Rebel wrote:Young Zach understood and wasn't afraid to speak out. And it didn't hurt him. In fact I respect him for that.
My recollection is that when "Young Zach" left (in no small part due to your antics) he stated that he voted based on the evidence as he saw it, and that he had couldn't say on what other Panel members based their vote. I took it mean that he would rather have had a (long) enumeration: "Regarding evidence X, do you find it convincing, and do you think it is relevant to Rule #2", and everyone would say yes/no/abstain on each. You can try to contact him to find out his opinion, but I doubt he will respond as he concluded that anything he said would be twisted, decontextualized, and used against him on top-5000.nl
Chris Whittington wrote:chose not to reapply is about the most ridiculous characterisation I read this year.
If you had no intention of re-applying, due to the smell of Hyatt/Levy exclusion, then why exactly did you waste my time trying to get me to liaise the situation in the first place? Simply to plead with me not to join the Panel?
Rebel wrote:You did not comment on my Assange / Snowden post.
Do I have to comment on every fanciful political parallel that you concoct?
Rebel wrote:I easily justify myself as a whistleblower
Whistleblowers are usually for things in the
public interest, often with (gasp) actual governments involved, and affecting significant sections of a population. I fail to see how the ICGA situation (relevant to maybe 100 people at best, of the highly esoteric "computer chess" special-interest group) is much comparable. There is much more of a "public interest" in being trustworthy and keeping private things private.
Or to cite (following Mr. Whittington) Wikipedia on
whistleblowing:
Whistleblower protection laws and regulations guarantee freedom of speech for workers and contractors in certain situations. Are you a worker or a contractor with the ICGA?
Rebel wrote:That the ICGA breached Vas privacy. By your logic Levy is not decent.
The ICGA published the
whole Rajlich/Levy correspondence, of which Levy was part, and as a matter of record in the investigation. When asked, it seems they removed it. You, on the other hand, publish snips and pieces, whatever fits your purpose, sometimes with asking, other times with not. A typical "conversation" with Mr. Schröder goes like this: "I'd like to post your (email) comments on my webpage. Can I do that?" Answer: no. A few weeks later, "Coming back to your comments about X, I'm done with subject and for completeness would like to record your views too. Everyone else has agreed, and I'll edit it as you see fit." Answer: no. A couple of months later: "I still feel that adding your comments to my webpage would help give the full picture blah blah." Answer: no reply to email. Webpage then says: "I contacted X about his comments, but he refused to respond." Some more time passes. Another email, another lack of reply. Webpage changed to: "I have contacted X, and since he no longer seems interested in the subject I have decided to record his comments so that they are not lost to posterity." (Of course, the comments are snipped down and variously highlighted in whatever manner Mr. Schröder prefers).
Incidentally, I was recently contacted by someone who had I not heard from for awhile (no, not Zach), saying that he had accidentally tripped over this thread, and that he didn't think I needed any more "ammunition" against Ed on the privacy issue, but in any case that I could mention XYZ (which I have not, up to this point). So I'm not exactly alone in my opinions here.
Rebel wrote:they both are relative safe and far away from the 60% gray area
This is your invention of the "60% gray area". As argued in Section 7 of the paper, 60% (under the given testing conditions) is much beyond merely "gray".
Rebel wrote:neglectful 3 percent in comparison with Fruit 2.1
Your so-called "neglectful 3 percent" is more than a standard deviation!
[Already on your clone.htm page Adam says "the standard deviation of the data was 2.86" for the 99 families, while depending on the comparison pool 3% is as high as 1.4σ]. Either you are ignorant in statistics (which I doubt), or are simply being disingenuous. Similarly with (see page 58 of Rybka Reloaded) your listing of "Different FEN parsing" and "different Zobrist hash keys" as Rybka/Fruit differences that Zach and I "missed", as if they would have anything to do with a Rule #2 violation. I might expect such silliness from Riis, but with you?
Rebel wrote: Now 58% similarity is a very poor number for 2 identical evaluations.
The actual numbers, as noted in the middle of page 19, were 56% and 60% similarity [see my email mid-April 2012]. Further, again as noted in the paper, having identical evaluation functions is only part of the picture, as by making
search changes the move selections could vary (cf. Thinker).
Rebel wrote:The fact you do not mention this experiment you have done yourself
I personally did not do the experiment you mention (LOOP/Fruit at 84%). It was part of Adam Hair's original data set over 378 engines. However, as explained in the paper, there was a family-based filtration (as Mr. Whittington had desired with EVAL_COMP) which tended to remove "obvious" clones -- indeed Adam's initial data analysis already preferred to discuss the 99 "families", or as Adam states:
The 99 engines listed above can be used to make judgements on the level of similarity found between engines from the larger group and other, supposedly unrelated engines. Before listing the highest move-matchings in the list of 378, he gives the five most extreme examples after delimiting by family (
there are some pairs in the unique engines sample that have greater then 59.46% (5σ) moves matched). I do not see why you concentrate on the extremal numbers in the data from the full 378 engines, as opposed to those from the "unique engines" sample (which we also further pared down from 99 to 86 in an alternative analysis), as the latter are much more relevant when making comparisons. The fact that ProgramX/CloneY matches at 85% does not say much about whether 60% should be considered a large percentage for a presumably unrelated pair.
Moreover, we discuss this further in Section 4.3, upon noting the comparisons a la Riis (where eg Rybka/Fruit is compared to an obvious clone pair like Houdini/Strelka) are not particularly useful. Similarly, we also didn't list numbers for intra-family comparisons like Robbolito/IPPOLIT/IvanHoe/Igorrit/..., which presumably again would be quite high. This is hardly a form of "bias", but rather a necessary filtration (removing obvious clones) to make the paper sensible and sensitive to the question of move similarity distribution in (presumably)
unrelated engines (and indeed, we discuss what it can really mean to be "unrelated", particularly in the Internet age).
Rebel wrote:in where you fail to mention that EVERY eval ingrediënt between Fruit and Rybka is coded DIFFERENTLY.
This is an utterly irrelevant point. It is trivial that a bitboard engine will be coded "DIFFERENTLY" than a mailbox engine.
This is also why most of the first 48 pages of your Rybka Reloaded (analysing Zach's work) is essentially worthless, as you concentrate on some inapplicable standard called "code theft". The EVAL_COMP analysis was whether the
evaluation features matched, not how they were coded. For the purposes of Rule #2 originality, the exact coding does not matter, as emphasised many times in many places, eg Levy in his ChessBase interview, or me in my rebuttal to Riis.
Chris Whittington wrote:did you look to see what readers were being invited to read. Levy invites readers to form their own opinion after reading ..... "Simply put, Rybka's evaluation is virtually identical to Fruit's."
And after reading Rajlich's reply:
Aside from that, this document is horribly bogus. All that "Rybka code" isn't Rybka code, it's just someone's imagination. But then, maybe it is just my making of excuses for Levy that lead me to quote both parts, rather than just one...
