This is the README for fungp, the genetic programming library I am working on as a student at CSUS. The current version is 0.3.2.
If you're reading this as HTML, it was automatically generated from the README.md file in the root of the source repository.
Genetic programming (GP) is the process of evolving computer programs using a process inspired by biological evolution. In GP, a computer program automatically writes new computer programs (in this case, by generating trees of new Clojure code), and judges them according to their ability to solve a problem or produce correct output.
In fungp, like in many GP systems, tree structures representing programs go through processes inspired by biological evolution, such as mutation and reproduction. Their chance of reproduction is decided by their "fitness," which is assigned by a fitness function. For example, a simple problem you might solve with GP is symbolic regression, in which the evolved trees represent mathematical functions mapping one or more inputs to an output, and the fitness function would compile and run the evolved programs on known data to test whether sets of inputs produce correct output.
There are only two hard things in Computer Science: cache invalidation, naming things, and off-by-one errors.
--- Paraphrased from Phil Karlton
fungp is a genetic programming library implemented in the Clojure programming language. It's pronounced fun-gee-pee, for functional genetic programming (or "genetic programming is fun").
A far more detailed explanation can be found in the core.clj source code. It is thoroughly documented. An
HTML document generated from the source is in the
docs/ folder, and is available hosted on my school's
web server here. It's a good place to start and it includes a
getting started tutoral (the source of which is of course available under src/fungp).
You'll need Clojure first, but that's easy. You can be up and running in seconds if you already have Java installed.
First you install Leiningen. This is how you'd do that in a Unix-like system.
mkdir ~/bin wget https://raw.github.com/technomancy/leiningen/preview/bin/lein mv lein ~/bin chmod 755 ~/bin/lein lein repl
~/bin is on your
$PATH. The lein executable could go anywhere as long as it's on your path, but I'd recommend you put it somewhere in your user's home directory. Use
echo $PATH to see what directories are on your path, or read this article to learn how to edit Unix environment variables.
Because you don't need root/sudo access to do this, it works fine on limited user accounts. To my friends at CSUS, that means it runs fine on Athena.
Once you have Clojure and Lein running you can grab fungp:
git clone https://github.com/probabilityZero/fungp.git cd fungp
You can also list fungp as a dependency in your
project.clj and make lein install it for you:
(defproject testfungp "0.1.0-SNAPSHOT" :description "FIXME: write description" :url "http://example.com/FIXME" :dependencies [[org.clojure/clojure "1.4.0"] [fungp "0.3.2"]])
lein deps and lein will download the library.
See the samples for example usage (located in src/fungp/samples). The general idea is to pass in a fitness function, a terminal set, a function set, and other various options, and the algorithm will attempt to evolve code out of the functions and terminals.
To use fungp you'll likely want to take advantage of Clojure's REPL (read-eval-print-loop). Once you have installed lein (see above), you can start the REPL with
lein repl and start using fungp. For more detail on what to do from there, read the documentation (either in the
docs/ folder, in the source code, or on the website linked above). You can experiment with the included sampls or run the tutorial code like this:
=> (use 'fungp.tutorial) nil => (test-genetic-program 5 10) ... => (use 'fungp.tutorial :reload-all) ;; reload file with changes
If you're starting out with fungp and Clojure, I recommend you start by copying and modifying the tutorial code or one of the samples. If you make a copy, be sure to give it a new name and namespace.
The run-genetic-programming function (the function you call to start the search) accepts the following options keywords:
- iterations : number of iterations between migrations
- migrations : number of migrations
- num-islands : number of islands
- population-size : size of the populations
- tournament-size : size of the tournaments
- mutation-probability : probability of mutation
- mutation-depth : depth of mutation-generated sub-trees
- max-depth : maximum depth of trees
- terminals : terminals used in tree building
- numbers : number literals to be used in tree building
- functions : functions used in tree building, in the form [function arity]
- adf-count : Number of automatically-defined functions
- adf-arity : Number of arguments for automatically-defined functions
- fitness : a fitness function that takes a tree and returns an error number, lower is better
- report : a reporting function passed [best-tree best-fit] at each migration
Some minimal knowledge of Clojure is probably necessary to use this library well. Not to worry, though! Clojure is a rather nice language, with excellent documentation and many great tutorials (like this one).
Basically, fungp uses a process of evolution (mimicing natural selection in nature) to create and rewrite Clojure code. Again, for a complete explanation look to fungp.core, or read Wikipedia's explanation of Genetic Programming.
Here's a sample output from a symbolic regression problem. Reports are printed to the screen periodically, not every generation.
fungp :: Functional Genetic Programming in Clojure Mike Vollmer, 2012 Test inputs: (-10 -8 -6 -4 -2 0 2 4 6 8) Test outputs: (300.0 192.0 108.0 48.0 12.0 0.0 12.0 48.0 108.0 192.0) (fn [a] (let  (- (+ (- (inc (inc 9.0)) (fungp.util/sdiv (dec a) (* a 0.0))) (+ (- (fungp.util/abs a) (- a a)) (inc (* a a)))) (dec (* (- (dec a) (inc a)) (+ (+ a 7.0) (* a a))))))) Error: 10210.0 (fn [a] (let  (+ (- (inc (inc 9.0)) (fungp.util/sdiv (dec a) (* a 0.0))) (+ (+ (inc (fungp.util/abs a)) (+ (+ (- (inc (inc 9.0)) (fungp.util/sdiv (dec a) (* a 0.0))) (+ (- (fungp.util/abs a) (- a a)) (inc (* a a)))) (fungp.util/abs (fungp.util/sdiv (fungp.util/abs 8.0) (dec 3.0))))) (inc (* a a)))))) Error: 6778.0 (fn [a] (let  (+ (* a a) (+ (+ (- (fungp.util/abs a) (- a a)) (inc (* a a))) (inc (* a a)))))) Error: 580.0 (fn [a] (let  (+ (* a a) (+ (* a a) (inc (* a a)))))) Error: 10.0 Done! (fn [a] (let  (+ (* a a) (+ (* a a) (* a a))))) Error: 0.0
Project created by Mike Vollmer and released under GPL. See the LICENCE file distributed with this code.