@Test
public void testBiDistribution() {
Algorithm algorithm = mock(Algorithm.class);
MOFitness fitness01 =
Fitnesses.create(new MinimisationFitness(3.0), new MinimisationFitness(0.0));
final Type type01 = mock(Type.class, "type2_01");
MOFitness fitness02 =
Fitnesses.create(new MinimisationFitness(2.0), new MinimisationFitness(1.0));
final Type type02 = mock(Type.class, "type2_02");
// Cloning returns the mocked objects themselves.
when(type01.getClone()).thenReturn(type01);
when(type02.getClone()).thenReturn(type02);
OptimisationSolution solution01 = new OptimisationSolution(type01, fitness01);
OptimisationSolution solution02 = new OptimisationSolution(type02, fitness02);
Archive archive = Archive.Provider.get();
archive.add(solution01);
archive.add(solution02);
Real distribution = new SolutionDistribution().getValue(algorithm);
Assert.assertEquals(0.0, distribution.doubleValue(), 0.0000000000001);
}
/** Test the get column method. */
@Test
public void testGetColumn() {
TypeList expected = new TypeList();
expected.add(Real.valueOf(0.2));
expected.add(Real.valueOf(1.2));
TypeList column = stringTargetPatterns.getColumn(1);
Assert.assertEquals(expected, column);
expected = new TypeList();
Vector target = Vector.of(1, 1, 0);
expected.add(target);
target = Vector.of(0, 0, 1);
expected.add(target);
column = vectorTargetPatterns.getColumn(5);
Assert.assertEquals(expected, column);
}
private void calculateAbsoluteAverages(PSO pso) {
int dimension = pso.getTopology().last().getDimension();
absoluteAverageVelocityVector = Vector.of();
averageSpeedVector = Vector.of();
for (Entity e : pso.getTopology()) {
Vector velocity = (Vector) e.getProperties().get(EntityType.Particle.VELOCITY);
for (int i = 0; i < dimension; i++) {
if (absoluteAverageVelocityVector.size() < dimension) {
absoluteAverageVelocityVector.add(velocity.get(i));
averageSpeedVector.add(Real.valueOf(Math.abs(velocity.doubleValueOf(i))));
} else {
absoluteAverageVelocityVector.setReal(
i, absoluteAverageVelocityVector.doubleValueOf(i) + velocity.doubleValueOf(i));
averageSpeedVector.setReal(
i, averageSpeedVector.doubleValueOf(i) + Math.abs(velocity.doubleValueOf(i)));
}
}
}
for (int i = 0; i < dimension; i++) {
absoluteAverageVelocityVector.setReal(
i, Math.abs(absoluteAverageVelocityVector.doubleValueOf(i) / (double) dimension));
averageSpeedVector.setReal(i, averageSpeedVector.doubleValueOf(i) / (double) dimension);
}
}
/**
* Given the datatype key, determines the datatype for a given column. Also sets up the HashTables
* to map a nominal attribute to a corresponding integer number.
*
* @param columnn the column to map to a datatype.
* @param datatype the datatype key.
* @return a corresponding CIlib type.
* @throws CIlibIOException {@inheritDoc}
*/
private Type getTypeData(int columnn, String datatype) throws CIlibIOException {
if (datatype.equalsIgnoreCase("NUMERIC")) {
return Real.valueOf(0.0);
}
if (datatype.equalsIgnoreCase("STRING")) {
return new StringType("");
}
if (datatype.equalsIgnoreCase("DATE")) {
throw new UnsupportedOperationException("Date format currently not supported" + " in CIlib.");
}
// If none of the above, has to be a nominal attribute.
