The model selection class

Model Selection class

The model selection is applied to find a neural network with a topology that minimizes the error for new data. Two ways to obtain an optimal topology are neuron selection and input selection.

Neuron selection algorithms aim to get the optimal number of neurons in the neural network’s hidden perceptron layer, while input selection algorithms are responsible for finding the optimal subset of inputs.

As we did in previous sections, we will continue working with the iris data set, which can be downloaded from the DataSet class chapter. Before continuing, reading the previous chapter on the TrainingStrategy class is also advisable.

To construct a model selection object associated with a training strategy object, we write:

ModelSelection model_selection(&training_strategy);

The default model selection consists of a growing inputs or incremental neurons selection algorithm. The following sentence allows us to change this:

model_selection.set_inputs_selection_method(ModelSelection::InputsSelectionMethod::GENETIC_ALGORITHM);
model_selection.set_neurons_selection_method(ModelSelection::NeuronsSelectionMethod::GROWING_NEURONS);
GeneticAlgorithm* genetic_algorithm = model_selection.get_genetic_algorithm();
genetic_algorithm->set_individuals_number(100);
GrowingNeurons* growing_neurons = model_selection.get_growing_neurons();
growing_neurons ->set_maximum_selection_failures(3);

Model selection’s main methods are the ones that perform the input and neuron selection. To call them, we use:

model_selection.perform_input_selection();
model_selection.perform_neurons_selection();

We can save the above object to an XML file.

model_selection.save("model_selection.xml");

For more information on the ModelSelection class visit the ModelSelection Class Reference.