using context-free grammars to generate CDFG files

[futrelle]

This grammar:

startshape r2
rule r1 0.753 {
SQUARE { x 0.06 y 1.89 s 0.77 r 1.350 b 0.7 }
SQUARE { x -1.180 y -0.943 s 0.46 r -97.1 b -0.4 }
}

rule r2 0.624 {
r2 { x 1.72 y 0.9 s 0.123 r -0.599 b 0.4 }
CIRCLE { x -0.354 y 1.833 s 0.74 r -42.60 b -0.1 }
r2 { x -0.758 y 1.6 s 0.7217 r 2.75 b 0.5 }
}

rule r2 0.948 {
r1 { x -0.9 y -1.5 s 1.9565 r 03.43 b 0.7 }
SQUARE { x -0.6 y -1.3 s 1.75 r 79.5 b 0.2 }
r1 { x -1.531 y 0.247 s 0.676 r 3.5 b -0.7 }
r2 { x 0.677 y 1.988 s 0.84 r -7.807 b -0.6 }
CIRCLE { x -0.3 y -0.218 s 0.7746 r -0.959 b 0.7 }
}

rule r2 0.902 {
r1 { x 0.11 y 1.9 s 0.6518 r 59.179 b -0.9 }
r2 { x -0.508 y -1.9 s 0.805 r 62.77 b 0.2 }
CIRCLE { x -0.560 y -1.20 s 0.9863 r 9.05 b -0.4 }
CIRCLE { x 0.513 y -0.0 s 0.9628 r 78.1 b 0.2 }
}

was generated by this rmutt grammar. It's ugly, because rmutt doesn't directly support random numbers:


grammar:
"startshape " rulename "\n"
rule{4}
;

rule: "rule " rulename " " prob " " rulebody "\n";

rulename: "r" ("1"|"2");
rulebody: "{\n" shape_clause rulename_clause* "}\n";
shape_clause: shape " " shapemod "\n";
rulename_clause: rulename " " mod "\n";
clause: shape_clause | rulename_clause;
shape: "SQUARE"|"CIRCLE";
mod: "{ x " nxy " y " nxy " s " ns " r " nr " b " nb " }";
shapemod: "{ x " nxy " y " nxy " s " nss " r " nr " b " nb " }";
nxy: "-"? "0." digit{1,3};
ns: "0." highdigit digit{1,3} &20 | "1." digit{1,3};
nss: ("0"&4|"1"&2|"2") "." highdigit digit{1,3};
nr: "-"? digit{1} "." digit{1,3};
prob: ns;
nb: "-"? "0." digit;
highdigit: "9"&9|"8"&8|"7"&7|"6"&6|"5"&5|"4"&4|"3"&3|"2"&2|"1";
digit: "0"|"1"|"2"|"3"|"4"|"5"|"6"|"7"|"8"|"9";

[chris]

That's fun!

I actually have a random cfdg generator now, along with a "mutator" program that randomly changes things in cfdg's (both rules and values), and I've been toying around with a genetic selection process to make cool images. The last part isn't working so well.. How do I tell if something is cool? Heh...if it starts producing anything of value, I'll release it sometime soon.

[mtnviewmark]

How do I tell if something is cool? Heh...

You generate sixteen genetic variations, display them on the screen for the user, and let the user pick one, two or three. Then you proceed from those.

[chris]

How do I tell if something is cool? Heh...

You generate sixteen genetic variations, display them on the screen for the user, and let the user pick one, two or three. Then you proceed from those.

I was considering something like that, with a web interface...which is actually a lot like something I saw in a science museum once. Another variation (which I've already made) is one that compares them to a jpeg, until it can find a grammar that draws your face -- or whatever -- well. It's gotten horrible results so far, though. I need to get smart about analyzing the jpeg.