Komodo 3 release

[Don]

We are working on the next Komodo release now.

Already available are Komodo 3 for Android and Komodo 3 for 64 bit Linux. Hopefully we will have Komodo 3 for windows in the next day or two.

You can get it here:

http://komodochess.com


We believe we have approximately 30 ELO of improvement and more at non-fischer time controls which we were deficient in. However it's difficult to say for sure how much we have gained as this depends on how it scales to time controls we don't test at.

The majority of the gain has come from evaluation improvements.

Hi,

thanks for the new version. Will it be "MP" too?

Best
Wolfgang
CEGT-Team

We are now working on MP - it will be in the next release. Komodo 3 is the last version of Komodo that will not come in an SMP version.

[Lion]

I do not agree with you.

Since your live is about Y Years, lets say the same as me, You will not be able to produce the same using 1 core vs 6 cores.

Time matters a lot since its limited !

rgds

[Don]

It's my belief that positional play is the next breakthrough in computer chess. We have studied many games of Komodo and and tried to determine why it wins and loses. That was the basis of some of the improvements in Komodo 3. You never win a game due to a flashy combination - it has to be because the opponent made an error that allowed you to win. Of course you can lose a won or drawn game by failing to see a tactic and thus making the wrong move that loses tactically, but that is actually extremely rare. In most of the tactics where you win for tactical reasons you win anyway even if you miss the tactics because you have a won position. Or you play the right move even though you do not yet see the tactics. Of course from time to time you can throw the win away - but again - that's not the big problem in computer chess.

When I did the similarity tester last year it was an eye opener for me. Not because of the clone issues or anything like that, but because it showed that a lot of decisions a program makes are "baked in" to each chess program. You can call it a stylistic preference and there is nothing wrong with that, but comparing any program against itself at a much deeper level doesn't change it's playing style and it's "baked in" preferences.

That is all well and good, but a percentage of those preferences can be considered biases that hurt the program and cause bad strategic decisions. In one case that we identified, Komodo plays a "weak" move and a super deep search does not correct it because it's not a tactical issue and it does not reveal anything that changes Komodo's mind. It might not have been a losing move technically but it turns the game from a possible win to a sure (and perhaps even difficult) draw.

That's why I believe this will be the next big thing that we must learn how to deal with and improve upon. Systematic error and misconceptions if you want to call it that. Of course knowing that and fixing it are two different things - but we are trying to attack the problem.

It's a little tricky because sometimes you see the odd move that no human would make that looks rather ugly - but the more you analyze it the more you see that it may not be so bad or even that it's good and YOU are the one that is wrong - the computer was not as stupid as you thought. But every program will occasionally plays the odd move that constitutes a bad decision point.

Another way to see this, and something that is along the same lines, is trying to win more games when you have an advantage. As programs get stronger you see more and more draws and that is how it should be. However, it's my belief that in some cases the advantage may have been enough for a win but it required an understanding of the position that the program did not possess. So we are looking into that too.

Hi Don, very good post!

It is refreshing after so many trashing posts about the Rybka-Fruit-Craifty affair.

Are you planning to use different techniques to accomplish it?

We are doing several things to attempt to accomplish this, but it starts with examples of positions where there is a problem so that we have something to work with. One of the things I'm doing starts with games of very strong grandmasters and a script which analyzes those games with Komodo. Grandmasters still play better chess than computers - at least in the strategic sense. The only reason computers now consistently beat them is that it is only a matter of time before any human will have some sort of minor oversight - or often a series of 2 or 3 of them. It's hard to beat the absolute consistency of computers.

This is a way to try to automate the process of identifying problems. The script runs each move played by the grandmaster through komodo and pays close attention to moves that Komodo would not play that they DID play. When one of these is found, the script determines if the difference is small or large and we ignore positions where the computer plays a different move but it's more or less a close call.

This typically generates just a very few positions where Komodo believes the human grandmasters played a bad move. Those are searched extra deeply to eliminate false matches or stylistic bias as much as possible. What we are left with is a few positions where there is more than just a trivial disagreement and Larry looks at these positions.

In some cases the human Grandmaster just played badly or even made a tactical blunder and Larry tries to determine that. We don't care about those positions. What we are looking for are positions where the human played correctly and Komodo did not know any better - those are the positions where there is something to be learned by Komodo. We also don't care too much about positions where one side is already clearly winning - those are not particularly interesting and they are certainty not game changing unless there is a serious blunder, which we are not interested in anyway.

At some point we are left with a few positions that the computer did not understand. The human played differently because he knew what he was doing and Komodo did not. These positions have generated a lot of ideas for us and evaluation improvements or just adjustments. In many cases we didn't actually fix the problem, but at least we reduced the disparity and this improves the program.

You CANNOT just take 20 positions where the computer plays the wrong positional move and fix all 20 to play the right move and then expect that the program is now stronger. For every position you fix you could cause 10 others to go bad! So the idea is to actually identify real evaluation issues or problems and to correct them, not to rig the program to solve specific positions. We look for issue, determine what concepts might be missing, implement and then autotest. Sometimes it fixes the specific position and sometimes it does not. What matters the most is whether the program as a whole is now stronger.

Another analysis we are doing is to classify games by score and stage of game. The idea of this analysis is to find scoring disparities. If a program reports a score of "35" for instance, it should mean the same thing regardless of the kind of position that was evaluated. For example 35 might mean that it scores 58% of the games. But if in the open stage it wins 58% of the games given that score but in some endgame phase it scores 51% of the games given that score, it means there is a major scoring disparity. In other words, it's scoring the position too high in the ending and/or too low in the opening. In a given search you might see BOTH positions with the same score - so which do you choose? The search thinks they are equal but the reality is that it is not and thus any decision involving these comparisons will be invalid. This is exactly the type of thing that would cause the program to make a bad decision. So the same score has to represent the same consequences if you expect your program to make good decisions.

