public void updateModelWeight(List<Sample> sampleList) {
for (FeatureValue featureValue : featureMap.values()) {
featureValue.setTempWeight(0);
}
for (Sample sample : sampleList) {
Map<String, Double> probMap = predict(sample.getFeatures());
for (String feature : sample.getFeatures()) {
for (String label : labelSet) {
this.setTempWeight(
label,
feature,
this.getTempWeight(label, feature)
+ probMap.get(label) * (1.0 / sampleList.size()));
}
}
}
// GIS;
features.setLastWeight();
for (FeatureValue featureValue : featureMap.values()) {
featureValue.setModelWeight(
featureValue.getModelWeight()
+ 1.0
/ maxFeaturePerSample
* Math.log(featureValue.getEmpiricalWeight() / featureValue.getTempWeight()));
}
}
// calculate p(y|x)=sum_i(alpha_i*f_i(x,y))/sum_y(sum_i(alpha_i*f_i(x,y))).
public Map<String, Double> predict(List<String> x) {
Map<String, Double> probMap = new HashMap<String, Double>();
double sum = 0;
for (String label : labelSet) {
double prob = 0;
for (String feature : x) {
prob += features.getModelWeight(label, feature);
}
prob = Math.exp(prob);
probMap.put(label, prob);
sum += prob;
}
for (String label : probMap.keySet()) {
probMap.put(label, probMap.get(label) / sum);
}
return probMap;
}
public boolean isConvergence(double threshold) {
double sum = 0;
for (FeatureValue featureValue : featureMap.values()) {
sum +=
(featureValue.getLastWeight() - featureValue.getModelWeight())
* (featureValue.getLastWeight() - featureValue.getModelWeight());
}
return Math.sqrt(sum) <= threshold ? true : false;
}
public void addFeature(String label, String feature) {
if (!featureMap.containsKey(new FeatureKey(label, feature))) {
featureMap.put(new FeatureKey(label, feature), new FeatureValue());
} else {
this.updateCount(label, feature);
}
if (!featureCount.containsKey(feature)) {
featureCount.put(feature, 1);
} else {
featureCount.put(feature, featureCount.get(feature) + 1);
}
}
public Map generateMaps() {
boolean generationSuccessful = false;
while (generationSuccessful == false) {
currentGeneration = new ArrayList<Map>();
currentLowestFitness = null;
secondBestFitness = null;
for (int i = 0; i < 8; i++) {
Map currMap = new Map();
currMap.generateMap();
currentGeneration.add(currMap);
}
setFitnesses();
System.out.println(currentLowestFitness.getFitness());
int xyz = 0;
long time = System.currentTimeMillis();
while (currentLowestFitness.getFitness() > 10 && xyz < 10000) {
// System.out.println(currentGeneration.size());
// größe größer 2
int iterator = 0;
int crossDivider = new Random().nextInt(currentLowestFitness.getCellList().size());
Map childMap = new Map();
ArrayList<Cell> childCellList = new ArrayList<Cell>();
Random random = new Random();
for (int i = 0; i < crossDivider; i++) {
if (random.nextInt(100) >= 2) {
childCellList.add(currentLowestFitness.getCellList().get(i));
} else {
childCellList.add(
new Cell(
new Point(
random.nextInt(MAXIMUM_X_POSITION), random.nextInt(MAXIMUM_Y_POSITION)),
random.nextInt(Map.CELL_MAXIMUM_SIZE - 4) + 4));
}
}
for (int i = crossDivider; i < secondBestFitness.getCellList().size(); i++) {
if (random.nextInt(100) >= 2) {
childCellList.add(secondBestFitness.getCellList().get(i));
} else {
childCellList.add(
new Cell(
new Point(
random.nextInt(MAXIMUM_X_POSITION), random.nextInt(MAXIMUM_Y_POSITION)),
random.nextInt(Map.CELL_MAXIMUM_SIZE - 4) + 4));
}
}
childMap.setCellList(childCellList);
// System.out.printf("Elems: %s\n",
// currentGeneration.keySet().stream().map(Ströing::valueOf).reduce((a,
// b) -> String.format("%s %s", a, b)).get());
// System.out.printf("Fitness: %f\n", childMap.getFitness());
currentGeneration.remove(secondBestFitness);
currentGeneration.add(childMap);
setFitnesses();
// System.out.println(currentLowestFitness.getFitness());
xyz++;
}
System.out.println(
"Number of Generations: " + xyz + " Time needed: " + (System.currentTimeMillis() - time));
if (xyz < 10000) {
generationSuccessful = true;
}
}
return currentLowestFitness;
}
private void setFitnesses() {
for (Map map : currentGeneration) {
if (currentLowestFitness == null) {
currentLowestFitness = map;
} else if (secondBestFitness == null
&& map.getFitness() > currentLowestFitness.getFitness()) {
secondBestFitness = map;
} else if (secondBestFitness == null
&& map.getFitness() <= currentLowestFitness.getFitness()) {
secondBestFitness = currentLowestFitness;
currentLowestFitness = map;
} else if (map.getFitness() < currentLowestFitness.getFitness()) {
secondBestFitness = currentLowestFitness;
currentLowestFitness = map;
} else if (map.getFitness() > currentLowestFitness.getFitness()
&& map.getFitness() < secondBestFitness.getFitness()) {
secondBestFitness = map;
}
}
}
public void setEmpiricalWeight(int sampleNumber) {
for (FeatureValue featureValue : featureMap.values()) {
featureValue.setEmpiricalWeight(1.0 * featureValue.getCount() / sampleNumber);
}
}
public double getModelWeight(String label, String feature) {
FeatureKey featureKey = new FeatureKey(label, feature);
return featureMap.containsKey(featureKey)
? this.featureMap.get(featureKey).getModelWeight()
: 0;
}
public void setTempWeight(String label, String feature, double weight) {
FeatureKey featureKey = new FeatureKey(label, feature);
if (featureMap.containsKey(featureKey)) {
this.featureMap.get(featureKey).setTempWeight(weight);
}
}
public void setLastWeight() {
for (FeatureValue featureValue : featureMap.values()) {
featureValue.setLastWeight();
}
}