Komodo 3 release

[Leto]

Jokes apart, I have laughed a lot when I read a post talking of Komodo as VERY weak in tactics....hahahaha.


Fern

You may continue laughing but the results are here (Arasan Test Suite (8-12) 500 pos. from immortal forum):

418-Critter 1.2 64-bit SSE4 1-cpu\
416-Houdini 1.5 x64 1-cpu\
403-Critter 1.01 64-bit SSE4 1-cpu\
400-spark-1.0 x64 1-cpu\
397-Rybka WinFinder 2.2 1-cpu\
390-Critter 0.90 64-bit SSE4 1-cpu\
377-Ivanhoe B49jA_1-cpu_x64p\
375-spark-0.4 1-cpu\
371-Rybka 3 Dynamic 1-cpu\
366-Rybka 4.1 x64\
362-Rybka 3 1-cpu\
358-Rybka 4 x64\
356-RobboLito 0.09 x64\
354-Rybka 3 Human 1-cpu\
347-Naum 4.2 1-cpu\
341-Critter 0.80 64-bit 1-cpu\
340-Stockfish 2.0 JA 64bit A200C50 1-cpu\
336-Stockfish 2.1 JA 64bit 1-cpu\
336-Stockfish 1.9.1 JA 64bit 1-cpu\
332-Zappa Mexico II Dissident Aggressor 1-cpu\
325-Stockfish 2.0 JA 64bit 1-cpu\
317-Spike 1.4 1-cpu\
311-HIARCS 13.1 SP\
307-HIARCS 13.2 SP\
307-HIARCS 11.2 SP\
304-HIARCS 12.1 SP\
297-bright-0.5c 1-cpu\
293-Rybka 2.3.2a\
287-Protector 1.4.0 x32 JA 1-cpu\
284-Hannibal 1.0a\
284-Deep Junior 12.5 UCI 1-cpu\
282-Rybka 1.0 Beta x64\
276-Gull 1.1 x64\
275-Komodo 2.01 x64 sse4\
275-Zappa Mexico II 1-cpu\
274-Deep Junior 12 UCI 1-cpu\
270-SmarThink 1.20 x64\
270-Deep Sjeng c't 2010\
262-Shredder 12 UCI\
261-Gull 1.2 x64\
259-Umko 1.1 SSE42 x64 1-cpu\
247-Chess Tiger 2007.1\
242-Komodo 1.3 x64\
241-Deep Sjeng WC2008 x64 1-cpu\
224-Ktulu 9\
216-Deep Onno 1-2-70 1-cpu\
216-Spike 1.2\
197-Pro Deo 1.6 Q3\
197-Booot 5.0.1\
173-Jonny 4.00 1-cpu\

That it plays so strongly despite being so weak on tactics seems to indicate its positional knowledge is exceptional.

[yanquis1972]

On the other hand, SMP is inefficient in the sense that given 4 cores, it's far better to run 4 programs and not one. So for some uses single core programs are better. If I had a number of positions that I need to deep search, it's far more efficient to have a number of instances of a single core program running.

interesting, i think this is the first time i've heard this argued. among engines of roughly x elo i always assumed it was most efficient to run 1 at full speed.

i think the emphasis you & larry are placing on positional play is extremely valuable & will help sell copies when you go that route. critter, houdini, and to a slightly lesser extent rybka are all tactical monsters but none have the style of komodo. since its earliest release even i could spot its 'weakness' in tactics but its incredible strength in positional play. i think you are well on your way to making the most human-like 3000 elo engine and that will be incredibly valuable to those who use it (humans).

[Christopher Conkie]

And so life goes on.....

A Crafty gets Shredded and gets made to feel Junior which in turn discovers a Fruity way only to get Zapped until they get Rybbed but not before a Critter comes and steals it all away.....

Mark my words. Listen very carefully. I shall say this only once......

If you like computer chess you will think twice before you put us all through what happened in these last two years.

The strongest chess engines are like buses.....there will be another one along in a minute.

That's called progress. Technology and the information that creates that, is the only way up. Dolly is dead....but then you are all sheep if you think about it so what would you know? Here endeth my critterique,,,,

Well done Don. A very nice program. Thank you.



Chris

[Robert Flesher]

Code: Select all

[Event "40 Züge in 10 min"]
[Site "swcr-32-2, A3"]
[Date "2011.08.16"]
[Round "19.17"]
[White "Houdini 1.5 w32"]
[Black "GullChess 1.2 w32"]
[Result "1-0"]
[PlyCount "67"]
[EventDate "2011.??.??"]