if (columnToNominalAttributesMap == null) {
columnToNominalAttributesMap = new HashMap<Integer, HashMap<String, Integer>>();
}
HashMap<String, Integer> nominalMap = new HashMap<String, Integer>();
datatype = datatype.replaceAll("[{}]", "");
String[] nominalAttributes = datatype.split("\\,");
if (nominalAttributes.length == 0) {
throw new CIlibIOException(
"Nominal attributes must be comma separated:"
+ "{<nominal-name1>, <nominal-name2>, <nominal-name3>, ...} ");
}
for (int i = 0; i < nominalAttributes.length; i++) {
String nominalAttribute = nominalAttributes[i];
nominalMap.put(nominalAttribute, i);
}
columnToNominalAttributesMap.put(columnn, nominalMap);
return Int.valueOf(0);
}
@Test
public void testSetColumn() {
TypeList newColumn = new TypeList();
List<Real> list = Arrays.asList(Real.valueOf(5.2), Real.valueOf(6.2));
for (Real real : list) {
newColumn.add(real);
}
stringTargetPatterns.setColumn(0, newColumn);
Assert.assertEquals(newColumn, stringTargetPatterns.getColumn(0));
newColumn = new TypeList();
Vector target = Vector.of(1, 1, 0);
newColumn.add(target);
target = Vector.of(0, 0, 1);
newColumn.add(target);
newColumn = vectorTargetPatterns.getColumn(5);
Assert.assertEquals(newColumn, vectorTargetPatterns.getColumn(5));
}
@Override
public List<Type> nextRow() {
List<Type> row = Lists.newArrayList();
for (int i = 0; i < this.columnCount; ++i) {
row.add(Real.valueOf(this.vector.get(this.index + i).doubleValue()));
}
this.index += this.columnCount;
return row;
}
@BeforeClass
public static void setUpBeforeClass() throws Exception {
Vector.Builder tmp = Vector.newBuilder();
set = new ArrayList<Pattern>();
for (int i = 1; i <= SIZE; i++) {
tmp.add(Real.valueOf(i));
}
set.add(new Pattern("class0", tmp.build()));
tmp = Vector.newBuilder();
for (int i = SIZE; i > 0; i--) {
tmp.add(Real.valueOf(i));
}
set.add(new Pattern("class1", tmp.build()));
set.add(new Pattern("class2", Vector.of(1.0, 1.0, 1.0)));
set.add(new Pattern("class1", Vector.of(2.0, 2.0, 2.0)));
set.add(new Pattern("class0", Vector.of(3.0, 3.0, 3.0)));
}
@Test
public void results() {
Algorithm algorithm = this.mockery.mock(Algorithm.class);
MOFitness fitness01 =
Fitnesses.create(new MinimisationFitness(0.0), new MinimisationFitness(1.0));
final Type type01 = this.mockery.mock(Type.class, "type_01");
MOFitness fitness02 =
Fitnesses.create(new MinimisationFitness(0.5), new MinimisationFitness(0.5));
final Type type02 = this.mockery.mock(Type.class, "type_02");
MOFitness fitness03 =
Fitnesses.create(new MinimisationFitness(1.0), new MinimisationFitness(0.0));
final Type type03 = this.mockery.mock(Type.class, "type_03");
// Cloning returns the mocked objects themselves.
this.mockery.checking(
new Expectations() {
{
oneOf(type01).getClone();
will(returnValue(type01));
oneOf(type02).getClone();
will(returnValue(type02));
oneOf(type03).getClone();
will(returnValue(type03));
}
});
OptimisationSolution solution01 = new OptimisationSolution(type01, fitness01);
OptimisationSolution solution02 = new OptimisationSolution(type02, fitness02);
OptimisationSolution solution03 = new OptimisationSolution(type03, fitness03);
Archive archive = Archive.Provider.get();
archive.add(solution01);
archive.add(solution02);
archive.add(solution03);
Real extent = new ParetoFrontExtent().getValue(algorithm);
Assert.assertEquals(Math.sqrt(2.0), extent.doubleValue(), 0.00001);
}
/**
* 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;
}
@Override
public <E extends Entity> List<E> crossover(List<E> parentCollection) {
Preconditions.checkArgument(
parentCollection.size() == 2, "BlendCrossoverStrategy requires 2 parents.");
// How do we handle variable sizes? Resizing the entities?
