anything great on your GPU etc. that will compete with Komodo. (Well that's my prediction anyway - hopefully I am wrong)

Isn’t it the opposite? Alpha zero only searched 80,000 nodes per second. How is that superior?

[Ovyron]

I simply like the Komodo playing style and love to observe Komodo's improvements over the years.

What is your favorite Komodo version's playing style? I think Komodo plays some sort of solid chess, not as exciting as Houdini 6 or SugaR XPro... Except for Komodo 10.1 which without needing contempt seems to like to avoid drawish positions naturally, making it seem like to most attractive Komodo chess, but I'd like to know how subjective is my view on this.

[glennsamuel32]

Could Komodo's evaluation be improved by Matthew's work on Giraffe ?

Reinforcement learning with deep neural networks ?

https://bitbucket.org/waterreaction/giraffe/overview

[syzygy]

Could Komodo's evaluation be improved by Matthew's work on Giraffe ?

Reinforcement learning with deep neural networks ?

https://bitbucket.org/waterreaction/giraffe/overview

The answer is no.

[glennsamuel32]

Could Komodo's evaluation be improved by Matthew's work on Giraffe ?

Reinforcement learning with deep neural networks ?

[url]

The answer is no.

Thanks !

That was quick

[syzygy]

Could Komodo's evaluation be improved by Matthew's work on Giraffe ?

Reinforcement learning with deep neural networks ?

[url]

The answer is no.

Thanks !

That was quick

Komodo's evaluation consists of probably hundreds of hand-crafted finely tuned evaluation features. It is not really feasible to replace that with a neural network without rewriting Komodo from scratch. The result would have only the name in common (and would require dedicated NN hardware to do well).