IMO, the efficiency of the high selectivity and aggressive pruning in top chess programs suggests that chess itself has some essential properties that can be better explored by using less brute-force.

Looking forward to the new Komodo "Botvinnik" personality.

Regards,

[Don]

I do not agree with you.

Since your live is about Y Years, lets say the same as me, You will not be able to produce the same using 1 core vs 6 cores.

Time matters a lot since its limited !

rgds

I'm sorry, I don't understand you. What is it that you don't agree with? Are you saying that because my life span is limited I cannot produce an SMP version? Or are you addressing someone else?

[Lion]

Sorry, I was not answering to you but Frank when he says that multi CPU is not important.

I do believe that it is very important since we cannot afford loosing time as it will never come back, unless.....

Aside from that, I am really looking forward to see Komodo SMP, I do believe that you are bringing as a real alternative to Rybka or Houdini & Co.

regards

[Don]

Sorry, I was not answering to you but Frank when he says that multi CPU is not important.

I do believe that it is very important since we cannot afford loosing time as it will never come back, unless.....

Aside from that, I am really looking forward to see Komodo SMP, I do believe that you are bringing as a real alternative to Rybka or Houdini & Co.

regards

SMP on 6 or less cores is still worth a few ELO, perhaps 100 - 150 or so at best. But the benefit from SMP doesn't scale up so well. So you might get 150 ELO from 6 cores (for example) but you won't get another 150 by using 12 cores.

That is why I have not been in a huge rush to get to this - but it's still important.

Thanks you for the kind words.

[Albert Silver]

Sorry, I was not answering to you but Frank when he says that multi CPU is not important.

I do believe that it is very important since we cannot afford loosing time as it will never come back, unless.....

Aside from that, I am really looking forward to see Komodo SMP, I do believe that you are bringing as a real alternative to Rybka or Houdini & Co.

regards

It is obvious he is wrong. I work with analysis on a daily basis. If I save myself 30 minutes a day as a result of more efficient results, I am gaining 180 hours per year.

[Lion]

Sorry, I was not answering to you but Frank when he says that multi CPU is not important.

I do believe that it is very important since we cannot afford loosing time as it will never come back, unless.....

Aside from that, I am really looking forward to see Komodo SMP, I do believe that you are bringing as a real alternative to Rybka or Houdini & Co.

regards

SMP on 6 or less cores is still worth a few ELO, perhaps 100 - 150 or so at best. But the benefit from SMP doesn't scale up so well. So you might get 150 ELO from 6 cores (for example) but you won't get another 150 by using 12 cores.

That is why I have not been in a huge rush to get to this - but it's still important.

Thanks you for the kind words.

With all due respect, I believe speed is very important right now since the gain between 1 core and 6 cores is what 4 times faster ?
You can not produce the same quality analysis when you have to spend 4 x more time.

Nevertheless, I understand the reason why you wanted to wait to create the SMP version as my understanding was that you first wanted to be 2nd on 1 core and then spend time on working on the SMP version.

regards

[Frank Quisinsky]

Hi,

it's totaly uninteresting.

2 or 4, 6 or more cores have nothing to do with the quality of a chess program. It's a time factor only. Same results you can produce with one core and more time.

That's what I mean!

OK, for sure, much programmer reach factor 1.8, much other 1.7 or 1.3. Speed factors 1 core to 2 cores for an example, speed factors only, not more not less.

Normaly we don't need it because Komodo with 1 core is stronger on actual hardware as the computer world champion. Why a normaly chess player with 1.600 ELO should have interest on more as 2.900 real ELO?

Two reasons:
I will be the best on chess server.
I will be a higher rating in correspondence chess.



Furthermore, with different engines you can't produce a analyze with more cores.

Clear is, more cores and more speed is a part of the future in hardware and software. But this have nothing to do with the quality of an engine.

You can try to find out the ELO strength with 4, 6 8 or one core. The only different in a ELO list you can find is, that a handfull engines have an advantage with more time. A good example is Junior, an other good example is Zappa. But we are speaking from a handfull engines and perhaps 30 ELO more as advantage to others.

For myself is only important the quality a program have with 1 Core. Commercials gave users the information ... you have to buy for a Deep version more money and the user pay for it.

If you pay more money for a product this product have an higher own personally real value.

The reality is you pay each year 50 EUR more for the same things

Perhaps I see that with other eayes, with eayes in testing a lot of engines and the reason are my activity with SWCR.

Not important for myself ... again!
Not in SWCR and not in personal analyzes of positions. If I have a systems with 12 cores, 12 different engines with 1 Core I would use for an analyzes to the same time with a clearly better result as one engine analyze with 12 Cores. To many time losses.

4 Cores 4x 99% - 100%.
4 Cores with one engine = factor 3.2.
I lost 0.8 of analyze time.
Made no sense in analyze positions too ... for myself.

Best
Frank

[Lion]

I do play OTB and have a relatively low ELO of 2050.

when analysing games/openings, time is important and the level of moves from a 2800 or 3200 ELO is very important. Even so I cannot play at that level, they help me to understand some chess mechanics. Trust me I see a difference in analysis between Shredder and Houdini.

Thus as time is limited, it does impact the quality.

regards

EDIT - It 1cpu in some cases doesnt see tactics because of time/speed.