Komodo 2.03 release is imminent

[Albert Silver]

After analysing the position over 1 hour, it is the best move for Shredder 12, on a Quad core machine.

I fail to see your point. You complained that it was absurd that Komodo did not have 0-0-0 in its top eight moves, and I pointed out that nor did Houdini or Critter. You can either claim that is absurd too, or come to terms that there is no reason why 0-0-0 should obligatorily be a top pick.

I am certain if you test enough engines you will find some that choose it. Personally, I like Komodo's style quite a bit. Attacking chess is fine, but it should be tempered with smarter chess. I think it is quite interesting that the latest generation of engines all focus very much on attack (Houdini, Rybka, Stockfish, and Critter), leading one to believe that is a big part of why they are dominating, and then along comes Komodo, and shows things are not so simple.

[Don]

Komodo definitely has some different evaluations. Which is nice! It is also working great now with the Shredder GUI. I don't really do much engine vs. engine stuff. I mostly use for analysis, opening study and for sparring partners. (which is why it would be neat to see some sort of limit strength option for Komodo---I would gladly pay for that.)

A few months ago I outlined a way to do that, which I intend to implement in the next version of Komodo.

The basic problem is that computers are so good that even at game in 1 second they crush you, so it's really difficult to set a time control so low you have a chance unless you are pretty strong. Many GUI's won't support time controls that are that low. And even if they did, it's disheartening when the computer responds instantly to every move you make while crushing you.

The solution I outlined involves slowing the program down by sleeping between nodes. It will have the effect of making the computer take the same amount of time on ANY computer system and you will now have a program that does something like 100 nodes per second (or whatever you set it for.) So you have normal time controls and such and the computer will behave as if it's running on a computer 20 or 30 years old and very slow. And best of all it will not consume much CPU resources (other than memory of course.)

So I will make sure the next version has that.

Don

[Don]

Thomas,

How come you are so obsessed with castling long in this position? You can find move all day long from games which some human played and the computer will not play.

I replayed the game of the century a while back, the one that Bobby Fischer played, to see how Komodo compares and I found some places where Komodo varies from the moves Fischer played. I did some analysis of my own and I cannot prove that Bobby's moves was any better. I'm talking about the interesting parts, not the opening. In fact, doing this exercise led to me to have a little less respect for that game. As great as the game was (and I don't deny Fischer's brilliant play) Komodo's analysis makes me think the game was over-hyped a bit. It was not always the case that Bobby played the "only" move that wins, etc. and it was also true that Donald Byrne made some mistakes which in many ways cheapen the game.

The Fischer/Spassky position where Fischer took the rook pawn with the bishop and it gets trapped is interesting. At the time there was a debate about whether it was a blunder or miscalculation or whether it was perfectly fine.

20 years ago Rexchess played in a human tournament and played Bxa7 due to the horizon effect. The opponent played b6 trapping the bishop. It was almost funny because the threat of losing the bishop caused Rexchess to play like a rabid dog - it launched this aggressive attack in order to horizon off the loss of the bishop.

As it turned out, the bishop capture probably was a good move and the opponent should not have taken the time to play b6 to trap it. Komodo didn't understand any of this but it played it perfectly as if it really knew what it was doing.

After analysing the position over 1 hour, it is the best move for Shredder 12, on a Quad core machine.

