sonar_tutorial_part4 - morinim/ultra GitHub Wiki

Choosing a different evaluator

Ultra comes with different pre-packaged evaluators: for classification tasks the default one is based on Gaussian distributions.

Usually, it's a good and versatile evaluator since it can handle multiclass classification problems.

However the sonar problem is a binary classification problem and could benefit from a specific evaluator: the binary_evaluator.

Now, consider another variant of the original source code:

src::basic_search<alps_es, src::binary_evaluator<gp::individual>> s(prob);
s.validation_strategy<src::holdout_validation>(prob);

const auto result(s.run(5));

std::cout << "\nCANDIDATE SOLUTION\n"
          << out::c_language << result.best_individual
          << "\n\nACCURACY\n" << *result.best_measurements.accuracy * 100.0
          << '%'
          << "\n\nFITNESS\n" << *result.best_measurements.fitness << '\n';

(full example in examples/sonar05.cc)

Here we use the src::basic_search class (instead of the usual src::search) because we want to specify the evaluator.

src::basic_search also allows specifying the evolution strategy. In this case, we use alps_es, which is the same strategy used by src::search and generally shouldn't be altered.