Milos wrote:Btw. writing 2600 elo engine is a piece of cake. Just put a bit of randomness in the evaluation (change PST values and remove kind safety and detune other parameters a bit) of any of strong open source engine, ...
You have a funny notion of what 'writing an engine' means. The rest of the world would call that 'copying an engine'.
yes, exactly...
like Vas with Rybka 1.0 beta
Houdart with Houdini 1.0
etc.
both of which (regardless) are 'widely accepted', 'tested by major testing groups', 'sold', etc.
i.e.: 'embraced' by the CC community...
PS-
as far as i remember didn't you do a fair amount of Rybka supporting/defending?
maybe i'm mistaken here...if so my apologies upfront
(sorry, i really don't feel like tediously 'searching' the forum again in an attempt to verify...)
Don wrote:According to Ed Shroeder, I would be a hypocrite if I implemented MP on Komodo because it's not my original idea, it would be plagiarism.
Correction, according to the ICGA.
There is absolutely no such opinion within the ICGA. Your statement is false, and you KNOW it is false. The ICGA journal proves that clearly.
Bob,
Do you feel guilty because you adapted a form of LMR after Fruit made it popular and all the other program started using it? And yet you still had the audacity to convict Vas for copying large sections of code! Shame on you.
Actually i invented reductions back in 1998 and entire 1999 Diep used reductions. LMR and History pruning is a subset of what i experimented with back then. Using history moves for reductions or even as a selective move mechanism was used in the 90s massively also by Said Koudache in his 10x10 international checkers program.
Many authors independant from each other invented reductions and some used it with more succes for their engine than others. Back then in 90s the nps of diep wasn't high enough to really profit from what i invented. Also i could prove on paper that the optimum to reduce was 1 ply; experiments reducing more i could prove on paper to have a more complicated break even point.
Another few selective algorithms i invented back then which i intend to retry now, as with the higher nps it is possible it might work.
The way how you wrote down your response to Bob here i find pretty disgusting.
Actually, Bruce Moreland played with them in 1996. He convinced me to give them a try. The only thing we did not do very well back then was to come up with ideas to selectively reduce some but not others. We were using a "true LMR" approach that in the days of 6-7-8-9 ply searches were killing us and we gave up.
hgm wrote:You have a funny notion of what 'writing an engine' means. The rest of the world would call that 'copying an engine'.
That's because majority of "original" authors do it this way today. If they are not at the very top they never get caught. Of course no one (or only few "fools") will admit it.
And this is what Ed is talking about. And rest of you "old school" unless threatened prefer to consider that great majority is "clean" and that only few are "rotten" while the reality is quite opposite.
Don wrote:Here is more irony. In this scenario I could take Houdini, reverse engineer it, and have the same chances as Houdini to win the tournament without a single original line of code. Since I reverse engineered it the source would superficially look different and with the highly relaxed rules you propose nobody could question it. And I could always use the noble statement that "I am standing on the shoulders of giants" if someone noticed the functional similarities.
Stop the BS. For you and 90% or other "original" authors, taking a 3000 elo program reverse engineering it and making a new program (without any change) that is working is an impossible task. Making a RE program that is 3000 elo strong is even more impossible.
Stop talking about things that you don't have a clue about.
Reverse engineering and making a new program that is equally strong or stronger is much harder than writing an original 3000 elo program.
If you thinks it's so easy just try it.
I'm really sick of you spreading that nonsense over and over again.
I have no idea what your asm skills are, but if you think it would be harder to RE a 3000 Elo program than to write one from scratch, you are "way out there in never-never-land". There's an order of magnitude difference between the two efforts, at the very LEAST. With RE being much easier.
mar wrote:What are you talking about? I certainly don't know Don and I don't care. Wasn't talking about him in the first place. I was simply reacting to what you wrote.
I can't make 3000 elo but I can make 2600. Good enough? I understand assembly well (unlike you obviously). I haven't said that I can't dissassemble a chess engine, you are shifting facts. I could possibly do it.
Will you pay me for that? You know it's a lot of work and don't do RE for living. Just give me a year.
How many chess engines have you written? How many programs in assembler? You know sometimes pure theory is not enough .
Well if someone pays me for a month I can provide him how much of your 2600 elo engine you actually copied from free sources . I'm pretty sure it's more than Vas did in Rybka 1.6 for example.
For the rest, sorry but you are too insignificant to be answered.
Btw. writing 2600 elo engine is a piece of cake. Just put a bit of randomness in the evaluation (change PST values and remove kind safety and detune other parameters a bit) of any of strong open source engine, change a bit LMR, rewrite UCI on your own, copy move generator from some other engine, recompile it and close a source. Since strengthwise you'll be too insignificant for anyone to pay attention, you'll be well below the radar, so you can freely brag how you are an "original author".