1. e4 {[%eval 0,0] [%emt 0:00:00]} e6 {[%eval 0,0] [%emt 0:00:00]} 2. d4 {
[%eval 0,0] [%emt 0:00:00]} d5 {[%eval 0,0] [%emt 0:00:00]} 3. Nd2 {[%eval 0,0]
[%emt 0:00:00]} Nc6 {[%eval 0,0] [%emt 0:00:00]} 4. Ngf3 {[%eval 0,0] [%emt 0:
00:00]} Nf6 {[%eval 0,0] [%emt 0:00:00]} 5. e5 {[%eval 0,0] [%emt 0:00:00]} Nd7
{[%eval 0,0] [%emt 0:00:00]} 6. Bd3 {[%eval 0,0] [%emt 0:00:00]} Nb4 {[%eval 0,
0] [%emt 0:00:00]} 7. Be2 {[%eval 0,0] [%emt 0:00:00]} c5 {[%eval 0,0] [%emt 0:
00:00]} 8. c3 {[%eval 0,0] [%emt 0:00:00]} Nc6 {[%eval 0,0] [%emt 0:00:00]} 9.
O-O {[%eval 19,18] [%emt 0:01:17]} Be7 {[%eval 7,18] [%emt 0:00:19]} 10. a3 {
[%eval 16,18] [%emt 0:00:05] (Sb3)} Qb6 {[%eval -1,17] [%emt 0:00:15] (a5)} 11.
Nb3 {[%eval 27,19] [%emt 0:00:13]} O-O {[%eval 7,17] [%emt 0:00:06] (c4)} 12.
Nxc5 {[%eval 33,19] [%emt 0:00:14]} Nxc5 {[%eval 15,17] [%emt 0:00:08]} 13.
dxc5 {[%eval 38,18] [%emt 0:00:00]} Qc7 {[%eval 10,17] [%emt 0:00:21]} 14. Bf4
{[%eval 44,19] [%emt 0:00:00]} Bxc5 {[%eval 9,17] [%emt 0:00:16]} 15. Rc1 {
[%eval 38,19] [%emt 0:01:11]} a5 {[%eval 8,18] [%emt 0:00:00] (Td8)} 16. b4 {
[%eval 44,19] [%emt 0:00:12] (Dd2)} Ba7 {[%eval 14,17] [%emt 0:00:26]} 17. Qd2
{[%eval 44,18] [%emt 0:00:00] (b5)} axb4 {[%eval 17,16] [%emt 0:00:46] (Td8)}
18. cxb4 {[%eval 48,17] [%emt 0:00:22]} Qb6 {[%eval 21,17] [%emt 0:01:17] (Dd7)
} 19. Bd3 {[%eval 55,17] [%emt 0:00:14]} Re8 {[%eval 18,16] [%emt 0:00:09]} 20.
Bg5 {[%eval 99,18] [%emt 0:00:38] (Tfd1)} Bd7 {[%eval 15,15] [%emt 0:00:19] 
(Se7)}

One of the latest SWCR games from today. Here move 20. black is interesting. Gullchess is very weak in tactic, not the attacker compare to Komodo.

Houdini found the right move very fast and that is the different in openings. In this case a very fast and big advantage Houdini found.

Test it:
Question 1: Found Komodo the a better move as GullChess.
Question 2: Found Komodo the move Houdini plays.

I think, Question 1 yes
I think, Question 2 no

without to test it.


Best
Frank

[D]r1b1r1k1/bp3ppp/1qn1p3/3pP3/1P3B2/P2B1N2/3Q1PPP/2R2RK1 w - - 0 20

Yes, I ran it, and it is interesting that in your game Houdini missed the best move 20.Bxh7+, nice find.

My creation finds it FAST! Good, old Dissident Aggressor!