E offspring1 = (E) parentCollection.get(0).getClone();
E offspring2 = (E) parentCollection.get(1).getClone();
Vector parentChromosome1 = (Vector) parentCollection.get(0).getCandidateSolution();
Vector parentChromosome2 = (Vector) parentCollection.get(1).getCandidateSolution();
Vector.Builder offspringChromosome1 = Vector.newBuilder();
Vector.Builder offspringChromosome2 = Vector.newBuilder();
int sizeParent1 = parentChromosome1.size();
int sizeParent2 = parentChromosome2.size();
int minDimension = Math.min(sizeParent1, sizeParent2);
for (int i = 0; i < minDimension; i++) {
double gamma =
(1 + 2 * alpha.getParameter()) * random.getRandomNumber() - alpha.getParameter();
double value1 =
(1 - gamma) * parentChromosome1.doubleValueOf(i)
+ gamma * parentChromosome2.doubleValueOf(i);
double value2 =
(1 - gamma) * parentChromosome2.doubleValueOf(i)
+ gamma * parentChromosome1.doubleValueOf(i);
offspringChromosome1.add(Real.valueOf(value1, parentChromosome1.boundsOf(i)));
offspringChromosome2.add(Real.valueOf(value2, parentChromosome1.boundsOf(i)));
}
offspring1.setCandidateSolution(offspringChromosome1.build());
offspring2.setCandidateSolution(offspringChromosome2.build());
return Arrays.asList(offspring1, offspring2);
}
/**
* Puts a token into a new object of the correct CIlib type.
*
* @param index the index of the token in the row (its column).
* @param token the token to be typed.
* @return a new CIlib object of the correct type.
*/
private Type mapTokenToType(int index, String token) {
Type type = columnTypePrototypes.get(index);
if (type instanceof Real) {
return Real.valueOf(Double.parseDouble(token));
}
if (type instanceof StringType) {
return new StringType(token);
}
// If none of the above, has to be a nominal attribute.
HashMap<String, Integer> nominalMap = this.columnToNominalAttributesMap.get(index);
return Int.valueOf(nominalMap.get(token));
}
@Override
public Vector get(Particle particle) {
Vector newPos = (Vector) delegate.get(particle);
Particle tmp = particle.getClone();
tmp.setPosition(newPos);
Fitness newFitness = particle.getBehaviour().getFitnessCalculator().getFitness(tmp);
final UniformDistribution uniform = new UniformDistribution();
Vector newPBest =
newPos.plus(
Vector.newBuilder()
.repeat(newPos.size(), Real.valueOf(1.0))
.build()
.multiply(
new P1<Number>() {
@Override
public Number _1() {
return uniform.getRandomNumber(
-granularity.getParameter(), granularity.getParameter());
}
}));
tmp.setPosition(newPos);
Fitness newPBestFitness = particle.getBehaviour().getFitnessCalculator().getFitness(tmp);
if (newPBestFitness.compareTo(newFitness) < 0) {
Vector tmpVector = Vector.copyOf(newPos);
newPos = newPBest;
newPBest = tmpVector;
newPBestFitness = newFitness;
}
double dot =
((Vector) particle.getNeighbourhoodBest().getBestPosition())
.subtract(newPos)
.dot(newPBest.subtract(newPos));
if (dot < 0) {
return (Vector) particle.getPosition();
}
particle.put(Property.BEST_POSITION, newPBest);
particle.put(Property.BEST_FITNESS, newPBestFitness);
return newPos;
}
/** {@inheritDoc} */
public Real getValue(Algorithm algorithm) {
PSO pso = (PSO) algorithm;
int numberParticles = pso.getTopology().size();
Iterator<Particle> k = pso.getTopology().iterator();
Particle particle = k.next();
Vector averageParticlePosition = (Vector) particle.getPosition().getClone();
while (k.hasNext()) {
particle = k.next();
Vector v = (Vector) particle.getPosition();
