Komodo 9.4

[mjlef]

Thanks for the new version, Larry!
Dynamism looks very interesting.
Contempt for White is helpful in analysis mode.
Pity Persistent Hash still doesn't work

Persistent Hash does work, and was made better in this release. I changed how aging works with the Persistent Hash. Did you try it?

Mark

[Nordlandia]

What is "Minimal Reporting"

Regarding Dynamism - Larry urge "80" for analysis.

Is anything less than 80 hurting raw analysis performance.

He mentioned that single digit numbers produce bizarre moves.

[mjlef]

I would really like people who buy Komodo to read the README file we include. It covers all these things in much more detail, along with a lot more.

Minimal Report is a bit like Suppress, only less aggressive.

I would not say Larry urged anything. He just thinks Dynamism = 80 produces values a GM finds more reasonable.

[lkaufman]

What is "Minimal Reporting"

Regarding Dynamism - Larry urge "80" for analysis.

Is anything less than 80 hurting raw analysis performance.

He mentioned that single digit numbers produce bizarre moves.

Minimal reporting saves time, but it is only noticeable at a level like game in onen second, so maybe we might not need to keep it. I think a value of Dynamism much below 80 will hurt analysis in general.

[Leto]

What is "Minimal Reporting"

Regarding Dynamism - Larry urge "80" for analysis.

Is anything less than 80 hurting raw analysis performance.

He mentioned that single digit numbers produce bizarre moves.

All I know is that D0 scored almost 800 elo below default in 50 games of 2 minute blitz.

Here's an example game from my D0 testing (I called it ND, short for maximum negative dynanism) :

[Event "Testing k94 ND, Blitz 2m"]
[Site "?"]
[Date "2016.03.18"]
[Round "4.1"]
[White "Komodo 9.4 64-bit ND"]
[Black "Komodo 9.4 64-bit"]
[Result "0-1"]
[ECO "A53"]
[PlyCount "40"]
[EventDate "2016.03.18"]
[EventType "tourn"]
[Source "Atreides"]

1. d4 Nf6 2. c4 d6 3. e3 g6 4. Qa4+ c6 5. b4 Bg7 6. Nc3 Bf5 7. c5 O-O 8. Ba3 e5
9. dxe5 Ng4 10. Rc1 dxc5 11. bxc5 Nxe5 12. Rd1 Nbd7 13. Qb4 b5 14. cxb6 axb6
15. Bb2 b5 16. h4 c5 17. Qf4 b4 18. Nce2 Rxa2 19. e4 Be6 20. Bc1 b3 0-1



7.c5 and 8.Ba3 are bizarre, seems like it places so much importance on its pawns.

[Nordlandia]

Thanks for reply!

Regarding long time controls such as 60+30 and above.

Is default parameters optimal? or maybe some modest change to few of the more influential ones can benefit in the long run. I'm not sure so i ask for advice.

I tend to use Table Memory 256 MB at 60+30 and 128 MB 30+30 TC.

"King Safety" 70*

"Dynamism" 100*

Also i presume "Contempt for White" work in case position get mirrored with colored reversed.

[mjlef]

Thanks for reply!

Regarding long time controls such as 60+30 and above.

Is default parameters optimal? or maybe some modest change to few of the more influential ones can benefit in the long run. I'm not sure so i ask for advice.

I tend to use Table Memory 256 MB at 60+30 and 128 MB 30+30 TC.

"King Safety" 70*

"Dynamism" 100*

Also i presume "Contempt for White" work in case position get mirrored with colored reversed.

If you have "Contempt for White" set in computer matches, the Contempt value would be applied to whoever is playing White. But in computer matches, you want Contempt to apply to whatever side Komooo is playing. So you should have "Contempt for White" off, and then set COntempt to a value based on the strength of the opponents. If you are playing a strong program like a recent Stockfish, use 0 or a small number. Against weaker opponents a higher number like 10 or 15.

Contempt for White is a feature to use in Analysis, where you think the goal of one side is to avoid draws and simplification, and the other side seeking a draw and exchanges. It locks the Contempt values to player color instead of assuming Contempt should be applied to whichever side Komodo is calculating for.

Mark

[peter]

Persistent Hash does work, and was made better in this release. I changed how aging works with the Persistent Hash. Did you try it?

I did so.

