DiGenO is a Distributed Genetic Optimizer framework in Erlang. It first emerged as a small part of a much larger “Top Secret Research Project” I was working on, but later I realized that it could be broken out and independently released under a permissive BSD license. This turned out to be a wise decision, because it enabled further evolution (no pun intended) of the framework long after the obsolescence of its original surroundings.
The codebase of DiGenO is incredibly small, under 1000 lines. It still handles the transparent addition or deletion of computation nodes, and the automatic distribution of operations that perform the genetic algorithm. It goes to show how compact and well fitting the Erlang language is when implementing distributed computing algorithms.
DiGenO implements a steady state evolution algorithm. By this we mean that after an initial population is built up, subsequent evolution is performed on a continuous basis, swapping out old instances in the population for new ones piecewise. This is in contrast with more conventional evolutional algorithms that iterate by generating a complete successor population at once. They usually do this by keeping a few percent of instances (the “elite”) intact, and filling up the rest by mutations and cross-overs of instances selected from the preexisting pool.
Instead of this, DiGenO generates, evaluates and releases into the population new instance candidates on a continuous basis, thereby performing a more gradual evolution. Each occasion of releasing a newly generated instance into the population is called a reduction. The number of reductions is a metric showing how many steps the genetic algorithm has done, and is similar to (but much larger than) the count of successive populations evolved with a conventional algorithm.
It is easy to formulate problems so that they can be optimized by DiGenO. Read this tutorial if you are interested in trying the framework.