Programmers Open Letter to ICGA on Rybka/Fruit

[Uly]

(this assumes Rybka 2.3.2a is the last one that contains any significant parts from Fruit).

Does it contain parts of Fruit?

[BB+]

(this assumes Rybka 2.3.2a is the last one that contains any significant parts from Fruit).

Does it contain parts of Fruit?

IMO, yes, the evaluation function from Rybka 2.3.2a is sufficiently similar to that for Rybka 1.0 Beta that I would still say it is "non-original". See this post for a brief summary, and the attached disassembly therein (to compare to the disassembly for Rybka 1.0 Beta). If this becomes an item for dispute, I will be more thorough and quantify this more.

Rybka 3 still has the "0.0" usage in integer-based time management code (which seems to come from Fruit), but this is so picayune for chess purposes that it shouldn't matter.

You can note that Rybka 2.3.2a was released on June 18 (about five days after 2.3.2), and the 2007 WCCC was June 11-18.

[BB+]

See this post for a brief summary

Another minor point that I just noted is that the criterion for a 7th rank rook bonus in Rybka 2.3.2a is to have an opposing king/pawn on the 7th/8th, while in Rybka 1.0 Beta it requires the opposing king on the 8th or a pawn on the 7th. Similarly with queens. Also, the "can-castle" criterion with pawn shelter/storm is no longer in the main eval, but it conceivably could be in pawn hash (especially if this was really "king location plus pawn hash" by this point). As I say, I will have to check to be sure.

[hyatt]

See this post for a brief summary

Another minor point that I just noted is that the criterion for a 7th rank rook bonus in Rybka 2.3.2a is to have an opposing king/pawn on the 7th/8th, while in Rybka 1.0 Beta it requires the opposing king on the 8th or a pawn on the 7th. Similarly with queens. Also, the "can-castle" criterion with pawn shelter/storm is no longer in the main eval, but it conceivably could be in pawn hash (especially if this was really "king location plus pawn hash" by this point). As I say, I will have to check to be sure.

In older versions of Crafty, that was the "enabler" for a rook on the 7th bonus. King on the 8th or a pawn on the 7th. There was a point in time where we evaluated "absolute 7th, which is best explained with a king on the 8th and rook on the 7th with no enemy pieces on the 7th. The king will not get out into the game and become active very easily.

But the Rybka 1 beta rook on 7th sounds to be what Crafty did:

Code: Select all

    if (Rank(square)==RANK7 && (BlackKingSQ>H7 ||
                                BlackPawns&rank_mask[RANK7])) {
      score+=ROOK_ON_7TH;
      if (tree->pawn_score.passed_w && BlackKingSQ>H7 &&
          !(BlackPieces&mask_abs7_w)) score+=ROOK_ABSOLUTE_7TH;
      if (AttacksRank(square)&(WhiteRooks|WhiteQueens))
        score+=ROOK_CONNECTED_7TH_RANK;
    }

This in version 19.x...

[kingliveson]

Surprisingly, Bob's post on Rybka forum showing Rybka took code (not ideas) from Crafty has not disappeared.

[hyatt]

I think that "the cat is out of the bag" now. Deleting posts probably seems worse than letting them stand...

[benstoker]

I think that "the cat is out of the bag" now. Deleting posts probably seems worse than letting them stand...

I haven't gone over there, but I would hope that the owner of the Rybka site would respond to a post on his own site indicating clear evidence of code copying. Surely he has responded at length, explaining how Bob and others are completely mistaken. Surely. You know ... being his website, his forum ...

[BB+]

You know ... being his website, his forum ...

I'm not sure Rajlich actually owns the site/forum. It could be Convekta (for instance). Dadi Jonsson administers the forum, and Felix Kling the website AFAIK.

[Uly]

From what I know (and I could be wrong) Felix decides on the forum policy. So it's up to Felix to decide what gets deleted and what doesn't.

I'm sure if it was Vas's decision, mostly everything would stay (i.e. mentioning the "derivatives"'s names would be no problem).

[Angel]

http://ippolit.wikispaces.com/message/v ... 9/35036490

The Rybka Cloning Issue

Anthony Cozzie