Leto wrote:Some authors are very protective of their work. If I remember right Anthony Cozzie said he never made a backup of the Zappa Mexico II source code, and when he retired from chess programming he deleted it to make sure no one could take it.
I thought that it was now in the hands of Zach?
Yea it is now in the hands of Zach Wegner, engine renamed to Rondo. I must have remembered his retirement post wrong. This is what it seems it originally said:
"The Zappa project is 100% finished. This includes both tournaments and future releases. Only one copy of the Zappa source code exists in the observable universe today, and that is safely squirreled away in a safe deposit box in the barbarous, provincial depths of rural Iowa, so I couldn't work on it even if I wished to."
Leto wrote:Some authors are very protective of their work. If I remember right Anthony Cozzie said he never made a backup of the Zappa Mexico II source code, and when he retired from chess programming he deleted it to make sure no one could take it.
kranium wrote:and might you admit that the fact that Vas never shared the source code with you, but instead had you work on Rybka 3 in such a strange manner is rather odd?
I imagine that most of the authors of strong commercial programs prefer not to share their source around, unless its absolutely necessary.
Wylie-
Larry is listed as an 'author'...and has indicated himself he is an 'author'.
it should have been completely normal for Vas to share the source code a trusted 'co-author'.
i.e. don't most 'authors' of a chess program see the source code once in awhile?
what was done in this case is pretty unusual...IMO, no matter how one spins it.
It is normal but that is not how it was done. In fact Larry does not have the source code to Komodo either. It's not because I'm holing anything back from him, but he wouldn't know what to do with it anyway. Larry is not a programmer.
So I see no reason for you to make something sinister out of this.
From what I understand and what Larry has told me, Larry did not have nearly the involvement with Rybka that he does with Komodo. In fact he was very excited working with me because he was able to get involved in a much more intimate way with Komodo. Vas kept him at arms length and I got the sense (even though Larry never said it in so many words) that it was not really Larry's program, it was Vas's. It had Larry's name on it because Larry was a contributor and that certainly didn't hurt the sales. Of course Rybka DID have Larry's influence, he contribution was significant because of the evaluation.
kranium wrote:and might you admit that the fact that Vas never shared the source code with you, but instead had you work on Rybka 3 in such a strange manner is rather odd?
I imagine that most of the authors of strong commercial programs prefer not to share their source around, unless its absolutely necessary.
Wylie-
Larry is listed as an 'author'...and has indicated himself he is an 'author'.
it should have been completely normal for Vas to share the source code a trusted 'co-author'.
i.e. don't most 'authors' of a chess program see the source code once in awhile?
what was done in this case is pretty unusual...IMO, no matter how one spins it.
It is normal but that is not how it was done. In fact Larry does not have the source code to Komodo either. It's not because I'm holing anything back from him, but he wouldn't know what to do with it anyway. Larry is not a programmer.
So I see no reason for you to make something sinister out of this.
From what I understand and what Larry has told me, Larry did not have nearly the involvement with Rybka that he does with Komodo. In fact he was very excited working with me because he was able to get involved in a much more intimate way with Komodo. Vas kept him at arms length and I got the sense (even though Larry never said it in so many words) that it was not really Larry's program, it was Vas's. It had Larry's name on it because Larry was a contributor and that certainly didn't hurt the sales. Of course Rybka DID have Larry's influence, he contribution was significant because of the evaluation.
sinister ?
ok Don, yes you can 'spin' it however you want...
please re-read my initial post...
i asked simple question in a simple and respectful manner.
funny, i suspected you'd be quick to jump in and answer for Larry.
kranium wrote:and might you admit that the fact that Vas never shared the source code with you, but instead had you work on Rybka 3 in such a strange manner is rather odd?
I imagine that most of the authors of strong commercial programs prefer not to share their source around, unless its absolutely necessary.
Wylie-
Larry is listed as an 'author'...and has indicated himself he is an 'author'.
it should have been completely normal for Vas to share the source code a trusted 'co-author'.
i.e. don't most 'authors' of a chess program see the source code once in awhile?
what was done in this case is pretty unusual...IMO, no matter how one spins it.