Here is one more time the well known Popov- Study (you know it till now from our many mails too, do you?
)

With empty hash (1024 Mb only, to make it easy to be stored and reloaded):

[D]8/2N4r/1p3pkp/8/5K1p/2P4N/P3Bn2/8 w - - 0 1

Analysis by Komodo 9.4 64-bit:

1.Nxf2 Rxc7
-/+ (-0.79 --) Depth: 7 00:00:00 24kN
...
1.Nxf2 Rxc7 2.Bd3+ Kg7 3.Nd1 Rd7 4.Nf2 Ra7 5.Bf5 Rxa2 6.Ne4 h5 7.Bh3 Ra3 8.Bf1 Ra4 9.Bh3 Ra2 10.Be6 Ra3 11.Bd7 Ra7 12.Bf5 Ra1 13.Bh3 Rh1 14.Nf2 Rc1 15.Ne4 Rd1 16.Bg2 Kg6 17.Bh3 Rh1 18.Nf2 Rc1 19.Ne4 Ra1 20.Nd6 Ra4+ 21.Ne4 Kg7
-/+ (-1.58) Depth: 37 00:05:43 9813MN


After having "got" the solution being "shown" so and having taken back again one ply only:

8/2N4r/1p3pkp/8/5K1p/2P4N/P3Bn2/8 w - - 0 1

Analysis by Komodo 9.4 64-bit:
...
1.Bh5+ Kg7 2.Nxf2 Kh8 3.Nb5 Rg7 4.Bf3 Kh7 5.Nd4 Rg1 6.Bg4 Rf1 7.Ke3 Kg6 8.Be6 Rc1 9.Ne4 Rh1 10.Kf3 h5 11.Nf2 Re1 12.Nd3 Rd1 13.Nf4+ Kg5 14.Nh3+ Kg6 15.Bf5+ Kf7 16.Nf2 Ra1 17.Be6+ Kf8 18.Bb3 Rc1 19.Nd1 Kg7 20.Kg2
+- (3.25) Depth: 32 00:00:19 591MN

After "Save Hash":

8/2N4r/1p3pkp/8/5K1p/2P4N/P3Bn2/8 w - - 0 1

Analysis by Komodo 9.4 64-bit:

1.Nxf2 Rxc7
-/+ (-0.77 --) Depth: 7 00:00:00 8037kN
...
1.Nxf2 Rxc7 2.Nd1 Rd7 3.Ne3 Rd2 4.Bf3 h3 5.Be4+ Kf7 6.Kg3 h2 7.Nc4 Rxa2 8.Bg2 Rc2 9.Nxb6 Rxc3+ 10.Kxh2 Ke6 11.Na4 Re3 12.Nb6 Ra3 13.Nd5 Ke5 14.Nb6 Kf5 15.Nc4 Rb3 16.Nd6+ Kg4 17.Bd5 Rd3 18.Ne4 Kf5 19.Bc6 Re3 20.Nf2 h5 21.Bd7+ Kf4 22.Nh3+ Ke5 23.Nf2 f5 24.Kg2 Rc3 25.Ba4
-/+ (-1.48) Depth: 39 00:08:21 13973MN
1.Nxf2 Rxc7
-/+ (-1.55 --) Depth: 40 00:13:40 22910MN

Solution is gone with the wind (with the GUI command "Save") and believe me, it will not be found again soon, as often as you try, probably with as many GUIs as you like, this time I tried Fritz14 and Arena 3.0 only, but with last Version I did so with chessbase12 GUI,Deep Shredder12, latest Hiarcs CE too.

Time to solution is just nothing better than without the already "knowing it" hash from before.

Simply starting and stopping the analysis doesn't do that at all, (as you can see in output) neither does going backward the line do that. If it did, backward analysis would be senseless.

Hash doesn't "age" by stopping analysis only, at least not as much as to lose solution, it "ages" by clicking the "Save Hash", I still call it being deleted, as for my understanding of deleted hash. I still don't know more about hash- "aging" as what you told me about it, but I do know how it looks like, if hash is deleted, which happens to komodo still by several GUI commands that should not do that, "Save Hash" is one of them for sure, and that's so for a long time now. You remember, when I wrote the first mail to you about it?


After "Load Hash":

8/2N4r/1p3pkp/8/5K1p/2P4N/P3Bn2/8 w - - 0 1

Analysis by Komodo 9.4 64-bit:

1.Nxf2 Rxc7
-/+ (-0.79 --) Depth: 7 00:00:00 3815kN
...
1.Nxf2 Rxc7 2.Nd1 Rd7 3.Bf3 h3 4.Ne3 Rd2 5.Be4+ Kf7 6.Kg3 h2 7.Nc4 Rxa2 8.Bg2 Rc2 9.Nxb6 Rxc3+ 10.Kxh2 Kg6 11.Nd5 Ra3 12.Nb6 Kf5 13.Nc4 Rb3 14.Bf1 h5 15.Nd6+ Ke5 16.Nf7+ Kf4 17.Nd6 h4 18.Bc4 Rc3 19.Bb5 Ke5 20.Nc4+ Ke4 21.Nd6+ Kf4 22.Bf1 Ke5 23.Nc4+ Ke6 24.Nb6 f5
-/+ (-1.57) Depth: 38 00:05:44 9693MN

No sign of the already "found" and "known" solution.
But how could it, having been deleted already by "Save Hash"- command?

