Crafty accused of copying Fruit PST

[oudheusa]

See http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=22635

[thorstenczub]

i wonder if bob is always working by himself on crafty, or if he has other people working on crafty in a team...

[kingliveson]

Foxnews has a negative story about Obama...

[BB+]

Yeah, I was wondering too about why Fruit/Crafty PST was in a "Rybka discussion" subforum, but I tend not to view information merely from the medium (for that matter, a pedant might note the title should say opening bishop PSTs, but I digress). The thread has already been spammed enough for me to have difficulty following the content.

If indeed Bob (or one of his subsidiaries) did derive Crafty's opening bishop PST via starting with the Fruit value and adding 8 [or maybe starting with Fruit's numbers, and then tuning via constant shifts], then this should be stated somewhere. Some "exterior" evidence for a specific Fruit role here would be that Crafty 22.2 (about 7 months after the previous sub-version) lists: Fruit-like scoring (interpolation between mg and eg scores) fully implemented. As Bob himself stated (I think), the bishop endgame PST is not strictly like the Fruit bishop endgame PST, though one might note that the Crafty PSTs at this point were simply the same for opening/endgame.

The same 22.1->22.2 changes also had

Code: Select all

int nval[64] = {
  -60, -29, -29, -29, -29, -29, -29, -60,
  -30, -24, -10, -10, -10, -10, -24, -30,
  -30, -6, -6, -6, -6, -6, -6, -30,
  -30, -6, -2, 0, 0, -2, -6, -30,
  -30, -6, -2, 0, 0, -2, -6, -30,
  -30, -6, -6, -6, -6, -6, -6, -30,
  -30, -24, -10, -10, -10, -10, -24, -30,
  -60, -29, -29, -29, -29, -29, -29, -60,
};

Code: Select all

int nval[2][2][64] = {
    -30, -20, -20, -10, -10, -20, -20, -30,
      0,  10,  16,  20,  20,  16,  10,   0,
      0,  12,  20,  24,  24,  20,  12,   0,
      0,  12,  20,  24,  24,  20,  12,   0,
      0,  10,  18,  20,  20,  18,  10,   0,  /* [mg][black][64] */
      0,   0,  16,  14,  14,  16,   0,   0,
      0,   0,   0,   0,   0,   0,   0,   0,
    -20, -20, -20, -20, -20, -20, -20, -20,

Code: Select all

int qval[64] = {
  -20, -20, 0, 0, 0, 0, -20, -20,
  -20, 0, 8, 8, 8, 8, 0, -20,
  0, 8, 8, 12, 12, 8, 8, 0,
  0, 8, 12, 16, 16, 12, 8, 0,
  0, 8, 12, 16, 16, 12, 8, 0,
  0, 8, 8, 12, 12, 8, 8, 0,
  -20, 0, 8, 8, 8, 8, 0, -20,
  -20, -20, 0, 0, 0, 0, -20, -20,
};

Code: Select all

int qval[2][2][64] = {
      0,   0,   0,   0,   0,   0,   0,   0,
      0,   0,   4,   4,   4,   4,   0,   0,
      0,   4,   4,   6,   6,   4,   4,   0,
      0,   4,   6,   8,   8,   6,   4,   0,
      0,   4,   6,   8,   8,   6,   4,   0,   /* [mg][black][64] */
      0,   4,   4,   6,   6,   4,   4,   0,
      0,   0,   4,   4,   4,   4,   0,   0,
      0,   0,   0,   0,   0,   0,   0,   0,

Pawn values stayed the same, as did the various king value arrays. At a glance, I don't immediately see any Fruit influence with knights/queens. Other numerical changes were in outpost scoring, and also rooks on open files.

[kingliveson]

I have no idea why Bob degrades himself to such lows by trying to have a civilized discussion with some of our friends over there of which 99.9% could not write "hello world" if their lives depended on it. Here's one reply:

For you l'll bet that was true! Coffee break is over Bob!

P.S.

I was checking out the forum yesterday Bobo!

That said, I hate to burst the bubble for those who think "aha, you too!," there's really nothing to see here. And no, this was not what Rybka was found guilty of doing. Both programs are open-source and as BB+ mentioned, Bob needs to note/acknowlege the derivation of, or influence of the Bishop PST according to GPL license which Fruit is released under.

GNU General Public License, version 2:

Code: Select all

a) You must cause the modified files to carry prominent notices
    stating that you changed the files and the date of any change.

edit: From The GPL:

Code: Select all

These requirements apply to the modified work as a whole. If identifiable sections of that work are not derived from the Program, and can be reasonably considered independent and separate works in themselves, then this License, and its terms, do not apply to those sections when you distribute them as separate works. But when you distribute the same sections as part of a whole which is a work based on the Program, the distribution of the whole must be on the terms of this License, whose permissions for other licensees extend to the entire whole, and thus to each and every part regardless of who wrote it. 