It is normal but that is not how it was done. In fact Larry does not have the source code to Komodo either. It's not because I'm holing anything back from him, but he wouldn't know what to do with it anyway. Larry is not a programmer.
So I see no reason for you to make something sinister out of this.
From what I understand and what Larry has told me, Larry did not have nearly the involvement with Rybka that he does with Komodo. In fact he was very excited working with me because he was able to get involved in a much more intimate way with Komodo. Vas kept him at arms length and I got the sense (even though Larry never said it in so many words) that it was not really Larry's program, it was Vas's. It had Larry's name on it because Larry was a contributor and that certainly didn't hurt the sales. Of course Rybka DID have Larry's influence, he contribution was significant because of the evaluation.
Don,
i am not making something sinister out of anything...
please re-read my initial post.
i asked simple question in a simple manner.
hmm, i suspected you'd be quick to jump in and answer for Larry...
Your initial post was not sinister, it's the one I'm responding to that is. Everyone here can read it just as you said it and make their own judgement, the part about 'no matter how one spins it", which implies that something is fishy about Larry claiming to not have access to the source code. Of course you immediately redirecting attention to your initial post, not the one that is relevant.
In the very same fashion you say, "I suspected you'd be quick to jump in and answer for Larry ..."
Why would I be quick to do that? Do you have a Mcarthy-like explanation for that statement? Is it because we are somehow in cahoots together to promote our evil agenda and you are a crime-buster trying to find truth and justice? I think you are an idiot to make these underhanded implied accusations.
kranium wrote:and might you admit that the fact that Vas never shared the source code with you, but instead had you work on Rybka 3 in such a strange manner is rather odd?
I imagine that most of the authors of strong commercial programs prefer not to share their source around, unless its absolutely necessary.
Wylie-
Larry is listed as an 'author'...and has indicated himself he is an 'author'.
it should have been completely normal for Vas to share the source code a trusted 'co-author'.
i.e. don't most 'authors' of a chess program see the source code once in awhile?
what was done in this case is pretty unusual...IMO, no matter how one spins it.
It is normal but that is not how it was done. In fact Larry does not have the source code to Komodo either. It's not because I'm holing anything back from him, but he wouldn't know what to do with it anyway. Larry is not a programmer.
So I see no reason for you to make something sinister out of this.
From what I understand and what Larry has told me, Larry did not have nearly the involvement with Rybka that he does with Komodo. In fact he was very excited working with me because he was able to get involved in a much more intimate way with Komodo. Vas kept him at arms length and I got the sense (even though Larry never said it in so many words) that it was not really Larry's program, it was Vas's. It had Larry's name on it because Larry was a contributor and that certainly didn't hurt the sales. Of course Rybka DID have Larry's influence, he contribution was significant because of the evaluation.
Don,
i am not making something sinister out of anything...
please re-read my initial post.
i asked simple question in a simple manner.
hmm, i suspected you'd be quick to jump in and answer for Larry...
Your initial post was not sinister, it's the one I'm responding to that is. Everyone here can read it just as you said it and make their own judgement, the part about 'no matter how one spins it", which implies that something is fishy about Larry claiming to not have access to the source code. Of course you immediately redirecting attention to your initial post, not the one that is relevant.
In the very same fashion you say, "I suspected you'd be quick to jump in and answer for Larry ..."
Why would I be quick to do that? Do you have a Mcarthy-like explanation for that statement? Is it because we are somehow in cahoots together to promote our evil agenda and you are a crime-buster trying to find truth and justice? I think you are an idiot to make these underhanded implied accusations.
Don,
I think you should refrain from such personal attacks as to call someone and idiot.
kranium wrote:and might you admit that the fact that Vas never shared the source code with you, but instead had you work on Rybka 3 in such a strange manner is rather odd?
I imagine that most of the authors of strong commercial programs prefer not to share their source around, unless its absolutely necessary.
Wylie-
Larry is listed as an 'author'...and has indicated himself he is an 'author'.
it should have been completely normal for Vas to share the source code a trusted 'co-author'.
i.e. don't most 'authors' of a chess program see the source code once in awhile?
what was done in this case is pretty unusual...IMO, no matter how one spins it.
