may 7
genetic algorithms
Last time there was interest in discussing how
"evolutionary algorithms" work.
The name comes from thinking of biological evolution
as a search for the "fittest" DNA sequence.
Consider each member of a population as a potential
solution to a problem you're trying to solve. Then
seaching for a good solution to the problem may be
done in a manner than mimics evolution.
The general approach looks like this:
(1) Generate a population. (Typically random initially.)
(2) Grow the population either by combining individuals
(e.g. sexual reproduction) or by mutating individuals
or both. Essentially this is a "local" search, searching
for small incremental improvements.
(3) Shrink the population by keeping the best.
("Survival of the fittest.")
(4) Repeat steps (2) and (3) many times.
For a good example, google search "genetic traveling salesman":
To run TSPGA (Johannes Sarg's 1998 Java visualization) on cs:
$ cd TSPGA
$ java TSPGA
# Simulation > Initialize ... (Set up default values)
# Options > Set Problem... (add towns)
# Options > GA Parameters... (mutate, recombine options, population size, ...)
# Simulation > Generation... (one time through grow/shrink loop; cntl-shift-G)
# repeat last and/or Simulation/Run...
To run pygene's TSP on cs:
$ cd pygene/pygene-0.2.1
$ ./demo_salesman.py
gen=0 best=6654.28258914 avg=7432.50187619
gen=1 best=5158.4389665 avg=6204.96477575
...
cntl-C
Best solution: ...
Running Steven Tanimoto's lisp version
(from Hampshire 2000 AI course; see comments at end of source code for invocations) :
$ cd lisp-TSP
$ clisp
[1]> (load "genetic-salesman")
T
[2]> (ga-salesman *sp* 10 2)
[3]> (ga-salesman *sp* 20 20)
There are lots of examples of this out there, including
general purpose libraries for dealing with problems using
this method.
Unix libraries :
$ apt-cache search genetic
python-genetic - genetic algorithms in Python
libevocosm-dev - a C++ framework for developing evolutionary algorithms
...
