There you go again with the untruths.

They are not identical.

Different weightings were used to create the tables. This creates differences.
The ratio between the different weightings are also different for differing piece types and game stages. This creates more differences within the tables.
Various kludges were applied, according to Zach, to create the tables. The kludges differ between the programs.

So kindly (or otherwise) stop with the untruths about "identical".

"You have a suspicion"???  Based on exactly what?

If you want to see copied code, read Zach's report, and then Mark's.

First, you say that the values were matched exactly, with respect to multiplicative factors.  This has been demonstrated in one instance with the initial board configuration (though it's VERY debatable whether there is anything wrong about this, let alone illegal...).  You also say that ALL values match.  Then, we have something that you say is the same, but is clearly different, even with respect to multiplicative factors, and you says, oh, "_some_ things were changed".

Do you see why it's getting more and more difficult to believe your statements about things being the same, especially with regard to code?  This just adds to the inability to show Rybka code that matches Fruit code in the evaluation function, even though it supposedly "matches exactly, other than translation to bitboards".  Your statements on these types of similarities must be taken with a grain of salt, to say the least.

I guess this whole situation has just turned into a bunch of nonsense, and it really doesn't interest me much anymore.  I'll wait to see what happens with any REAL investigations by entities that can be trusted.

Bob isnt going to listen, and get out while he can. Shame, cause I sorta like him.

>> The probably of mere chance alone having such a perfect match of these is beyond reasonable belief.

Yeah, and you, as an accuser, are definitely an authority appointed to decide what is beyond reasonable belief and what's not.

When you have a program that uses formulas, vs one that uses tables, the most rational way to compare is to take the one that uses tables, and see if the formulas fit from the other program.  If they do...  And in this case, they do...  then the conclusion is quite obvious.  I have used formulas, and then later used static initialization once it was obvious that either the formulas were OK, or perhaps that one or two values needed tweaking, which can be much harder with a formula.  Either way, the report stands pretty clearly, if you read and think.  It is not a simple thing to understand, as comparing two programs like this is a painful process.  But it is understandable, with some effort expended...

As far as "intentionally trying to produce code that matches fruit."  that is the idea here.  One could take a pc/sq table and produce an infinite number of different code segments that would generate those values.  What is the probability?  Basically 0.  But if the tables are shown to match perfectly or almost perfectly with Fruit, it is not so impossible to understand...

> Well, interesting times ....
>
> The panel clearly either did not read the technical reports or they didn't read with a critical eye. I posit that most of them just read the conclusions and skipped over the mass of technical detail. The technicals are an unproofed  disaster-zone. More on this as I delve ......
>
> But, for now, the BB "results" of
>
> 74% Rybka-Fruit "overlap"
> 54% Crafty-Fruit "overlap"
> and around 30% overlap elsewhere
>
> is perfectly explainable in a quite benign way.
>
> The programs tested for "evaluation function overlap" fall into three categories and the results just reflect the categories.
>
> Without wishing to be rude or denigrating to the programs used by BB, they are
>
> a) a group of quite primitive bean-counter programs, the programmers are relative amateurs and the programs are old.
>
> b) a group of two highly talented, and, looking at Fabien's work, with good chessic knowledge. Vas is in this group of two, he is of course a very strong player, an IM.
>
> c) a group of one, Hyatt, very competent programmer, but with limited chessic knowledge, dependent being told this knowledge by others (Valvo etc etc etc as often quoted) but with 25 years experience of the chess from programming.
>
> a = low chess knowledge, amateur programmers (all the others)
>
> b = high chess knowledge, highly competent programmers (Fabien, Vas)
>
> c = medium chess knowledge, highly competent programmer (Hyatt)
>
> which fits to the resultant data of BB.
>
> b > c > a
>

I have been hoping that you would make a sincere effort to refute the findings of the Panel or at least call to question some of the assumptions that the analysis of the Rybka/Fruit is based on. Unfortunately, you seem to be much more interested in making highly speculative, unsupported comments than to actually work at poking holes in the analysis. Your desire to oppose any position that Dr. Hyatt takes seems to lead you to make statements that are less than substantive. Though, given the volume of such statements made by you, I am guessing you don't care.

The final report from the ICGA is definitely incohesive mess. Yet, it still is disingenuous to deliberately mix numbers from two different analyses contailed in the report in order to better support your contentions. Let's stick with the numbers from the EVAL_COMP analysis,
which is a more refined attempt to measure evaluation feature overlap and where the ~74% number for Fruit/Rybka 1.0 comes from. From that same report, the measured overlap of Crafty/Fruit is 34.20%, not ~54%.

