Hi Pilou, 
Making the number of points, less 1, in the script, divisible by 4, results in 4 long lines at angles PI/2, PI, 3PI/2, and 2PI. 
Then the "radial" curves seem to be more symmetric. 
There are 4 quadrants, so being congruent to 0 modulus 4 seems helpful. 
The original script uses 21, 161, and 101, which give quite a lot of curves, and after fixing up the outer perimeter, are network-able. 
I think this is because the values near these angles change a great deal with small perturbations. 
The equations are not nice and uniform. 
 
- Brian 
 |