Just once again being "shown" the solution:

8/2N4r/1p3pkp/8/5K1p/2P4N/P3Bn2/8 w - - 0 1

Analysis by Komodo 9.4 64-bit:

1.Bh5+ Kg7 2.Nxf2 Kg8 3.Ne6 Ra7 4.Ne4 Rxa2
+- (3.20) Depth: 7 00:00:00 1913kN
...
1.Bh5+ Kg7 2.Nxf2 Kh8 3.Nd5 Ra7 4.Kg4 Ra5 5.Nxf6 Kg7 6.N2e4 Rxa2 7.Kxh4 Ra5 8.Kg4 Kh8 9.Ne8 Kg8 10.N8d6 Re5 11.Kf4 Ra5 12.Nf5 Kh8 13.Bf3 Kh7 14.Nf6+ Kg6 15.Nd5 h5 16.Nd4 Ra3 17.Be4+ Kh6 18.Kg3 Ra7 19.Bg2 Ra3 20.Bf3 Kg5 21.Be4 Ra8 22.Ne6+ Kh6 23.Bf3 Ra2 24.Nef4
+- (3.56) Depth: 37 00:01:17 2246MN

Komodo will keep best move in output for sure, as long as you don't "Save Hash" (or, as I wrote to you several times also, rather annoying too, e.g. change the number of variants in MV- mode and back to single variant again).

http://www.talkchess.com/forum/viewtopi ... 76&t=59139

Stopped and restarted analysis:

8/2N4r/1p3pkp/8/5K1p/2P4N/P3Bn2/8 w - - 0 1

Analysis by Komodo 9.4 64-bit:

1.Bh5+ Kg7 2.Nxf2 Kf8 3.Ne6+ Ke7 4.Nd4 Rh8 5.Bg4
+- (3.29) Depth: 7 00:00:00 30kN
...
1.Bh5+ Kg7 2.Nxf2 Kh8 3.Nd5 Ra7 4.Kg4 Ra5 5.Nxf6 Kg7 6.N2e4 Rxa2 7.Kxh4 Ra4 8.Kg3 Ra5 9.Kf4 Rb5 10.Be2 Rb1 11.Nd5 Rb2 12.Bh5 Rh2 13.Ng3 Rf2+ 14.Bf3 Rb2 15.Nf5+ Kg6 16.Nd6 h5 17.Be4+ Kh6 18.Nf5+ Kg6 19.Nd4+ Kh6 20.Kg3 Ra2 21.Nxb6 Ra1 22.Kh4 Rg1 23.Bf3 Rb1 24.Nd5
+- (3.64) Depth: 38 00:01:30 2621MN

[lkaufman]

What is "Minimal Reporting"

Regarding Dynamism - Larry urge "80" for analysis.

Is anything less than 80 hurting raw analysis performance.

He mentioned that single digit numbers produce bizarre moves.

All I know is that D0 scored almost 800 elo below default in 50 games of 2 minute blitz.

Here's an example game from my D0 testing (I called it ND, short for maximum negative dynanism) :

[Event "Testing k94 ND, Blitz 2m"]
[Site "?"]
[Date "2016.03.18"]
[Round "4.1"]
[White "Komodo 9.4 64-bit ND"]
[Black "Komodo 9.4 64-bit"]
[Result "0-1"]
[ECO "A53"]
[PlyCount "40"]
[EventDate "2016.03.18"]
[EventType "tourn"]
[Source "Atreides"]

1. d4 Nf6 2. c4 d6 3. e3 g6 4. Qa4+ c6 5. b4 Bg7 6. Nc3 Bf5 7. c5 O-O 8. Ba3 e5
9. dxe5 Ng4 10. Rc1 dxc5 11. bxc5 Nxe5 12. Rd1 Nbd7 13. Qb4 b5 14. cxb6 axb6
15. Bb2 b5 16. h4 c5 17. Qf4 b4 18. Nce2 Rxa2 19. e4 Be6 20. Bc1 b3 0-1



7.c5 and 8.Ba3 are bizarre, seems like it places so much importance on its pawns.

With Dynamism at zero (or very near zero) Komodo will essentially play random moves among those that don't lose material or hurt pawn structure. So 800 elo loss is no surprise. A value like 10 will at least play reasonable moves in general, but probably would still lose hundreds of elo.

[cma6]

The Readme file is very well done as background material and explanation.
For analysis, it is unclear why or when one would ever use Contempt for White, since this parameter only weakens the analysis.
Also, my default value is "0", not "10".