/**
* Calculates the AUC measurment.
*
* @param algorithm The optimisation algorithm with a NNTrainingProblem.
* @return A Vector with the AUC for each NN output.
*/
@Override
public Vector getValue(Algorithm algorithm) {
Vector solution = (Vector) algorithm.getBestSolution().getPosition();
NNTrainingProblem problem = (NNTrainingProblem) algorithm.getOptimisationProblem();
StandardPatternDataTable generalisationSet = problem.getGeneralisationSet();
NeuralNetwork neuralNetwork = problem.getNeuralNetwork();
neuralNetwork.setWeights(solution);
// Arrange outputs and target values into ArrayLists.
ArrayList<ArrayList<Real>> targets = new ArrayList<ArrayList<Real>>();
ArrayList<ArrayList<Real>> outputs = new ArrayList<ArrayList<Real>>();
// case of multiple outputs
if (generalisationSet.getRow(0).getTarget() instanceof Vector) {
int size = ((Vector) generalisationSet.getRow(0).getTarget()).size();
for (int i = 0; i < size; ++i) {
targets.add(new ArrayList<Real>());
outputs.add(new ArrayList<Real>());
}
for (StandardPattern pattern : generalisationSet) {
Vector target = (Vector) pattern.getTarget();
Vector output = neuralNetwork.evaluatePattern(pattern);
for (int curOutput = 0; curOutput < target.size(); ++curOutput) {
targets.get(curOutput).add((Real) target.get(curOutput));
outputs.get(curOutput).add((Real) output.get(curOutput));
}
}
}
// case of single output
else {
targets.add(new ArrayList<Real>());
outputs.add(new ArrayList<Real>());
for (StandardPattern pattern : generalisationSet) {
Real target = (Real) pattern.getTarget();
Vector output = neuralNetwork.evaluatePattern(pattern);
targets.get(0).add(target);
outputs.get(0).add((Real) output.get(0));
}
}
// Calculate the Vector of AUC values
Vector results = Vector.of();
for (int curOutput = 0; curOutput < outputs.size(); ++curOutput) {
results.add(Real.valueOf(areaUnderCurve(targets.get(curOutput), outputs.get(curOutput))));
}
return results;
}
/**
* Reinitialises all the output weights of a cascade network within a PSO.
*
* <p>{@inheritDoc}
*/
@Override
public void performReaction(E algorithm) {
NNDataTrainingProblem problem = (NNDataTrainingProblem) algorithm.getOptimisationProblem();
NeuralNetwork network = problem.getNeuralNetwork();
int precedingLayersSize =
network.getArchitecture().getArchitectureBuilder().getLayerConfigurations().get(0).getSize()
+ network
.getArchitecture()
.getArchitectureBuilder()
.getLayerConfigurations()
.get(1)
.getSize();
if (network.getArchitecture().getArchitectureBuilder().getLayerConfigurations().get(0).isBias())
precedingLayersSize++;
int outputLayerSize =
network
.getArchitecture()
.getArchitectureBuilder()
.getLayerConfigurations()
.get(2)
.getSize();
int nrOfweightsToDo = precedingLayersSize * outputLayerSize;
Topology<? extends Entity> entities = algorithm.getTopology();
for (Entity entity : entities) {
DynamicParticle particle = (DynamicParticle) entity;
Vector position = (Vector) particle.getPosition();
Vector velocity = (Vector) particle.getVelocity();
for (int curElement = position.size() - nrOfweightsToDo;
curElement < position.size();
++curElement) {
((Real) position.get(curElement)).randomize(randomGenerator);
velocity.setReal(curElement, 0.0);
}
}
}
@Test
public void testBuildArchitecture() {
NeuralNetwork network = new NeuralNetwork();
network.getArchitecture().getArchitectureBuilder().addLayer(new LayerConfiguration(5));
network.getArchitecture().getArchitectureBuilder().addLayer(new LayerConfiguration(3));
network.getArchitecture().getArchitectureBuilder().addLayer(new LayerConfiguration(3, false));
network.getArchitecture().getArchitectureBuilder().addLayer(new LayerConfiguration(2));
network.getArchitecture().getArchitectureBuilder().getLayerBuilder().setDomain("R(-3,3)");
network.initialize();
// assert num layers
Assert.assertEquals(4, network.getArchitecture().getNumLayers());
// assert fully connected
Assert.assertEquals(36, network.getWeights().size());
// assert biasses
Assert.assertEquals(true, network.getArchitecture().getLayers().get(0).isBias());
int layerSize = network.getArchitecture().getLayers().get(0).size();
Assert.assertEquals(6, layerSize);
Assert.assertEquals(
-1,
network.getArchitecture().getLayers().get(0).getNeuralInput(layerSize - 1),
Maths.EPSILON);
Assert.assertEquals(true, network.getArchitecture().getLayers().get(1).isBias());
layerSize = network.getArchitecture().getLayers().get(1).size();
Assert.assertEquals(4, layerSize);
Assert.assertEquals(
-1,
network.getArchitecture().getLayers().get(1).getNeuralInput(layerSize - 1),
Maths.EPSILON);
Assert.assertEquals(false, network.getArchitecture().getLayers().get(2).isBias());
layerSize = network.getArchitecture().getLayers().get(2).size();
Assert.assertEquals(3, layerSize);
Assert.assertEquals(false, network.getArchitecture().getLayers().get(3).isBias());
layerSize = network.getArchitecture().getLayers().get(3).size();
Assert.assertEquals(2, layerSize);
}