Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Crafty/Fruit projection
1 2 Previous Next  
Parent - - By Banned for Life (Gold) Date 2013-09-05 04:46 Edited 2013-09-05 04:49
Rybka 1.4 - 1.6.1 DID cache align.

Rybka 1.4 through 1.6.1 were testbeds that didn't use Vas' TT code, as you well know. You're naturally clueless, so it's really disconcerting when you make an effort to be even more clueless than normal.

The next versions seen did NOT cache align.  Wonder why?  Aha.  The "better" 2-bound table?  Alignment would STILL be a good thing.

No, I don't wonder why. I had discussions with Vas about this in the early days. He wasn't too concerned about the minor slowdown due to accessing multiple cache lines. It turned out that the slowdown was actually very small on top of the random memory access and high frequency TLB page walks. Maybe it's worth a couple Elo. Vas had bigger fish to fry. When he finally fixed the problem, it was more to reduce memory utilization in a relatively new MP configuration than to reduce latency.

I am not full of anything other than certainty about that code's contents.  You NEVER saw any mtd(f) versions, yet you can make statements of fact?  :)

Absolutely. I deal with experts in my field all of the time. They never show me their source code, but I know what they are working on, and have worked on in the past. Vas has spent a fair amount of time working on MTD(f). I doubt that Fabien or Tord would dispute this fact.

The Socrates argument demonstrates a logical fallacy included in your argument. Rybka used floats. Fruit used floats. Therefore Rybka is Fruit. Of course this went way over your oversized head. No surprise there...
Parent - - By bob (Gold) Date 2013-09-05 17:28
The argument was, "Fruit used floats correctly, Rybka copied the float usage, but had integer variables, making the float usage incorrect.  Therefore this is just another clue that code was copied."

Not fallacious at all.
Parent - - By Ugh (*****) Date 2013-09-05 17:44
Clues that code was copied, are just that: CLUES.

If you then find NO copied code, then they remain CLUES only, worthless clues in fact, clues that lead nowhere.

Simple, my Dear Watson.

If you'ld like it in black and white, you need code that was copied to prove that Fruit was copied. And you need quite a lot of it too. So far, your only shot that gets anywhere near "copied code" is your hash tt store. There's evidence that the providence of that is wrong for your theory, and there's evidence that Vas knowledge in that region was so high that it is just foolish to talk of "copy". Talk to Ed, he studied it and produced good evidence against your simplistic Vas-copy-modify BS.
Parent - - By bob (Gold) Date 2013-09-05 18:53
There is code that was copied.  I notice everyone has exited stage left on the TT table code?  Why?
Parent - - By Ugh (*****) Date 2013-09-05 19:00 Edited 2013-09-05 19:02
There is no copied code, your best and only shot is the tt code and Ed has been busy refuting you on that. I didn't see you refute his refutations yet.

I noticed you left stage left on the BS non literal copied mobility code.

And the big question remains. How did you get a guilty verdict and a public humiliation and a business destruction through in 2011 with no code copied to show at all? Oh, yes, it was a witchhunt with predetermined result.
Parent - - By bob (Gold) Date 2013-09-06 01:03
Everyone was convinced about the rybka/fruit connection due to the evaluation information that was published in the evidence.  It is MORE than enough.

You guys keep right on "refuting".  Only it seems you have a completely different definition of "refute" than what most of us use.  But then you have used other oddball definitions of well-known terms as well.  Ed's "semantic equivalence" applied to one line of code comes to mind...
Parent - - By turbojuice1122 (Gold) Date 2013-09-06 01:11
You mean, the evaluation information that didn't compare Rybka with any programs of even a remotely similar level, but instead showed that Rybka was quite different from programs rated many hundreds of Elo points weaker?  That evaluation information?
Parent - - By bob (Gold) Date 2013-09-06 01:29
No.  The evaluation in Zach's report, numb-nuts...

THAT information.  Which compared rybka 1.0b to Fruit, or Mark's report which compared Rybka 2.3.2 to Fruit.  THAT information.
Parent - - By turbojuice1122 (Gold) Date 2013-09-06 01:55
The information in Mark's report was what I was referring to.  It should be obvious that you can't make statistical comparisons that way and use them to conclude that one engine was derived from another one.
Parent - - By bob (Gold) Date 2013-09-06 18:44
I don't see why not.  The more similar, the less original...
Parent - - By turbojuice1122 (Gold) Date 2013-09-06 21:04
The problem is that there was no basis for comparison.  Other programs of similar strength were not compared.  It would be reasonable to hypothesize that nearly all strong programs of that era and after would have very high correlations because they have many concepts derived from Fruit (nothing wrong with that).  I also recall that even if there were strong, modern engines, the statistical methodology itself was quite flawed, though I don't remember all of the details; Alan probably does.
Parent - - By user923005 (****) Date 2013-09-06 21:42
Then engines chosen to perform the comparison was beyond absurd.
All of them were weak sisters of the poor compared to Rybka and Fruit.
I complained about that in my initial letter.