FEN: rn1qk1nr/pp2bpp1/2p3p1/3p4/3P1B2/2N1P3/PPQ2PPP/R3K1NR w KQkq - 0 10

DeepShredder 12:
1/2 00:00 26 151 -0,33 Lf4xb8 Ta8xb8
1/2 00:00 69 401 +0,63 Sg1f3
2/6 00:00 565 3.284 +0,30 Sg1f3 Sb8d7
3/6 00:00 2.129 12.377 +0,51 Sg1f3 Sg8f6 OO Sb8d7
4/6 00:00 3.868 22.488 +0,51 Sg1f3 Sg8f6 OO Sb8d7
5/14 00:00 10.113 53.792 +0,48 Sg1f3 Sb8a6 a2a3 Sg8f6 OO OO
6/13 00:00 15.429 82.069 +0,50 Sg1f3 Sb8a6 a2a3 Sg8f6 OO OO h2h3
7/14 00:00 36.774 181.152 +0,50 Sg1f3 Sb8d7 OO Sg8f6 a2a4 Sd7b6 e3e4 OO
8/16 00:00 70.182 320.465 +0,38 Sg1f3 Sb8d7 OO Sg8f6 a2a4 OO h2h3 Le7b4 Sc3a2 Lb4a5
9/24 00:00 123.156 492.624 +0,41 Sg1f3 Sb8d7 OO Sg8f6 a2a4 OO a4a5 Sf6h5 Lf4e5 a7a6 h2h3
10/22 00:00 249.854 761.750 +0,35 Sg1f3 Sb8d7 OO Sg8f6 a2a4 a7a5 e3e4 d5xe4 Sc3xe4 OO Tf1e1 Le7b4 Te1e2 Sf6d5
11/23 00:00 516.088 1.100.400 +0,29 Sg1f3 Sb8d7 OO Sg8f6 h2h3 OO Lf4g3 Sd7b6 Sf3e5
11/24 00:00 917.472 1.305.081 +0,32 Sg1e2 Sg8f6 OO Sb8d7 f2f3
12/27 00:00 1.100.597 1.353.747 +0,32 Sg1e2 Sg8f6 OO Sb8d7 f2f3 Sd7b6 e3e4 OO e4e5 Sf6h5 Lf4d2 Sb6c4 Sc3a4 Le7g5 f3f4
13/30 00:01 1.895.465 1.516.372 +0,28 Sg1e2 Sg8f6 OO Sb8d7 a2a4 OO a4a5 a7a6 Dc2b3 b7b5 a5xb6/ep Dd8xb6 Se2c1 a6a5 Db3xb6 Sd7xb6 Sc1d3 Sb6c4 f2f3
14/31 00:03 6.336.158 1.661.725 +0,19 Sg1e2 Sg8f6 OO Sb8d7 a2a4 Sd7b6
14/31 00:05 8.700.931 1.672.291 +0,27 Sg1f3 Sb8d7 OO Sg8f6 a2a3 OO b2b3 Dd8a5 Tf1e1 a7a6 b3b4 Da5d8 e3e4 d5xe4 Sc3xe4 Sf6xe4 Dc2xe4 Tf8e8 De4c2 Sd7f6
15/32 00:06 11.077.849 1.712.451 +0,27 Sg1f3 Sb8d7 OO Sg8f6 a2a3 OO b2b3 Dd8a5 Tf1e1 a7a6 h2h3 Da5d8 Te1d1 Sf6h5 Lf4e5 f7f5 Sc3a4 Sh5f6 Sa4c5 Sd7xc5 d4xc5
16/40 00:10 17.509.328 1.731.882 +0,26 Sg1f3 Sb8d7 OO Sg8f6 a2a3 OO b2b3 Dd8b6 h2h3 Ta8c8 Tf1d1 Db6a6 b3b4 Sd7b6 Td1b1 Sf6h5 Lf4h2 Sb6c4 a3a4
17/41 00:18 33.074.046 1.755.150 +0,23 Sg1f3 Sb8d7 OO Sg8f6 a2a3 OO h2h3 Sd7b6 a3a4 Sb6c4 b2b3 Sc4d6 Sf3e5 Sf6h5 Lf4h2 Tf8e8 Se5d3 a7a6 Sd3c5 b7b6 Sc5d3
