ECF - Tutorial (v 1.0)

1. In short - an overview (below)
2. Minimal examples – by genotype (or “I don't care about evolutionary algorithms, I just want to solve my problem”)
3. Customizing the Tree genotype (or “Where's GP? I want genetic programming”)
4. Using the parameters (or “Hm, it doesn't work that well, can I make it any better”)
5. Adding components to ECF (or “I know EC, I just want to use my special genotype/algorithm/whatever, how do I do it”)
6. Using parallel ECF (or "Can it go any faster")

In short...

To use ECF, you can combine/parametrize the following components:

• Genotype. The basic component of ECF is Genotype. Each individual in the population has one or more genotypes – any genotypes of any number. The simplest way of defining genotypes is listing them (and their parameters) in the configuration file. Genotype is the only component that has to be defined in a configuration file – all the others may be used by default.
  Each genotype in ECF has its data structure and associated genetic operators – of crossover and mutation type (so there's no possibility of using wrong operator on wrong genotype). Knowing the genotype's data structure is a requirement for defining the fitness function that will evaluate the individual containing the given genotype.

• Algorithm. The ECF runs an algorithm. The desired algorithm (and its parameters) can be stated in the configuration file.

• Evolutionary system – the state. Algorithm and genotypes exist in a context that takes care of everything else: the size of the population, termination condition, logging etc.

The above components are fully configurable without compilation via configuration file. The only element you need to program and define programmatically is the evaluation operator which embodies the fitness function. Once you write the evaluation operator, you only need to reference it in main and then run the ECF. So the typical main function might look like this:

int main(int argc, char **argv)
{
        StateP state = static_cast<StateP> (new State);
        state->setEvalOp(static_cast<EvaluateOpP> (new MyEvalOperator));

        state->initialize(argc, argv);
        state->run();
        return 0;
}

while the configuration file might look like this:

<ECF>
  <Algorithm>
    <SteadyStateTournament>
      <Entry key="tsize">3</Entry>
    </SteadyStateTournament>
  </Algorithm>

  <Genotype>
    <BitString>
      <Entry key="size">20</Entry>
    </BitString>
    <Tree>
      <Entry key="functionset">sin cos + - / *</Entry>
      <Entry key="terminalset">X Y</Entry>
    </Tree>
  </Genotype>

  <Registry>
    <Entry key="population.size">30</Entry>
    <Entry key="term.maxgen">100</Entry> 
  </Registry>
</ECF>

Implementation note: each Class in ECF has an associated type ClassP which is a boost smart pointer to an object of that class.
See more information on using boost.