Entry for http://codegolf.stackexchange.com/questions/69930/paint-starry-night-objectively-in-1kb-of-code/

I didn't win the contest, but I'm happy with the score I got, and I subjectively like the resulting image:

I'm putting this in a public GIT repository for fun, and as an exercise. I've used GIT a lot over the past few years, but never had reason to publish anything.

There are several versions of code.c that show the evolution of my solution, 2 of which are worth noting.

code5.c contains the "scaffolding" code that I mention in my entry description. It includes the file v.i with the R,G,B,x,y,r values, modifies whichever fields are uncommented, iterating values {-10,10}, saving the values that produced the lowest score, then writes a new version of the array to v.i. Re-compile and repeat.

code6.c is my final entry, weighing in at 999 bytes.

validate.py is the original validator by the post author.

cvalidate.c is a C version of the validator that I wrote for speed.

StackExchange user @neocpp pointed out a faster filled-circle algorithm in a comment, and I found an example of it here,

http://stackoverflow.com/a/1237519/488095

So I wrote a new version using this other algorithm, and not only does it run 170 times faster than my trig function version, it is smaller and made room to add 1 more circle. So there are now 2 additional versions.

fastcircle1.c includes optimizer.c and regnerates v2.i.

fastcircle2.c is the standalone version, and the contents of v2.i should be pasted into the v[] array.

Here's the general idea of how it works,

$ make cvalidate
$ make fastcircle1 && ./fastcircle1 && ./cvalidate o.ppm
$ ./fastcircle1 optimize
$ make fastcircle1 && ./fastcircle1 optimize && ./cvalidate o.ppm

Repeating the last line will tick down the score slightly with each pass, up to a point where people better at maths than me would say it reaches "local minima".