18/46 01:24 150.274.384 1.774.447 +0,18 Sg1f3 Sg8f6 a2a4 Sb8a6 OO OO Dc2d1
19/44 03:32 381.105.150 1.798.463 +0,16 Sg1f3 Sg8f6 h2h4 Sb8d7 OOO OO Kc1b1 Tf8e8 Td1c1 Sd7b6 Lf4g5 Sb6d7 Sc3a4 Dd8c7 g2g3 Le7d6 Lg5f4 Ld6xf4 g3xf4
20/42 05:36 609.036.622 1.808.905 +0,16 Sg1f3 Sg8f6 h2h4 Sb8d7 OOO OO Kc1b1 Tf8e8 Td1c1 Sd7b6 Lf4g5 Sb6d7 Sc3a4 Dd8c7 g2g3 Le7d6 Lg5f4 Ld6xf4 g3xf4 Dc7a5 Th1g1 Sf6e4 Sa4c3
21/48 10:48 1.176.539.697 1.814.421 +0,26 Sg1f3 Sg8f6 h2h4 Sb8d7 OOO Le7b4 Kc1b1 Dd8e7 Sc3e2 Sf6e4 Se2g3 Sd7f6 Sf3e5 Se4xg3 Lf4xg3 OO Se5d3 a7a5 Lg3e5 Lb4d6 Le5xf6 De7xf6 h4h5 g6g5
22/59 16:31 1.798.659.240 1.815.166 +0,26 Sg1f3 Sg8f6 h2h4 Sb8d7 OOO Le7b4 Kc1b1 Dd8e7 Sc3e2 Sf6e4 Se2g3 Sd7f6 Sf3e5 Se4xg3 Lf4xg3 OO Se5d3 a7a5 Lg3e5 Lb4d6 Le5xd6 De7xd6 g2g4 Sf6xg4
23/52 21:17 2.311.309.474 1.809.598 +0,24 Sg1f3 Sg8f6 h2h4 Sb8d7 OOO Le7b4 Kc1b1 Dd8e7 Sc3e2 Sf6e4 Se2g3 Sd7f6 Sf3e5 Se4xg3 Lf4xg3 OO Se5d3 Lb4d6 Lg3xd6 De7xd6 Sd3c5 Dd6c7
24/56 31:30 3.400.551.210 1.798.876 +0,22 Sg1f3 Sg8f6 h2h4 Sb8d7 OOO Le7b4 Kc1b1 Dd8e7 Sc3e2 Sf6e4 Se2g3 Sd7f6 Sf3e5 Se4xg3 Lf4xg3 OO Se5d3 Lb4d6 Lg3xd6 De7xd6 Sd3c5 Dd6c7
24/56 1:09:32 7.301.101.065 1.749.703 +0,28 OOO Sg8f6 Kc1b1 Sb8d7 h2h4 OO Sg1e2 Sd7b6 Se2g3 Sf6g4 f2f3 Sg4h6 Sg3f1 Tf8e8 g2g3 Le7d6 Lf4g5 Ld6e7 Lg5xe7 Dd8xe7 e3e4 d5xe4
25/60 1:47:50 11.279.670.360 1.743.355 +0,29 OOO Sg8f6 Kc1b1 Sb8d7 h2h4 a7a5 Sg1f3 a5a4 a2a3 Dd8b6 Sc3xa4 Db6a6 Sa4c3 b7b5 Dc2b3 b5b4 a3xb4 OO b4b5 Da6b7 Kb1c2
26/61 2:22:53 15.091.458.550 1.760.247 +0,27 OOO Sg8f6 Kc1b1 Sb8d7 h2h4 a7a5 Sg1f3 a5a4 a2a3 Dd8b6 Sc3xa4 Db6a5 Sa4c3 b7b5 Dc2b3 OO Sc3a2 Sf6e4 Td1c1 Tf8c8 Lf4g5 Le7d6
27/67 4:07:47 26.507.267.951 1.782.883 +0,26 OOO Sg8f6 Kc1b1 Sb8d7 h2h4 a7a5 Sg1f3 Sd7b6 Sf3e5 a5a4 a2a3 Sf6d7 e3e4 Sd7xe5 d4xe5 d5d4 Lf4g5 Le7xg5 h4xg5 Th8h5 f2f4 Sb6c4 Th1xh5

