Genetic algorithm defined on generic grammars, currently learns LISP functions (anti-skynet weight included)

Firma defines a simple genetic algorithm which uses a traditional programming language grammar. The current implementation generates LISP functions, but is written to use any programming language in principle. All that is needed for a new language both a generate() and an interpret() function, where generate() generates a random piece of code, and interpret() compiles and runs that code.

This project is based on my other project, sence.

For example, suppose we want to learn the following function:

def f(x):
    return 2 * x + 1

The function genetic(f) in genetic.py can be used to learn it as follows:

genetic(f)

Resulting in the potential output:

grid {firma}: python genetic.py
Selected:
('expression',
 [('operator', '+'),
  ('expression', [('operator', '%'), ('atom', 'x'), ('atom', 6)]),
  ('expression', [('operator', '+'), ('atom', 1), ('atom', 'x')])])