This repository contains my project for COSC401, a fourth year machine learning paper at the University of Canterbury. See the report for details on the implementation.
At the moment this source is very system specific. My original plan was to produce a genetic programming gem and utilise this for my assignment, but because of time constraints I had to simple hack together a working program. Then I wound up using over 60GiB of memory during my initial testing run and had to throw together a distributed system with Resque. All-in-all this source has become a tangled mess held together with coffee and energy drinks.
If you do wish to try and get this running yourself there are a few pointers I can give you:
-
You will need a Redis database running, Resque requires Redis and the distribution of the population also utilises it.
-
Look at
lib/environment.rbfirst. The Redis location at least will need to be changed. -
run-clients.rbshows how I spawned the workers into a screen session if they weren't already running, should be mostly directly usable. The hostname matching was used to make sure that jobs were only processed on one host to keep the filesystem based logging usable. -
spawn-clients.rbssh's into each host and runsrun-clients.rb, should simply need to change line 3 to a list of your available hosts. -
shutdown-clients.rbandkill-clients.rbare similar to the spawner, they ssh in to each host and then either sendQUIT(shutdown after finishing current job) orTERM(kill child and quit now) to all running Resque workers. NOTE: this doesn't try and find only the workers that are working on this, it will attempt to kill any Resque workers on the host.
The generation of the results used in the report can be found in this gist.