[mhull]

Komodo definitely has some different evaluations. Which is nice! It is also working great now with the Shredder GUI. I don't really do much engine vs. engine stuff. I mostly use for analysis, opening study and for sparring partners. (which is why it would be neat to see some sort of limit strength option for Komodo---I would gladly pay for that.)

A few months ago I outlined a way to do that, which I intend to implement in the next version of Komodo.

The basic problem is that computers are so good that even at game in 1 second they crush you, so it's really difficult to set a time control so low you have a chance unless you are pretty strong. Many GUI's won't support time controls that are that low. And even if they did, it's disheartening when the computer responds instantly to every move you make while crushing you.

The solution I outlined involves slowing the program down by sleeping between nodes. It will have the effect of making the computer take the same amount of time on ANY computer system and you will now have a program that does something like 100 nodes per second (or whatever you set it for.) So you have normal time controls and such and the computer will behave as if it's running on a computer 20 or 30 years old and very slow. And best of all it will not consume much CPU resources (other than memory of course.)

So I will make sure the next version has that.

Don

See the second paragraph of my reply to you last September:
http://www.talkchess.com/forum/viewtopi ... 561#369561

It was in a different context, though. And I think you were too busy discussing things with Bob to acknowledge it as an idea with any merit. But I recall having also suggested it even before this as a way to reduce strength in Crafty in particular and programs in general.

[Don]

Komodo definitely has some different evaluations. Which is nice! It is also working great now with the Shredder GUI. I don't really do much engine vs. engine stuff. I mostly use for analysis, opening study and for sparring partners. (which is why it would be neat to see some sort of limit strength option for Komodo---I would gladly pay for that.)

A few months ago I outlined a way to do that, which I intend to implement in the next version of Komodo.

The basic problem is that computers are so good that even at game in 1 second they crush you, so it's really difficult to set a time control so low you have a chance unless you are pretty strong. Many GUI's won't support time controls that are that low. And even if they did, it's disheartening when the computer responds instantly to every move you make while crushing you.

The solution I outlined involves slowing the program down by sleeping between nodes. It will have the effect of making the computer take the same amount of time on ANY computer system and you will now have a program that does something like 100 nodes per second (or whatever you set it for.) So you have normal time controls and such and the computer will behave as if it's running on a computer 20 or 30 years old and very slow. And best of all it will not consume much CPU resources (other than memory of course.)

So I will make sure the next version has that.

Don

See the second paragraph of my reply to you last September:
http://www.talkchess.com/forum/viewtopi ... 561#369561

It was in a different context, though. And I think you were too busy discussing things with Bob to acknowledge it as an idea with any merit. But I recall having also suggested it even before this as a way to reduce strength in Crafty in particular and programs in general.

I'm going to respond to that old post and see if it comes back to life ...

[georgerifkin]

The solution I outlined involves slowing the program down by sleeping between nodes. It will have the effect of making the computer take the same amount of time on ANY computer system and you will now have a program that does something like 100 nodes per second (or whatever you set it for.) So you have normal time controls and such and the computer will behave as if it's running on a computer 20 or 30 years old and very slow. And best of all it will not consume much CPU resources (other than memory of course.)

So I will make sure the next version has that.

Don

Yes that should work. It's also what Spike does.

[Dann Corbit]

After analysing the position over 1 hour, it is the best move for Shredder 12, on a Quad core machine.

I fail to see your point. You complained that it was absurd that Komodo did not have 0-0-0 in its top eight moves, and I pointed out that nor did Houdini or Critter. You can either claim that is absurd too, or come to terms that there is no reason why 0-0-0 should obligatorily be a top pick.

I am certain if you test enough engines you will find some that choose it. Personally, I like Komodo's style quite a bit. Attacking chess is fine, but it should be tempered with smarter chess. I think it is quite interesting that the latest generation of engines all focus very much on attack (Houdini, Rybka, Stockfish, and Critter), leading one to believe that is a big part of why they are dominating, and then along comes Komodo, and shows things are not so simple.

I believe that Thinker gained strength from a similar approach to Komodo (just a wild guess, based upon the names of the binaries).

[Thomas Lagershausen]

Thomas,

How come you are so obsessed with castling long in this position? You can find move all day long from games which some human played and the computer will not play.

I replayed the game of the century a while back, the one that Bobby Fischer played, to see how Komodo compares and I found some places where Komodo varies from the moves Fischer played. I did some analysis of my own and I cannot prove that Bobby's moves was any better. I'm talking about the interesting parts, not the opening. In fact, doing this exercise led to me to have a little less respect for that game. As great as the game was (and I don't deny Fischer's brilliant play) Komodo's analysis makes me think the game was over-hyped a bit. It was not always the case that Bobby played the "only" move that wins, etc. and it was also true that Donald Byrne made some mistakes which in many ways cheapen the game.

The Fischer/Spassky position where Fischer took the rook pawn with the bishop and it gets trapped is interesting. At the time there was a debate about whether it was a blunder or miscalculation or whether it was perfectly fine.

