perlmonker303 / functionalprogramgenerator

Generation of Functional programs using Genetic Programming

Functional Program Generation using Genetic Programming

Developed for the AI in Software Engineering course for my MSc. course.

Features so far:

• parse LISP expressions (operators, conditions, comparisons, null, functions)
• evaluate LISP expression
• defun token + recursive functions
• identify all tokens to use as elements for reproduction in genes
• check validity of S-expression (using "parse_sexp" or pre-existing compiler "SBCL")
• use an existing lisp compiler to ensure program correctness and output (difficult to implement everything)
• considered a different chromosome representation: s-expressions
• mutation techniques: shrink, subtree, swap
• inserted perfect candidate in initial pool and see what happens
• cache fitness computations for candidates without changes - use unique candidate ids
• improved mutation fairness
• add similarity coefficient computation in fitness function (ex: https://www.mdpi.com/2076-3417/11/10/4673)
  • caching won't work this way, cause the fitness of an individual is dependent on all other individuals now

Notes:

• "Null" checks whether a list is empty or not.
• consider idea: take some existing LISP code, evaluate it, scan it, then generate it.
• SBCL command for factorial program sbcl --noinform --eval "(defun factorial (n) (cond ((zerop n) 1) (t (* n (factorial (- n 1))))))" --eval "(print (factorial 3))" --quit --disable-debugger