[hyatt]

Some things never worked for me. Tracy took a look (specifically) at Fruit's queen scoring since we did almost nothing in that regard. First thing he noticed was mobility. We tried Fruit's simple approach (count squares * constant). We tried our weighted approach where attacking central squares is worth more than edge squares, etc. Not one dang thing worked for queen mobility as it appeared to in Fruit. However, we were doing what is commonly called "second order" evaluation with respect to queens, which is namely that the queen scoring is dependent upon the opponent's king safety and how many other of our pieces are close by to attack. And it seems that doing that made the mobility term worse than useless. 22.2 was, I believe, a significant Elo jump. That was our first formal "tuning" using the cluster. We played roughly .7 million games per day for at least 6 months, tuning each piece's material value, each scoring term, and even overall scaling on some of the PST arrays. Huge number of different values tested. Where the Fruity bishop PST values came from is, for the moment, unknown. At the time, we did not have a bishop PST if I recall, so I had to write some code to "populate" the arrays using a minimum number of terms (I think I used something like "mult" for the per-rank/file bonus, offset, which set the smallest value to that and everything else was relative to that starting point (values from -20 to 0 do behave differently than values from 0 to 20, because when they are added in evaluation, they can change the relationships between the value of each piece type), and "diag" which was just a main diagonal bonus. I had done this multiple times in the past to try to encourage keeping the bishop on the critical diagonal when you play a fianchetto type opening (g3/Bg2, then you look pretty stupid when you promptly play Bg2xNc6 and now you have critical weaknesses around your king). I played a lot of matches, trying different values, and based on my notes, none of them helped and the code was supposed to be removed. I am re-running the test right now with the bishop PST code (2 lines in evaluate.c) commented out to see if it changes the Elo at all...

The "ploy" here was to divert attention. I have exactly one table that looks like one from fruit. How it ended up that way I do not know. Rybka has every PST looking suspect since they can all be produced perfectly by taking fruit PST initialization code and changing the basic multipliers for ranks and files and back rank and whatever. Supposedly, my having 1 is just as bad as Rybka having all 10 match. Even though my evaluation doesn't look like Fruit's at all, while Rybka's does. But it was a nice attempt to divert attention since everyone had apparently given up on refuting Zach's PST value comparisons...

[Samer]

and this stupid guy post here...

Uly
Crafty a clone of Fruit.

Also, michiguel just took a quick, casual look and found that, I think that if Crafty was investigated (say, to the extent that Rybka was) they could find more similarities.

For me and many people, Bob is a legend...
but for stupid guys is a cloner.
Shit !!

[hyatt]

I'm certainly surprised by the bishop PST issue. I'll be more surprised if there is other code that is similar.

I'm certainly not concerned since I have written 99% of the code in Crafty...

[Uly]

Thanks for the praise Samer.

I was actually replying to someone that asked that if a PST table was both in Fruit and Crafty (+8), that from where the table was coming from. It came from Fruit.

But saying "Crafty a clone of Fruit" does that in fewer words (he knew what I meant when I read that), even though "clone" doesn't actually have a definition besides genetics.

*Checks dictionary*

Oh wait, it does:

Clone:
noun: an unauthorized copy or imitation

I don't think bob was authorized to use Fruit's PST table in Crafty, so from this angle, it counts.

And I stand for what I said, Crafty hasn't been scrutinized because people thought they wouldn't find anything a priori. Now that something has been found, maybe that will change.

[veritas]

Thanks for the praise Samer.

I was actually replying to someone that asked that if a PST table was both in Fruit and Crafty (+8), that from where the table was coming from. It came from Fruit.

But saying "Crafty a clone of Fruit" does that in fewer words (he knew what I meant when I read that), even though "clone" doesn't actually have a definition besides genetics.

*Checks dictionary*

Oh wait, it does:

Clone:
noun: an unauthorized copy or imitation

I don't think bob was authorized to use Fruit's PST table in Crafty, so from this angle, it counts.

And I stand for what I said, Crafty hasn't been scrutinized because people thought they wouldn't find anything a priori. Now that something has been found, maybe that will change.

ALL your "hot air " and red herrings do NOT distract from simple FACT

Vas is a REAL CLONER , PLAGIARIZER . CHEAT & BARE FACED LAIR , a bully who thought to toss dirt at others and treat most as dirt who has been caught red handed by his arrogance and stupidity
or from the fact that you and Prima are a pair of muppets


have a nice day