It is normal but that is not how it was done. In fact Larry does not have the source code to Komodo either. It's not because I'm holing anything back from him, but he wouldn't know what to do with it anyway. Larry is not a programmer.
So I see no reason for you to make something sinister out of this.
From what I understand and what Larry has told me, Larry did not have nearly the involvement with Rybka that he does with Komodo. In fact he was very excited working with me because he was able to get involved in a much more intimate way with Komodo. Vas kept him at arms length and I got the sense (even though Larry never said it in so many words) that it was not really Larry's program, it was Vas's. It had Larry's name on it because Larry was a contributor and that certainly didn't hurt the sales. Of course Rybka DID have Larry's influence, he contribution was significant because of the evaluation.
Don,
i am not making something sinister out of anything...
please re-read my initial post.
i asked simple question in a simple manner.
hmm, i suspected you'd be quick to jump in and answer for Larry...
Your initial post was not sinister, it's the one I'm responding to that is. Everyone here can read it just as you said it and make their own judgement, the part about 'no matter how one spins it", which implies that something is fishy about Larry claiming to not have access to the source code. Of course you immediately redirecting attention to your initial post, not the one that is relevant.
In the very same fashion you say, "I suspected you'd be quick to jump in and answer for Larry ..."
Why would I be quick to do that? Do you have a Mcarthy-like explanation for that statement? Is it because we are somehow in cahoots together to promote our evil agenda and you are a crime-buster trying to find truth and justice? I think you are an idiot to make these underhanded implied accusations.
Don,
I think you should refrain from such personal attacks as to call someone and idiot.
You can talk nicer than that!
Idiot is not really the most accurate way to describe his behavior, but it's flattering in comparison to words that would be more correct. I just have low tolerance for this kind of injustice where you attack someone who is innocent with innuendo. I just don't' have a lot of tolerance for bullies.
I'll try to answer the various questions posed in this thread as well as I can.
1. Of course, Rybka 3 is not 100% new program from Rybka 2.32a, maybe 85% would be a reasonable figure.
2. I never asked for source code, because I wouldn't have been able to understand it anyway. But Vas always explained everything about the code that I needed to know for my work, and a lot that I didn't need to know. Now I'm actually able to read a bit of source code, but not enough to do much by myself.
3. I did work on tuning search parameters for a couple of months just prior to Rybka 2.3.2a, so I am fairly familiar with the search in that version. I think it was already quite different than Rybka 1 search, though I was never very interested in Rybka 1.
4. Regarding the eval, I was given a file with all the terms in Rybka 2.2 and their values, and Vas answered enough questions about them so that I understood how everything worked, just not how it was coded. I didn't concern myself with its relation to Rybka 1 or to other programs like Fruit. I changed many values and tables even by Rybka 2.3.2a, but probably more than half the eval was untouched by me at that point. My most important contribution, which is a feature of all the Ippo programs as well as Rybka, was the idea that piece values should rise steadily as the pieces come off the board, so that the program will automatically try to simplify when ahead or complicate when behind. Anyway if we assume that Rybka 1 was more or less a functional clone or derivative of Fruit (I don't know how close they are, I'm just assuming here), I think that Rybka 2.3.2a had pretty much a new search and maybe a third new eval.
5. Soon after Rybka 2.3.2a was released there was an incident in which someone tricked Vas into sending him all the stuff he normally sent me. Fortunately that did not include source code. It was all published on the internet, presumably helping the competition. After that Vas stopped sending me anything relating to the search, as I was no longer working on it anyway. I know that the search was heavily modified in Rybka 3, based in part on the Ippo/Rybka study you all know about. But as I've said, the Rybka 2.3.2a search was already quite different than anything I'd ever heard about before.
6. By the time Rybka 3 came out, the eval size had roughly doubled, and the older terms were mostly modified by me, often substantially. There may have been a few eval terms that still resembled Rybka 2.3.2a or even Rybka 1, but probably nothing of any great significance. Although I did not write any code, I would say that I was responsible for the algorithms and parameter values of perhaps 90% of the Rybka 3 eval. I also tuned the time control parameters, and chose all the eval values for the alternate Rybka 3 versions called Dynamic and Human.