20 years ago Rexchess played in a human tournament and played Bxa7 due to the horizon effect. The opponent played b6 trapping the bishop. It was almost funny because the threat of losing the bishop caused Rexchess to play like a rabid dog - it launched this aggressive attack in order to horizon off the loss of the bishop.

As it turned out, the bishop capture probably was a good move and the opponent should not have taken the time to play b6 to trap it. Komodo didn't understand any of this but it played it perfectly as if it really knew what it was doing.

Don,

you are absolutly right when you came to the result that the Byrne-Fischer game is totally over-hyped.

It was only great tactics for this time.

But the people were very impressed that such a young boy, as Fischer was in year the game was played, could play such a firework of tactics with the black pieces.

Today, in the age of machines, alot of this tactic moves are less brilliant and show the weakness of human chess.

And here we are, why are so obsessed to see the move 10.O-O-O in the Carlsen-Nakamura game by a computer.

Meanwhile the topchess is living in period where most games ended in a draw.

If you want to win a game at this level you have to play more risky,or more aggressive.

Chess at the top is in big danger to be dead by draw.

So the only way to give chess back the old live is to become a attacking player.

This is the only solution to excape from the death of chess by draw.

Believe me, chess was my religion.

Fischer was the greatest player of his time. But he didn´t play only bestmoves.

He won because of his aggressiveness.

That´s his secret.

And Magnus Carlson knows this too.

[Thomas Lagershausen]

First i want to show that it has to be done further work to play chess in all options to reach the olymp of perfect chess.

It´s not a question of style, it´s a question of ability to play the move that is neccessary to make progress in the position.

Secondly with my example of how Shredder it handles i wanted to show that someone has made the first steps in this area and the flag is on the moon.

There is lot of work to do, because computers are not complete chessplayers.

They won they games mainly by search and tactics.

Now the aliens of chess are stronger than humans. But they have enormous weaknesses and will never play perfect chess if they didn´t learn from human chess.

And the first lesson is to get a better feeling for attackingchess.

Only that was my mission.

And i want to show that SMK and his shredder have started the race to a complete chessplayer in silicon.

[Don]

First i want to show that it has to be done further work to play chess in all options to reach the olymp of perfect chess.

Why don't you tell us something we don't know

I have posted many times here that I believe there are still hundreds of ELO left before perfect play. As a computer chess developer don't you think I have any perspective on this?

It´s not a question of style, it´s a question of ability to play the move that is neccessary to make progress in the position.

We don't strive to make it play like humans because we think that makes it play stronger. We just feel that is more pleasing and more fun to play against.

In fact I think making it play "like" humans probably is not best for the ultimate strength.

A perfect player will not play that much like a human in my opinion.

Secondly with my example of how Shredder it handles i wanted to show that someone has made the first steps in this area and the flag is on the moon.

By showing us 1 example? I think Junior is a better example for this type of thing.

I honestly do not know if we play more like a human that other programs as I have never spent any serious time trying to evaluate the "humanness" of any give program. All I really know is that Larry and I prefer it to play like a human and have put some effort into this. We are not claiming we have succeeded or that we do it well.

A project Larry has been pestering me to start on at some point is a serious attempt to tune our program to play more like humans. The trick is to do this without sacrificing ELO. So we are far from our ultimate goal and I don't know if we will ever fully be satisfied that we have the most human playing program. Maybe we never will - but it's not something we lose any sleep over.

There is lot of work to do, because computers are not complete chessplayers.

Programs have the same flaws they did 40 years ago, they are just hidden better. The horizon effect is minimized and disguised by much greater depths but it's still there. And programs still do not reason about positions so they do approach the game differently that humans.

Again I have to say you are rambling on here - tell us something we don't already know. You act like you are just now figuring this all out but everything you are saying is common knowledge that has been known for years and your insights are not original. They are correct, but not original or even interesting.

They won they games mainly by search and tactics.

Now the aliens of chess are stronger than humans. But they have enormous weaknesses and will never play perfect chess if they didn´t learn from human chess.

And the first lesson is to get a better feeling for attackingchess.

Only that was my mission.

And i want to show that SMK and his shredder have started the race to a complete chessplayer in silicon.