It's not surprising at all that since first Ippolit appeared the number of young "original" chess authors with closed-source chess engines in the 2500-2800 elo bracket increased ten-fold...
What is wrong with you? Taking and modifying an open-source chess program is NOT "writing a 2600 engine". That's what is wrong with the world today in computer chess. many believe they can change one line and they 'wrote" a new chess engine. Ain't so...
hgm wrote:You have a funny notion of what 'writing an engine' means. The rest of the world would call that 'copying an engine'.
That's because majority of "original" authors do it this way today. If they are not at the very top they never get caught. Of course no one (or only few "fools") will admit it.
And this is what Ed is talking about. And rest of you "old school" unless threatened prefer to consider that great majority is "clean" and that only few are "rotten" while the reality is quite opposite.
Based on what? reading tea leaves? Horoscope? Fortune-teller? You certainly have ZERO supporting evidence to show that most copy and only a few write...
bob wrote:I have no idea what your asm skills are, but if you think it would be harder to RE a 3000 Elo program than to write one from scratch, you are "way out there in never-never-land". There's an order of magnitude difference between the two efforts, at the very LEAST. With RE being much easier.
You all assume reverse engineering to find out ideas. Reverse engineering to make a new working program is an order of magnitude more difficult.
So if you need for example 1000 hours to get all new ideas from the reverse engineered program and to understand how it works in detail, to get a new working program out of disassembled one that is equally strong and without major bugs requires at least 10000 hours.
However, I'm pretty sure you also know that, but just pretend that you don't get it...
bob wrote:
We explicitly voted to allow nalimov EGTB code. It is the sort of code where for any input, there is exactly one output, no variability whatsoever. Most of a chess engine does NOT fit in that category. Certainly nothing to do with search or evaluation..
As I understand this rationalization is your contribution. Historically it would be nice to understand if it was expressed that way during that first player's meeting? I looked through the rules and I can't find this written in the ICGA rules or journals discussing the rules. Maybe I looked in the wrong place. Either way, those player meeting rules are not written in stone either: It is possible that a current player thinks different from a past player. Some go as far suggesting the past player's opinion is irrelevant. I wouldn't take it that far myself if I were a player.
The historian looking back might be tempted to conclude "The Nalimovs got allowed because everybody had them included already when the issue was brought up, everybody was excited about them and it was way too much work to rewrite it all by themselves (imagine getting that nasty en-passant rule right in retrograde analyses, or compressing while still allowing reasonable random access), so they conveniently excused themselves because Nalimov didn't seem to object either".
He might also think "That one-input one-output because-it-is-neutral is a kind of a weird justification because why then did everyone at more or less the same moment stopped using Ken's databases and started using the Nalimovs, which are functionally entirely equivalent. Maybe because Ken's CD-ROMs can't easily be probed from within the search tree and that was a perceived big advantage. That is funny: one-input one-output but orders-more-faster is suddenly not exactly neutral anymore. Wonder why they would allow that while the discovery that one was illegally using the number 0.0 in a time-related function was a big deal a few years later..."
This hypothetical historian might then shrug his shoulders and instead study the ways of the banned programmers who were much closer to solving chess at the time but outcast by their peers.
I'm not an expert but I've heard of a program called 'bayeselo', many rating lists seem to use it.
What is the INPUT for BayesElo? (BTW I use it daily in my cluster testing). The answer is "raw PGN". From WHAT games? Under WHAT conditions? Under WHAT time controls? Etc. Too easy to manipulate.
Conditions set by the TD, or prior winners, whatever, its implementation is solvable.
You are using a recursive process. You don't let them compete until they show a significant improvement. They can't show a significant improvement if they can't compete. How, exactly, does that seem like a rational process???
I think that bear is dead. It is possible to play vetting games before the tournament to show added value over what you started out with. A bit how it was done for one of the programs in the olympiad this year. (One program was asked to play games before being allowed in its tournament, and lo and behold, it turned out to be possible to do so without creating a logical fallacy or infinite recursion.).
I don't think it was ever "formalized". If you specify exactly what can be copied, you run into trouble when something new comes up. As it is, the players meet before each tournament, and any rule issues are discussed and agreed on before round 1 starts.
This egtb issue pre-dates Nalimov, however, as Steven Edwards released 3-4 piece EGTBs way before Eugene took up the challenge. And participants agreed that was OK. Ditto for Thompson egtbs...
bob wrote:Based on what? reading tea leaves? Horoscope? Fortune-teller? You certainly have ZERO supporting evidence to show that most copy and only a few write...
And you also have ZERO supporting evidence to show that most write and only few copy...
And your programming authority has nothing to do with it. It's human nature that is in question...
. I'm pretty sure it's more than Vas did in Rybka 1.6 for example.