for (int j = 0; j < averageParticlePosition.size(); ++j) {
averageParticlePosition.setReal(
j, averageParticlePosition.doubleValueOf(j) + v.doubleValueOf(j));
}
}
for (int j = 0; j < averageParticlePosition.size(); ++j) {
averageParticlePosition.setReal(
j, averageParticlePosition.doubleValueOf(j) / numberParticles);
}
Iterator<Particle> i = pso.getTopology().iterator();
double particleSum = 0.0;
while (i.hasNext()) {
particle = i.next();
double dimensionSum = 0.0;
Vector v = (Vector) particle.getPosition();
for (int j = 0; j < particle.getDimension(); ++j) {
dimensionSum +=
(v.doubleValueOf(j) - averageParticlePosition.doubleValueOf(j))
* (v.doubleValueOf(j) - averageParticlePosition.doubleValueOf(j));
}
particleSum += Math.sqrt(dimensionSum);
}
double diversity = particleSum / numberParticles;
DiameterVisitor diameterVisitor = new DiameterVisitor();
pso.accept(diameterVisitor);
double diameter = diameterVisitor.getResult();
return Real.valueOf(diversity / diameter);
}
@Test(expected = IllegalArgumentException.class)
public void testVectorDistance() {
DistanceMeasure distanceMeasure = new CosineDistanceMeasure();
Vector v1 = new Vector();
Vector v2 = new Vector();
v1.add(Real.valueOf(4.0));
v1.add(Real.valueOf(3.0));
v1.add(Real.valueOf(2.0));
v2.add(Real.valueOf(2.0));
v2.add(Real.valueOf(3.0));
v2.add(Real.valueOf(4.0));
double distance = distanceMeasure.distance(v1, v2);
assertTrue(distance >= -1 && distance <= 1);
assertEquals(1 - (25.0 / 29.0), distance, 0.000000000000001);
v1.add(Real.valueOf(22.0));
distanceMeasure.distance(v1, v2);
}
@Test
public void algorithmExecution() {
NNDataTrainingProblem problem = new NNDataTrainingProblem();
problem.getDataTableBuilder().setDataReader(new ARFFFileReader());
problem.getDataTableBuilder().setSourceURL("library/src/test/resources/datasets/iris.arff");
problem.setTrainingSetPercentage(0.7);
problem.setGeneralizationSetPercentage(0.3);
problem
.getNeuralNetwork()
.getArchitecture()
.setArchitectureBuilder(new CascadeArchitectureBuilder());
problem.getNeuralNetwork().setOperationVisitor(new CascadeVisitor());
problem
.getNeuralNetwork()
.getArchitecture()
.getArchitectureBuilder()
.addLayer(new LayerConfiguration(4));
problem
.getNeuralNetwork()
.getArchitecture()
.getArchitectureBuilder()
.addLayer(new LayerConfiguration(0));
problem
.getNeuralNetwork()
.getArchitecture()
.getArchitectureBuilder()
.addLayer(new LayerConfiguration(1));
problem
.getNeuralNetwork()
.getArchitecture()
.getArchitectureBuilder()
.getLayerBuilder()
.setDomain("R(-3:3)");
problem.initialise();
PSO pso = new PSO();
pso.getInitialisationStrategy().setEntityType(new DynamicParticle());
pso.addStoppingCondition(new MeasuredStoppingCondition());
pso.setOptimisationProblem(problem);
pso.performInitialisation();
CascadeNetworkExpansionReactionStrategy reaction =
new CascadeNetworkExpansionReactionStrategy();
Assert.assertEquals(5, ((Vector) pso.getBestSolution().getPosition()).size());
Assert.assertEquals(5, problem.getNeuralNetwork().getWeights().size());
for (int i = 0; i < Topologies.getBestEntity(pso.getTopology()).getDimension(); ++i) {
((Vector) Topologies.getBestEntity(pso.getTopology()).getPosition())
.set(i, Real.valueOf(0.0));
((Vector) Topologies.getBestEntity(pso.getTopology()).getVelocity())
.set(i, Real.valueOf(0.0));
((Vector) Topologies.getBestEntity(pso.getTopology()).getBestPosition())
