private DynamicIntArray initHistogram(Set<? extends Attributable> elements, String key) {
DynamicIntArray array = new DynamicIntArray(12, 0);
for (Attributable element : elements) {
String strVal = element.getAttribute(key);
if (strVal != null) {
double value = Double.parseDouble(strVal);
int bucket = discretizer.index(value);
array.set(bucket, array.get(bucket) + 1);
}
}
return array;
}
private void calculateBuckets(
Set<? extends Person> persons,
DynamicDoubleArray sums,
DynamicIntArray counts,
String xAttrKey,
String yAttrKey) {
TIntDoubleHashMap sumBuckets = new TIntDoubleHashMap();
TIntIntHashMap countBuckets = new TIntIntHashMap();
for (Person person : persons) {
String xValStr = person.getAttribute(xAttrKey);
String yValStr = person.getAttribute(yAttrKey);
if (xValStr != null && yValStr != null) {
double xVal = Double.parseDouble(xValStr);
double yVal = Double.parseDouble(yValStr);
int bucketIdx = xDataDiscr.index(xVal);
sumBuckets.adjustOrPutValue(bucketIdx, yVal, yVal);
countBuckets.adjustOrPutValue(bucketIdx, 1, 1);
}
}
TIntDoubleIterator it = sumBuckets.iterator();
for (int i = 0; i < sumBuckets.size(); i++) {
it.advance();
int bucketIndex = it.key();
double sum = it.value();
int cnt = countBuckets.get(bucketIndex);
sums.set(bucketIndex, sum);
counts.set(bucketIndex, cnt);
}
}
private void initReferenceValues(
Set<? extends Person> persons, String xAttrKey, String yAttrKey) {
referenceValues = new DynamicDoubleArray(100, Double.NaN);
DynamicDoubleArray sums = new DynamicDoubleArray(100, Double.NaN);
DynamicIntArray counts = new DynamicIntArray(100, -1);
calculateBuckets(persons, sums, counts, xAttrKey, yAttrKey);
for (int i = 0; i < sums.size(); i++) {
double sum = sums.get(i);
int cnt = counts.get(i);
if (!Double.isNaN(sum) && cnt > 0) {
referenceValues.set(i, sum / (double) cnt);
}
}
}
private double changeBucketContent(int bucketIndex, double value, boolean add) {
double oldDiff = calculateDiff(bucketIndex);
double sum = bucketSums.get(bucketIndex);
int count = bucketCounts.get(bucketIndex);
if (add) {
bucketSums.set(bucketIndex, sum + value);
bucketCounts.set(bucketIndex, count + 1);
} else {
bucketSums.set(bucketIndex, sum - value);
bucketCounts.set(bucketIndex, count - 1);
}
double newDiff = calculateDiff(bucketIndex);
return newDiff - oldDiff;
}
public UnivariatFrequency(
Set<? extends Attributable> refElements,
Set<? extends Attributable> simElements,
String attrKey,
Discretizer discretizer) {
this.discretizer = discretizer;
this.attrKey = attrKey;
refFreq = initHistogram(refElements, attrKey);
simFreq = initHistogram(simElements, attrKey);
scaleFactor = simElements.size() / (double) refElements.size();
normFactor = 1; // simElements.size();
int size = Math.max(simFreq.size(), refFreq.size());
for (int i = 0; i < size; i++) {
double simVal = simFreq.get(i);
double refVal = refFreq.get(i) * scaleFactor;
// hamiltonianValue += (Math.abs(simVal - refVal)/ refVal)/normFactor;
hamiltonianValue += calculateError(simVal, refVal) / normFactor;
}
}
private double calculateDiff(int bucketIndex) {
double refValue = referenceValues.get(bucketIndex);
if (Double.isNaN(refValue)) {
// There is no reference value for this bucket. We cannot do any comparison.
return 0.0;
} else {
double sum = bucketSums.get(bucketIndex);
int cnt = bucketCounts.get(bucketIndex);
if (!Double.isNaN(sum) && cnt > 0) {
double simValue = sum / (double) cnt;
return Math.abs(refValue - simValue);
} else {
return 0.0;
}
}
}
public MultivariatMean(
Set<? extends Person> refPersons,
Set<CachedPerson> simPersons,
String xAttrKey,
String yAttrKey,
Discretizer xDataDiscr) {
this.xDataKey = Converters.getObjectKey(xAttrKey);
this.yDataKey = Converters.getObjectKey(yAttrKey);
this.xDataDiscr = xDataDiscr;
initReferenceValues(refPersons, xAttrKey, yAttrKey);
initSimulationValues(simPersons, xAttrKey, yAttrKey);
// Calculate the initial hamiltonian value.
hamiltonianValue = 0;
int size = Math.max(referenceValues.size(), bucketCounts.size());
for (int i = 0; i < size; i++) {
hamiltonianValue += calculateDiff(i);
}
}
private double changeBucketContent(int bucketIndex, int value) {
double simVal = simFreq.get(bucketIndex);
double refVal = refFreq.get(bucketIndex) * scaleFactor;
// double oldDiff = Math.abs(simVal - refVal)/refVal;
double oldDiff = calculateError(simVal, refVal);
simFreq.set(bucketIndex, simFreq.get(bucketIndex) + value);
simVal = simFreq.get(bucketIndex);
refVal = refFreq.get(bucketIndex) * scaleFactor;
// double newDiff = Math.abs(simVal - refVal)/refVal;
double newDiff = calculateError(simVal, refVal);
return newDiff - oldDiff;
}