Todo

- Make README in rst or markdown or similar

- Use compile if the evolved function will be used multiple times

- Let fitness return tuples to allow e.g. parsimony preassure

- Enable line breaks in the rules when parsing, i.e. avoid everything on one line

- Print stats to file

- Add a cmdline parameter and code to set the save-directory (and make
  it automatically if necessary)

- Look at Rangoli: what are the rules? Can we learn from them?
  http://en.wikipedia.org/wiki/Rangoli

- In L-systems, the main command which changes state without drawing
  is rotate (aka +/-). The key thing about this is, you can have lots
  of rotate commands either together, or spread out through the
  program. Rotation "wraps around", so if you have "too many" +
  commands, you just end up back at the start. This is a crucial
  property because it is in the nature of L-systems to sometimes give
  too many of a particular command. Our colour scheme (now) does the
  same: it's circular, going from colour1 -> colour2 -> black ->
  colour1 in 100 steps. The scheme for changing the line-width should
  do the same. Or maybe there should be just a few possible
  line-widths.

- Add Jennifer's PonyGE code as a branch.

- Add James' fixed-steps bricks code as a branch.

- July 19 (Erik): add a "spawn" command which can be part of an
  L-system. When executed, it presumably pushes onto the stack, and
  starts a new L-system which inherits some similarities from the
  existing one. Maybe it uses a smaller line-step, to keep the
  children L-systems smaller? Maybe not.

- July 19 (James): "random canvas". Idea is to evolve a picture as a
  population rather than a single individual. We remove the grid and
  allow individuals to have arbitrary xy origins. User can select and
  kill an individual. A new individual can be added by crossover from
  selected individuals.

- Use a parser/generator library, maybe
  http://www.acooke.org/lepl/features.html, to experiment with
  reverse-mapping from user edits and grammar induction.

- Contact the company who made the Irish EU presidency logos and see
  if they'd be interested in some kind of endorsement or field-test:
  http://www.irishtimes.com/newspaper/breaking/2011/1130/breaking53.html.
  The company was http://www.reddog.ie/

- Keep generating individuals until a picture is actually drawn, ie
  invalid and null pictures should be replaced.

- For the symbolic regression and boolean problems, use Numpy.

- Try using a quotient operator
  http://www.cs.bham.ac.uk/~wbl/biblio/gp-html/Ni_2012_ieeeTEC.html

- Use logging and pdb and pylint for nicer
  code. http://www.reddit.com/tb/rwj1a

- Extended BNF (Erik)

- Force Fields (Jonathan)

- 3D (Eoin)

- Generalise the GUI so that it can be used for any type of
  interactive EC -- maybe a pure web-GUI in JS/HTML5/WebGL? (James)

- Murphy Cup and enter GECCO art competition

- A button/shortcut for saving desktop backgrounds and screensavers

- Web interface

- Dave suggests: a library of shapes etc that could be used to inject
  specific structures. Given an existing structure, the user decides
  that a specific new shape is needed, so goes to the inject menu to
  select from some predefined shapes.

- Miguel suggests the mutation rate should be inversely proportional
  to the effective genome length (it is currently just per-codon).