New game
r1b1r1k1/bp3ppp/1qn1p3/3pP3/1P3B2/P2B1N2/3Q1PPP/2R2RK1 w - - 0 1

Analysis by Zappa Mexico II: Dissident Aggressor

1.Bh6 Ne7 2.Qg5 Ng6 3.Bxg7 Kxg7 4.Qf6+ Kg8 5.Ng5 Rf8 6.Qf3 h6 7.Nxf7 Rxf7
+- (1.66) Depth: 9/33 00:00:00 658kN
1.Be3 Qd8 2.Ng5 g6 3.Bxa7 Rxa7 4.Qf4 Qe7
+- (1.70) Depth: 9/33 00:00:00 838kN
1.Be3 Qd8 2.Ng5 g6 3.Bxa7 Rxa7 4.Qf4
+- (1.70) Depth: 9/33 00:00:00 1350kN
1.Be3 Qd8 2.Ng5 g6 3.Bxa7 Rxa7 4.Qf4 Qe7 5.b5
+- (1.85) Depth: 10/33 00:00:00 1444kN
1.Be3 Qd8 2.Ng5 g6 3.Bxa7 Rxa7 4.Qf4 Qe7 5.b5
+- (1.85) Depth: 10/34 00:00:01 1867kN
1.Be3 Qd8 2.Ng5 g6 3.Bxa7 Rxa7 4.Qf4 Qe7 5.b5
+- (1.85) Depth: 11/34 00:00:01 2110kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.Bg3 f5 6.Qh8+ Ke7 7.Qxg7+
+- (2.12) Depth: 11/34 00:00:01 2781kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.Bg3 f5 6.Qh8+ Ke7 7.Qxg7+
+- (2.12) Depth: 11/34 00:00:01 2963kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.Bg3 Ke7 6.Qxg7 Kd8 7.b5 Kc7 8.Rfd1
+- (2.86) Depth: 12/35 00:00:02 4760kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.Bg3 Ke7 6.Qxg7 Kd8 7.b5 Kc7 8.Rfd1
+- (2.86) Depth: 12/35 00:00:02 5081kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Qb6 6.Qh8+ Ke7 7.Qxg7 Kd7 8.Be3 Qa6 9.b5 Qxb5
+- (2.59) Depth: 13/39 00:00:03 8315kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Qb6 6.Qh8+ Ke7 7.Qxg7 Kd7 8.Be3 Qa6 9.b5 Qxb5
+- (2.59) Depth: 13/39 00:00:04 10820kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Qb6 6.Qh8+ Ke7 7.Qxg7 Kd7 8.Be3 Qa6 9.b5 Qxb5
+- (2.59) Depth: 14/39 00:00:05 13255kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Qb6 6.Qh8+ Ke7 7.Qxg7 Kd7 8.Be3 Qa6 9.b5 Qxb5
+- (2.59) Depth: 14/39 00:00:05 15993kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Qb6 6.Qh5 Ke7 7.Nxf7 Bb8 8.Be3 d4 9.Bg5+ Kd7 10.Nd6 Rxa3 11.Qxe8+
+- (3.39) Depth: 15/47 00:00:12 34904kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Qb6 6.Qh5 Ke7 7.Nxf7 Bb8 8.Be3 d4 9.Bg5+ Kd7 10.Nd6 Rxa3 11.Qxe8+
+- (3.39) Depth: 15/47 00:00:13 39029kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Ke7 6.Qxg7 Kd7 7.b5 Re7 8.Qf6 Qb6 9.a4 Nd4 10.Be3 Ne2+ 11.Kg2 d4
+- (3.65) Depth: 16/51 00:00:23 72777kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Ke7 6.Qxg7 Kd7 7.b5 Re7 8.Qf6 Qb6 9.a4 Nd4 10.Be3 Ne2+ 11.Kg2 d4
+- (3.65) Depth: 16/51 00:00:25 79678kN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Ke7 6.Qxg7 Kd7 7.b5 Bc5 8.Nh7 Rxa3 9.Nf6+ Kd8 10.Nxe8 Kxe8 11.Qg8+ Kd7 12.bxc6+ Kxc6 13.Qxc8+ Kb6
+- (4.20) Depth: 17/53 00:00:49 159mN
1.Bxh7+ Kxh7 2.Ng5+ Kg8 3.Qd3 Kf8 4.Qh7 Qd4 5.g3 Ke7 6.Qxg7 Kd7 7.b5 Bc5 8.Nh7 Rxa3 9.Nf6+ Kd8 10.Nxe8 Kxe8 11.Qg8+ Kd7 12.bxc6+ Kxc6 13.Qxc8+ Kb6
+- (4.20) Depth: 17/53 00:00:52 168mN

(, Microsoft 16.08.2011)

[Don]

On the other hand, SMP is inefficient in the sense that given 4 cores, it's far better to run 4 programs and not one. So for some uses single core programs are better. If I had a number of positions that I need to deep search, it's far more efficient to have a number of instances of a single core program running.

interesting, i think this is the first time i've heard this argued. among engines of roughly x elo i always assumed it was most efficient to run 1 at full speed.

That's essentially what I am saying. However if you have 4 cores, the most efficient way to utilize all 4 cores for chess is not SMP but to run 4 single core programs individually. Of course that means you have to have 4 separate chess "tasks" to perform. You could have 4 tests running for instance.

