paul-hoehne/ea

Evolutionary Computation in Golang

This is really a re-implementation of a set of libraries I've developed over the years to perform evolutionary computation. I've had versions in C++, Java, C#, LISP, and even Ada. I'm re-implementing it in Go for a couple of reasons. The first is I just like the subject matter. It's fun to play "guess what converges faster."

The second is because I've always solved this problem with generics. Except for LISP (where I passed in specific algorithms as functions), the easy way to tackle a generic evolutionary computing problem was to parameterize a generic EC framework with specific implementations of behaviors. The behaviors involve how populations are constructed, how reproduction occurs, and how offspring and parents are selected.

While the behavior of specific subtype of evolutionary algorithms might involve sexual or asexual reproduction, mutation, encoding as bit-strings, etc., they all had roughly the same general outline.

1. Generate a candidate population
2. Produce some number of offspring
3. Have the offspring and parents compete of space in the new generation.

Being able to parameterize an EA so as to mimic a (highly explorative) Genetic Algorithm with sexual reproduction and fitness-proportional selection or more of a μ + λ (and highly exploitive) Evolutionary Strategy is possible in a framework that views them as points along a continuum. That allows you to "shift the slider" left or right as you explore a problem space.

So what's this like without generics (which may be coming in Go 2.0). Without opening up a can of worms, generics are actually a hard language feature to implement correctly. (cough cough Java). Even then, not all parameterized constructs are equally generic. For example, a generic parameter may be required to support a specific set of operations or inherit from a specifc parent.

So it's also just interesting to take a problem that I've solved multiple times with generics and solve it with nothing more than interfaces and first-class functions.

ea's Issues

Add individuals that are just bit strings

Explain why genes are the crossover level

Explain why genes and not individual bits are where crossover happens.

Add the ability to specify 1, 2, or n point crossover

Add crossover support.