> BB wasn't measuring "overlap" he was measuring chess programming skill. If anybody cares to put the program ELO's against BB's "scores" he will find high correlation.

Now, here is a list of the comparisons along with the Elo difference for each pair. Rybka 2.3.2a has been left out, and I have removed the Fruit/Rybka 1.0 entry for the moment.

Eng      Eng     Overlap  Elo Diff
--------------------------------
Crafty  RESP  39.30  185
Crafty  Ryb1  31.80  301
Crafty  Phal  33.10  134
Crafty  Fail  21.20  518
Crafty  Fr21  34.20  188
Crafty  Pepi  41.10  4
Crafty  EX5b  30.30  275
RESP  Ryb1  31.40  486
RESP  Phal  28.80  51
RESP  Fail  29.50  333
RESP  Fr21  33.60  373
RESP  Pepi  30.60  189
RESP  EX5b  43.80  90
Ryb1  Phal  30.50  435
Ryb1  Fail  25.20  819
Ryb1  Pepi  31.40  297
Ryb1  EX5b  30.00  576
Phal  Fail  21.10  384
Phal  Fr21  30.40  322
Phal  Pepi  42.00  138
Phal  EX5b  29.10  141
Fail  Fr21  27.40  706
Fail  Pepi  23.80  522
Fail  EX5b  28.40  243
Fr21  Pepi  29.90  184
Fr21  EX5b  30.20  463
Pepi  EX5b  36.80  279

This is the resulting plot and regression line.



As you can see, there seems to be some correlation between overlap and Elo diff (smaller difference in Elo corresponds to a larger overlap). Yet, the correlation is not that strong (the coefficient of determination is ~0.366). Given the plot (which does not seem to reveal some other possible functional relationship between Elo diff and overlap) and the low coefficient, one can not say that smaller differences in Elo is a major contributor in larger overlap percentages.

Now, let's add the Fruit/Rybka 1.0 data point (113, 74.40) in.



As pointed out in Mark Watkins' EVAL_COMP, the Fruit/Rybka data point is an extreme outlier. Now, Fruit 2.1 and Rybka 1.0 are at least 200 Elo stronger than the rest of the engines, so perhaps engines at that level must contain a great many things in common. Much more so than engines 200 Elo or more weaker do among theirselves. Which of course leads us to the assumption that even stronger engines share even more things in common. And so on. The only problem with this line of thought is that there are no signs that it holds true. Yes, stronger engines appear to be more similar as a group. But not necessarily at the level of Fruit 2.1 and Rybka 1.0.

So, your explanation for the overlap measurements is not without its weaknesses. I'm not confident that anyone can explain why the eval overlap for Fruit/Rybka is so much higher than the other overlaps without evoking plagiarism. Though, it would be nice if some of the authors that had engines of strengths comparable to Fruit 2.1 offered their source code from that period for comparison.

I am a new member so please forgive my ignorance.  I get the distinct impression that some members are accusing ICGA of simply wanting to destroy Vas, but while those same members are screaming for "evidence", I have not seen any concrete evidence from their side as to why ICGA would have an axe to grind.  Of course jealousy is part of our fabric as human beings, but why would there be such an intense jealousy/animosity against Vas?  He is obviously a very intelligent man but Bill Gates he is not.  Isn't it possible, just remotely possible, that ICGA wants to enforce it's rules?
With all due respect, I strongly disagree with the previous poster stating that the best way to resolve this is with lawyers arguing in front of a judge.  Nothing will ever get done if every organization had to go to a judge before making a statement which is what in fact ICGA has done.  Thank you.

Poster "awrist" has stated that Vas has figuratively been lynched.  I find the word "lynching" somewhat questionable and exaggerated in this context.   I can tell you without hesitation that "lynching" is quite a sensitive word on the American side of the ocean with a lot of racial historical significance and that we don't throw that word around lightly.   Would it be analogous if I told our dear poster Awrist that Vas was taken to the gas chambers?
I understand that some of us do not possess the sophisticated "democratic" cultural understanding that he does.  However, at the core of the matter is that ICGA is an independent organization and they can say what they like.  Vas can also say what he likes or say nothing at all.  ICGA also has the right to ban people who it thinks have broken it's rules.  Again however, Vas has the right to even start his own competitions if he so chooses.   He has publicly stated that he doesn't care.   I say good for him.   That doesn't sound like a lynched man!