.set(i, Real.valueOf(0.0));
}
reaction.performReaction(pso);
Assert.assertEquals(11, ((Vector) pso.getBestSolution().getPosition()).size());
Assert.assertEquals(11, problem.getNeuralNetwork().getWeights().size());
Assert.assertEquals(
Vector.of(
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getPosition());
Assert.assertEquals(
Vector.of(
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getVelocity());
Assert.assertEquals(
Vector.of(
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getBestPosition());
for (int i = 0; i < Topologies.getBestEntity(pso.getTopology()).getDimension(); ++i) {
((Vector) Topologies.getBestEntity(pso.getTopology()).getPosition())
.set(i, Real.valueOf(0.0));
((Vector) Topologies.getBestEntity(pso.getTopology()).getVelocity())
.set(i, Real.valueOf(0.0));
((Vector) Topologies.getBestEntity(pso.getTopology()).getBestPosition())
.set(i, Real.valueOf(0.0));
}
reaction.performReaction(pso);
Assert.assertEquals(18, ((Vector) pso.getBestSolution().getPosition()).size());
Assert.assertEquals(18, problem.getNeuralNetwork().getWeights().size());
Assert.assertEquals(
Vector.of(
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getPosition());
Assert.assertEquals(
Vector.of(
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getVelocity());
Assert.assertEquals(
Vector.of(
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getBestPosition());
for (int i = 0; i < Topologies.getBestEntity(pso.getTopology()).getDimension(); ++i) {
((Vector) Topologies.getBestEntity(pso.getTopology()).getPosition())
.set(i, Real.valueOf(0.0));
((Vector) Topologies.getBestEntity(pso.getTopology()).getVelocity())
.set(i, Real.valueOf(0.0));
((Vector) Topologies.getBestEntity(pso.getTopology()).getBestPosition())
.set(i, Real.valueOf(0.0));
}
reaction.performReaction(pso);
Assert.assertEquals(26, ((Vector) pso.getBestSolution().getPosition()).size());
Assert.assertEquals(26, problem.getNeuralNetwork().getWeights().size());
Assert.assertEquals(
Vector.of(
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getPosition());
Assert.assertEquals(
Vector.of(
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getVelocity());
Assert.assertEquals(
Vector.of(
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
Double.NaN,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
Double.NaN),
(Vector) Topologies.getBestEntity(pso.getTopology()).getBestPosition());
}
private void updateInertia(PSO pso) {
int dimension = pso.getTopology().last().getDimension();
if (inertiaWeight.size() < dimension) {
Vector.Builder builder = Vector.newBuilder();
builder.repeat(dimension, Real.valueOf(initialInertiaWeight.getParameter()));
inertiaWeight = builder.build();
}
Vector.Builder builder = Vector.newBuilder();
for (int i = 0; i < dimension; i++) {
builder.add(
Math.sqrt(
Math.pow(absoluteAverageVelocityVector.doubleValueOf(i), 2)
+ Math.pow(averageSpeedVector.doubleValueOf(i), 2)));
}
Vector d = builder.build(); // get the degree of convergence vector
double max_d = 0;
for (Numeric component : d) {
if (component.doubleValue() > max_d) {
max_d = component.doubleValue();
}
}
if (max_d != 0) {
Vector.Builder builder2 = Vector.newBuilder();
for (Numeric component : d) {
builder2.add(max_d / (max_d + component.doubleValue()));
}
Vector w = builder2.build();
/*double sum_w = 0;
for(Numeric component : w) {
sum_w += component.doubleValue();
}
/*
Vector.Builder builder3 = Vector.newBuilder();