If you were playing 4 correspondence chess games it would far more efficient to search each game in a separate core, not to run MP for 1/4 the time then switch.

i think the emphasis you & larry are placing on positional play is extremely valuable & will help sell copies when you go that route. critter, houdini, and to a slightly lesser extent rybka are all tactical monsters but none have the style of komodo. since its earliest release even i could spot its 'weakness' in tactics but its incredible strength in positional play. i think you are well on your way to making the most human-like 3000 elo engine and that will be incredibly valuable to those who use it (humans).

[Frank Quisinsky]

Hi Fernando,

and I laugh if I read that a user ask for Equinox if 600 other engines are available and each of this engines is min. 500, max. 1.500 ELO stronger as the own level.

Wy you compare chess engines?
You will find out a bit.

If that is not important because tactic is strong enough, and all other thing is good enough, all is good we all are happy each time ... we can used Fritz 1 and the World is fantastic too.

Not the ELO strength is the most important point we should find out, clearly more interesting is the playing style.

If you have to play in a club vs. an other player, better if a GM play vs. an other GM he have to analyze the weaknesses and strength from the opponent. All what he need are games.

Best
Frank

Must laughing if I read your posting because the main point in computer chess generally you haven't understand.

[Frank Quisinsky]

Hi,

the most problem in test suites is ....
We await a clear answer, a best move. Our brain need a clear solving.

A best move indicate in 99% a tactial move for the most of us. You can really search a good test suite which find out positional strengths.

I think the most difficult topic in testing. If you replay Komodo games the positional idea started often with move x, example move 40 and ended with move 60.

Unfortunately, we cann't see it if we look on a part or on one move in the game only.

Often I am thinking if I replay Komodo games ... that's die a slow and agonizing death for the opponents of Komodo.

That's really an interesting topic here!

Best
Frank

[Frank Quisinsky]

Hi,

if I looking on the result (IvanHoe so strong, Junior so weak ... must setting "weak" in "", Fernando will read it) I think it's a good tactical test ... means if I compare the table of this test with complete games results.

The different from a test suite to games ...
Each game is on own test suite, each engine-engine game. Why engines, like Spark, Junior, Hiarcs produced so many fast win games, clearly more as others. Why IvanHoe have a high remis quote and don't produce clearly more short games for one possible example only.

If we wait in test suite of a clear "best move", yes ... many engines can find it but not many engine can find the way to this move. Engines are dummy. They must have a good search to find out an easy combination. We can look on the board and can see it directly. The problem is that our style is not accurate enough and we working in the most cases with "position pics". So we need the engines

Better we need the engines and optimal is to have from each of the engines a clear description in playing style, weaknesses and strengths. Means a visiting card of an engine should be develops and before we install a new engine we should look on this card.

My opinion to test suites is very terrible, for me a fun factor to use such things. I need complete games and not test suites if I have interest on an engine. If I look on this results, I can read nothing! Same if I look on ratinglist. I produced such a list but more interesting as ELO only are other things.

Best
Frank

[Frank Quisinsky]

Hi Chistopher,

your mean topic
Forget clones and looking on the games.
You will have more fun ...

Chain of causation by yourself is broken on a wrong position and you spend to many time on this broken position in your brain.

Time you can use for other things, have nothing todo with computer chess, generally problem human have in bigger groups, like internet.

Best
Frank

[yanquis1972]

On the other hand, SMP is inefficient in the sense that given 4 cores, it's far better to run 4 programs and not one. So for some uses single core programs are better. If I had a number of positions that I need to deep search, it's far more efficient to have a number of instances of a single core program running.

interesting, i think this is the first time i've heard this argued. among engines of roughly x elo i always assumed it was most efficient to run 1 at full speed.

That's essentially what I am saying. However if you have 4 cores, the most efficient way to utilize all 4 cores for chess is not SMP but to run 4 single core programs individually. Of course that means you have to have 4 separate chess "tasks" to perform. You could have 4 tests running for instance.

If you were playing 4 correspondence chess games it would far more efficient to search each game in a separate core, not to run MP for 1/4 the time then switch.

i think the emphasis you & larry are placing on positional play is extremely valuable & will help sell copies when you go that route. critter, houdini, and to a slightly lesser extent rybka are all tactical monsters but none have the style of komodo. since its earliest release even i could spot its 'weakness' in tactics but its incredible strength in positional play. i think you are well on your way to making the most human-like 3000 elo engine and that will be incredibly valuable to those who use it (humans).

ok, so yr basically talking about the IDeA (chessok's aquarium) method of chess analysis. iow, in a single position its best to use random engine x at full capacity but in multiple positions its best to run either multiple instances of engine x or use u, x, y & z on a four core machine.