7. Soon after Rybka 3 I went off the payroll and my only further contribution to Rybka of any significance was the parameter values to set for Rybka 4 to make a "Rybka 4 Human" version. I have remained on friendly terms with Vas but of course now I am a competitor with Komodo.
In conclusion, I cannot say that there was no Fruit or Crafty code or values in Rybka 3, but I am confident that there was no SIGNIFICANT such code or values. I do not dispute the panels' findings about Rybka 1.
lkaufman wrote:I'll try to answer the various questions posed in this thread as well as I can.
1. Of course, Rybka 3 is not 100% new program from Rybka 2.32a, maybe 85% would be a reasonable figure.
2. I never asked for source code, because I wouldn't have been able to understand it anyway. But Vas always explained everything about the code that I needed to know for my work, and a lot that I didn't need to know. Now I'm actually able to read a bit of source code, but not enough to do much by myself.
3. I did work on tuning search parameters for a couple of months just prior to Rybka 2.3.2a, so I am fairly familiar with the search in that version. I think it was already quite different than Rybka 1 search, though I was never very interested in Rybka 1.
4. Regarding the eval, I was given a file with all the terms in Rybka 2.2 and their values, and Vas answered enough questions about them so that I understood how everything worked, just not how it was coded. I didn't concern myself with its relation to Rybka 1 or to other programs like Fruit. I changed many values and tables even by Rybka 2.3.2a, but probably more than half the eval was untouched by me at that point. My most important contribution, which is a feature of all the Ippo programs as well as Rybka, was the idea that piece values should rise steadily as the pieces come off the board, so that the program will automatically try to simplify when ahead or complicate when behind. Anyway if we assume that Rybka 1 was more or less a functional clone or derivative of Fruit (I don't know how close they are, I'm just assuming here), I think that Rybka 2.3.2a had pretty much a new search and maybe a third new eval.
5. Soon after Rybka 2.3.2a was released there was an incident in which someone tricked Vas into sending him all the stuff he normally sent me. Fortunately that did not include source code. It was all published on the internet, presumably helping the competition. After that Vas stopped sending me anything relating to the search, as I was no longer working on it anyway. I know that the search was heavily modified in Rybka 3, based in part on the Ippo/Rybka study you all know about. But as I've said, the Rybka 2.3.2a search was already quite different than anything I'd ever heard about before.
6. By the time Rybka 3 came out, the eval size had roughly doubled, and the older terms were mostly modified by me, often substantially. There may have been a few eval terms that still resembled Rybka 2.3.2a or even Rybka 1, but probably nothing of any great significance. Although I did not write any code, I would say that I was responsible for the algorithms and parameter values of perhaps 90% of the Rybka 3 eval. I also tuned the time control parameters, and chose all the eval values for the alternate Rybka 3 versions called Dynamic and Human.
7. Soon after Rybka 3 I went off the payroll and my only further contribution to Rybka of any significance was the parameter values to set for Rybka 4 to make a "Rybka 4 Human" version. I have remained on friendly terms with Vas but of course now I am a competitor with Komodo.
In conclusion, I cannot say that there was no Fruit or Crafty code or values in Rybka 3, but I am confident that there was no SIGNIFICANT such code or values. I do not dispute the panels' findings about Rybka 1.
Very interesting and thanks for sharing! I always thought that evaluation should be dynamic and change according to how many pieces are still on. That is a very human approach to chess where you always try to simplify when winning and complicating when behind. I had no idea this was already being done in today's engines, and was under the impression that this human "translation" would backfire as complications tend to favor the stronger side if they are of equal strength and on equal hardware. I guess the next big breakthrough would be for engines to handle endgames better, and in this case I think there really needs to be more specialized code as each endgame has specific details that are hard to put in an algo. A big step though would be to provide a module that is specialized in rook endgames. Since most endgames end up in a rook endgames, that could give a nice boost to engine strength. Also, I really don't think it would be too difficult to translate the known specific conditions where a rook endgame is drawn or losing and avoid or try to reach that particular line. Anyone for a PCI express endgame add on card?