for(Numeric component : w) {
builder3.add(Math.pow(dimension * component.doubleValue() / sum_w, pwr.getParameter()));
} */
/*
for(Numeric component : w) {
//builder3.add(component.doubleValue() - w_mean / w_stdDiv);
builder3.add(component.doubleValue() * initialInertiaWeight.getParameter());
}
for(int i = 0; i < inertiaWeight.size(); i++) {
builder3.add(w.doubleValueOf(i) * inertiaWeight.doubleValueOf(i));
}
*/
/*
Vector m = builder3.build();
double sum_m = 0;
for (Numeric num : m) {
sum_m += num.doubleValue();
}
double m_mean = sum_m / (double) dimension;
double sum_diff_squared = 0;
for(Numeric component : m) {
sum_diff_squared += Math.pow(component.doubleValue() - m_mean, 2);
}
double m_stdDiv = Math.sqrt(sum_diff_squared / (double) dimension);
*/
// System.out.println("VEL: StdDiv of M: " + m_stdDiv + ", mean of M: " + m_mean);
for (int i = 0; i < inertiaWeight.size(); i++) {
inertiaWeight.setReal(
i,
(1 - filter.getParameter()) * w.doubleValueOf(i)
+ filter.getParameter()
* inertiaWeight.doubleValueOf(i)); // w.doubleValueOf(i));//;
}
}
}
/**
* Calculates the AUC of one output.
*
* @param targets The expected outputs. Each value is discetised to the closest bound.
* @param outputs The outputs of the NN.
* @return The AUC in the range [0,1].
*/
public double areaUnderCurve(ArrayList<Real> targets, ArrayList<Real> outputs) {
double negatives = 0.0;
double positives = 0.0;
double centerBound = (upperBound + lowerBound) / 2.0;
// determine total positives and negatives
for (Real curTarget : targets) {
if (curTarget.doubleValue() > centerBound) positives += 1;
else negatives += 1;
}
// plot ROC curve coordinates
double threshold = lowerBound;
ArrayList<Real> tpCoords = new ArrayList<Real>();
ArrayList<Real> fpCoords = new ArrayList<Real>();
// when all outputs are seen as true
tpCoords.add(Real.valueOf(1.0));
fpCoords.add(Real.valueOf(1.0));
double newThresholdL;
double newThresholdU;
do {
// calculate next threshold
newThresholdL = upperBound;
newThresholdU = upperBound;
for (Real curOutput : outputs) {
if (curOutput.doubleValue() >= threshold && curOutput.doubleValue() < newThresholdL) {
newThresholdL = curOutput.doubleValue();
}
}
for (Real curOutput : outputs) {
if (curOutput.doubleValue() > newThresholdL && curOutput.doubleValue() < newThresholdU) {
newThresholdU = curOutput.doubleValue();
}
}
threshold = (newThresholdL + newThresholdU) / 2.0;
// calculate tp and fp for this threshold
double tPositive = 0;
double fPositive = 0;
for (int curOutput = 0; curOutput < outputs.size(); ++curOutput) {
if (targets.get(curOutput).doubleValue() > centerBound
&& outputs.get(curOutput).doubleValue() > threshold) {
tPositive += 1;
} else if (targets.get(curOutput).doubleValue() <= centerBound
&& outputs.get(curOutput).doubleValue() > threshold) {
fPositive += 1;
}
}
tpCoords.add(Real.valueOf(tPositive / positives));
fpCoords.add(Real.valueOf(fPositive / negatives));
} while (newThresholdU < upperBound);
// when all outputs are seen as false
tpCoords.add(Real.valueOf(0.0));
fpCoords.add(Real.valueOf(0.0));
// calculate area
double area = 0.0;
for (int curCoord = 1; curCoord < fpCoords.size(); ++curCoord) {
area +=
((tpCoords.get(curCoord - 1).doubleValue() + tpCoords.get(curCoord).doubleValue()) / 2)
* (fpCoords.get(curCoord - 1).doubleValue()
- fpCoords.get(curCoord).doubleValue()); // fpCoords sorted in decending order
}
return area;
}