This is what I said:

The use of weak programs like Resp, Phalanx, Faile, Pepito and Exchess for comparisons is literally absurd.
It is like saying, “This Fairlane 500 does NOT have sodium cooled valves, but BOTH the Ferrari and the Porsche open wheeled roadsters DO have sodium cooled valves!”

The comparison was beyond stupid, mindless and sluggish.  It was literally incompetent to do that.  I would certainly have expected more from that particular committee.
Parent - - By user923005 (****) Date 2013-09-06 21:46
I should mention that ExChess has put on a couple hundred Elo since that comparison was made.
The modern ExChess would have been more of a fair comparison.

All of the engines used were also engines that sat for a very long time without any modification from their authors (at the time).

Everything was wrong with using those engines as a model.  I can't believe that they even did it.
Parent - - By bob (Gold) Date 2013-09-07 00:08
One needed to compare with "contemporary" engines of the time.  Where any OTHER examples of copying would be apparent.

Made perfect sense.  Strength is not the issue, it is what is inside the code.  One can take a program like Crafty, which was (in that time frame) the strongest open source engine around until Fruit, and turn it into a dud.  IE Rybka 1.4-1.6.1...  but it was STILL a copy.  You compare to what was available at the time.
Parent - - By user923005 (****) Date 2013-09-07 00:15
I can't believe that this is even coming out of your mouth.  How can you say that without feeling deep, heartfelt shame?
Parent - - By bob (Gold) Date 2013-09-07 01:00
If you are looking at a patent infringement suit today, but the infringement, although recently discovered, happened 10 years ago, WHAT do you look at?  Today's products?  Or those from 10 years ago.

No idea what you are thinking, but I see nothing shameful at all.  This was a study of contemporaries of fruit 2.1, circa 2005, nothing more nothing less.  Seems to be completely correct to me...  Why on earth would you compare evals from programs written 8 years ago with those of today???  Wouldn't you expect 8 years of differences?
Parent - - By user923005 (****) Date 2013-09-07 01:18
Comparing Fruit to any of the engines in the comparison other than Rybka is comparing a Volkswagen beetle to a formula one race car and you know it.
What!? Rybka has more and better evaluation terms like Fruit does!!!  The others don't!
What a shock!!

It really makes it look like you were trying to fix the outcome of the investigation in a disgusting and dishonest way.
Probably, you simply failed to think it through.  But it looks really, really bad.
Parent - - By bob (Gold) Date 2013-09-07 14:38
If you are trying to determine if someone ripped off a 1956 Chrysler New Yorker design, by looking for similarities, would you compare cars from the 50's or from the 2000's?  Surely you see the difference.  In 1956, every manufacturer came out with a new body style, every year.  New engines.  New transmissions.  New interiors.  New dimensions.  Etc.  Compare that to the 2000's where things change every 5-7 years.

It is not about the "strength".  It is about the "parts".

What looks REALLY bad is your not thinking about how things SHOULD be compared.  If you suspect A copied B, in 2005, you would NATURALLY look at C, D, E, ..., Z from that same time period to see how similar THEY looked.  NOT programs from almost 10 years later...
Parent - - By Banned for Life (Gold) Date 2013-09-07 18:55
Your argument is just plain silly. Nobody should be surprised that Rybka's evaluation was much closer to Fruit's than to engines that were 500 Elo weaker, just as you wouldn't learn much about the 1956 Chrysler New Yorker by studying the cars that Juan Manuel Fangio and Stirling Moss were driving in Formula 1 that year.

Anthony Cozzi noted that Rybka had a very traditional evaluation function. It would have been a valid test to see how three top engines, Fruit, Rybka, and Zappa, stacked up in terms of evaluation features. Of course you guys weren't looking for a valid test...
Parent - By user923005 (****) Date 2013-09-07 19:23
Even Crafty would have been a much better test.
Parent - - By RFK (Gold) Date 2013-09-06 20:12
Hey! He called you "numb-nuts"! That is a first!  :lol:  :roll: You're pissing this guy off! :smile:
Parent - By turbojuice1122 (Gold) Date 2013-09-06 21:05
Yes, that was quite entertaining; didn't bother me one bit.  His majesty, the baby! :lol:
Up Topic Rybka Support & Discussion / Rybka Discussion / Crafty/Fruit projection
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill