diep wrote:lucasart wrote:diep wrote:
First of all i'm one of those who invented a bunch of algorithms himself and several of my tricks boosted the elo of Rybka, therefore also the rybka clones.
And no i didn't like that.
To give one example of what i invented somewhere start 21th century and which i used in Diep for nullmove:
if( static eval >= beta + S ) then R = R+1;
Just two questions
1/ Is Diep source code available ?
2/ Don't you think that this is a rather logical thing to test, and someone else (maybe after you but independantly if Diep's code is closed) could have thought about it ?
Not at all logical to test. Especially if yo usee what else is in that executable of rybka3. Total trivial stuff. Initially what was kept behind for the clones was a passed pawn evaluation trick that won elo for rybka3. It wasn't in the ippolits....
So you see there also that most of the cloning is source code based.
I dare to say that the guys who built rybka3, they simply had source codes of many private engines. Obviously illegal.
They tried some cheapskate tricks and the ones that worked they kept.
Nothing was selfinvented.
The razoring idea in rybka (cheap nullmove i call it) was not new either. Revived by Omid David Tabibi in Falcon around 2001. He didn't publicly post this.
How did it end up in rybka? I don't know.
However THAT was KNOWN.
Diep's source code is not available.
In all those 20 years no one ever tried that until i tried it in Diep.
The way how i order my root moves is also not total trivial. In all those 20 years no one i ever have been seeing doing it not until after i started doing it.
There is indeed a few engines made by guys known how to debug who also do it.
Maybe 1 or 2 authors i told the trick - they didn't spread that.
Then a year after it is in production executables of diep you see it everywhere suddenly. That's how it works with all tricks.
Debugging assembler is too easy.
I'm not doing it, yet it's total trivial to do.
You know there is a lot of ways to steal source code, but even then, majority of chessprogrammers reads assembler like it's C. Most know assembler better than C actually. Which makes sense if you think about it.
Computerchess is a low level sport. A few C++ guys now know hardly assembler, but that's about the only one.
There is so many things to try - and nearly no one ever tries anything you know. You just don't have the SYSTEM time to measure 1 elopoint.
Bullet some years ago also could not measure anything of that, as you just didn't get through the tactical barrier enough to measure even simple stuff.
In fact even today it doesn't work of course. Some people just tell fairy tales here.
The simple problem is TESTING whether something works and the fact that those who easily invent algorithms have something better to do than to test a chessprogram.
I hope you realize in end 90s most were doing hashtables with a SINGLE lookup. Despite that i posted it publicly here in CCC that i was doing 8 SEQUENTIAL lookups. Bob still was posting in 21th century that this was 'bad for chaining'. You had to do random lookups according to Bob and that's what crafty did do. 2 random lookups.
Until nalimov changed it for him to sequential probing...
that happened when it was obvious AMD would have a faster memory controller than intel with the opteron. Nalimov worked for wintel at that opint. Microsoft. Now he's at google.
Cilkchess at the supercomputer was doing a SINGLE probe.
They just didn't have the time to test it all you know and there is a billion tricks possible which oneas are you gonna test, if you FIGURE OUT the idea of doing more than 1 probe anyway?
Oh it was posted in CCC/RGCC?
Ah well. Have fun with that.
The tricks i described were NEVER posted.
The number of guys on this planet who have good ideas how to possibly improve, you can count on 1 hand.
the number of guys who knows how to test these ideas from others in a correct manner, now that's a LOT of folks...
they wouldn't even invent alfabeta themselves. Now speaking of a simple algorithm, alfabeta sure is something you can selfinvent.
In fact i know of at least 2 math guys who invented alfabeta themselves.
And that isn't a joke.
Even the most obvious things even posted in CCC, a forum that's the last to know something new always, wasn't getting tried, unlike some of the stronger commercials historically. they tried everything. However a number of the 'stronger' commercial engines always had programmers who stole everything by means of debugging.
There is a group who like to invent and doesn't steal and there is a rather large group that has no ethical problems with stealing algorithms.
Guys like me never did do this of course, though i AM capable of checking what goes on in other programs. I have enough ideas myself.
And no you will NOT find this trick.
Additionally it's not just 1 trick mine that was taken over. I described 2. There is more however.
I can point you to another 10 tricks that NO ONE in CCC found here which all give elo, be it little, for simple beancounters like the rybka clones. I have them on paper here. Never saw anyone refer to it and it's trivial most of those tricks never tried. Usually what doesn't work well you'll find on CCC.
They're just NOT finding things that works until someone has an executable around where they just DEBUG it. First of all it takes 6 months maybe to invent something new. Debugging it is 5 minutes.
World champs 2004. Statement from Stefan Meyer-Kahlen.
"Oh i have this wonderful new search now in Shredder it works MAGNIFICENT, but what i find so sad is that 6 months after it's released it ends up in Fritz as well".
