/**
* Returns KC distance between event sets es1 and es2
*
* @param es1 The first EventSet
* @param es2 The second EventSet
* @return the KC distance between them
*/
@Override
public double distance(EventSet es1, EventSet es2) {
EventHistogram h1 = es1.getHistogram();
EventHistogram h2 = es2.getHistogram();
Set<Event> s = new HashSet<Event>();
List<Pair<Event, Double>> l1 = new ArrayList<Pair<Event, Double>>();
List<Pair<Event, Double>> l2 = new ArrayList<Pair<Event, Double>>();
HashMap<Event, Integer> hm1 = new HashMap<Event, Integer>();
HashMap<Event, Integer> hm2 = new HashMap<Event, Integer>();
double oldfreq = Double.POSITIVE_INFINITY;
double correlation = 0.0;
s.addAll(es1.uniqueEvents());
s.addAll(es2.uniqueEvents());
// System.out.println(h1.toString());
// System.out.println(h2.toString());
/* make lists of the histograms */
for (Event e : h1) {
l1.add(new Pair<Event, Double>(e, h1.getRelativeFrequency(e), 2));
}
for (Event e : h2) {
l2.add(new Pair<Event, Double>(e, h2.getRelativeFrequency(e), 2));
}
/* sort the list so the most frequent items are at the top */
/* NOTE : THIS MAY BE USEFUL ELSEWHERE : SAVE THIS CODE */
Collections.sort(l1);
Collections.reverse(l1);
Collections.sort(l2);
Collections.reverse(l2);
/* DEBUGGING STUFF
for (Pair <Event,Double> p : l1) {
System.out.println("L1: " + p.toString());
}
for (Pair <Event,Double> p : l1) {
System.out.println("L2: " + p.toString());
}
*/
/* Convert lists into a hashmap of event:rank pairs */
int rank = 0;
int count = 0;
for (Pair<Event, Double> p : l1) {
Event e = (Event) (p.getFirst());
double f = (Double) (p.getSecond());
count++;
if (f != oldfreq) {
rank = count;
oldfreq = f;
}
hm1.put(e, rank);
}
/* reset and do second list */
rank = 0;
count = 0;
for (Pair<Event, Double> p : l2) {
Event e = (Event) (p.getFirst());
double f = (Double) (p.getSecond());
count++;
if (f != oldfreq) {
rank = count;
oldfreq = f;
}
hm2.put(e, rank);
}
/* More debugging stuff
System.out.println(hm1.toString());
System.out.println(hm2.toString());
System.out.println(s.toString());
*/
Integer x1, x2, y1, y2;
Set<Event> s2 = new HashSet<Event>(s);
for (Event e1 : s) {
// s2.remove(e1);
for (Event e2 : s2) {
if (e1.equals(e2)) continue;
/* get ranks of events e1 and e2 in both x and y distributions */
x1 = hm1.get(e1);
/* if not present, rank is size + 1 */
if (x1 == null) x1 = hm1.size() + 1;
x2 = hm2.get(e1);
if (x2 == null) x2 = hm2.size() + 1;
y1 = hm1.get(e2);
/* if not present, rank is size + 1 */
// broke because if (y1 == null) x1 = hm1.size()+1; x1 should be y1
if (y1 == null) y1 = hm1.size() + 1;
y2 = hm2.get(e2);
if (y2 == null) y2 = hm2.size() + 1;
/* more debugging stuff
System.out.println(e1.toString() + " is ("+x1+","+x2+")");
System.out.println(e2.toString() + " is ("+y1+","+y2+")");
System.out.println(sgn(x1.compareTo(y1)) + " " +
sgn(x2.compareTo(y2)) );
System.out.println("");
*/
correlation += (sgn(x1.compareTo(y1)) * sgn(x2.compareTo(y2)));
// System.out.println(correlation);
}
}
// System.out.println(correlation);
correlation /= (hm1.size() * (hm2.size() - 1));
// System.out.println(correlation);
// System.out.println("---");
return 1.0 - correlation;
}
/**
* Converts the extracted document information into a JStylo DataMap
*
* @param features
* @param relevantEvents
* @param cumulativeFeatureDriver
* @param documentData
* @return
*/
public ConcurrentHashMap<Integer, FeatureData> createDocMap(
List<String> features,
List<EventSet> relevantEvents,
CumulativeFeatureDriver cumulativeFeatureDriver,
List<EventSet> documentData) {
// generate training instances
ConcurrentHashMap<Integer, FeatureData> documentMap =
new ConcurrentHashMap<Integer, FeatureData>();
// remove metadata event
EventSet metadata = documentData.remove(documentData.size() - 1);
// go through all eventSets in the document
for (EventSet es : documentData) {
// initialize relevant information
ArrayList<Integer> indices = new ArrayList<Integer>();
ArrayList<Event> events = new ArrayList<Event>();
EventHistogram currHistogram = new EventHistogram();
// whether or not we actually need this eventSet
boolean eventSetIsRelevant = false;
// find out if it is a histogram or not
if (cumulativeFeatureDriver.featureDriverAt(documentData.indexOf(es)).isCalcHist()) {
// find the event set in the list of relevant events
for (EventSet res : relevantEvents) {
if (es.getEventSetID().equals(res.getEventSetID())) {
eventSetIsRelevant = true;
break;
}
}
// if it is relevant
if (eventSetIsRelevant) {
// find the indices of the events
// and count all of the events
for (Event e : es) {
int currIndex = 0;
boolean hasInner = false;
// for the events n the set
for (EventSet res : relevantEvents) {
boolean found = false;
for (Event re : res) {
hasInner = true;
// if they are the same event
if (e.getEvent().equals(re.getEvent())) {
boolean inList = false;
for (Event el : events) {
if (el.getEvent().equals(e.getEvent())) {
inList = true;
break;
}
}
if (!inList) {
indices.add(currIndex);
events.add(e);
}
// Old location revert if change breaks
currHistogram.add(e);
found = true;
}
if (found) {
break;
}
currIndex++;
}
if (found) {
break;
}
// if there's no inner, it was a non-hist feature.
// increment by one
if (!hasInner) {
currIndex++;
}
}
}
// calculate/add the histograms
int index = 0;
for (Integer i : indices) {
documentMap.put(
i,
new FeatureData(
cumulativeFeatureDriver.featureDriverAt(documentData.indexOf(es)).getName(),
cumulativeFeatureDriver
.featureDriverAt(documentData.indexOf(es))
.getNormBaseline()
.getTitle(),
currHistogram.getAbsoluteFrequency(events.get(index))));
index++;
}
}
} else { // non histogram feature
// initialize the index
int nonHistIndex = 0;
// find the indices of the events
// and count all of the events
for (EventSet res : relevantEvents) {
if (es.getEventSetID().equals(res.getEventSetID())) {
break;
}
// count to find the index
boolean hasInner = false;
for (@SuppressWarnings("unused") Event re : res) {
hasInner = true;
nonHistIndex++;
}
// if ther's no inner feature, increment by one; we just passed a non-histogram
if (!hasInner) nonHistIndex++;
}
// Extract and add the event
String eventString = es.eventAt(0).getEvent();
int startIndex = eventString.indexOf("{");
int endIndex = eventString.indexOf("}");
eventString = eventString.substring(startIndex + 1, endIndex);
FeatureData fd =
new FeatureData(
cumulativeFeatureDriver.featureDriverAt(documentData.indexOf(es)).getName(),
cumulativeFeatureDriver
.featureDriverAt(documentData.indexOf(es))
.getNormBaseline()
.getTitle(),
Math.round((float) Double.parseDouble(eventString)));
documentMap.put(nonHistIndex, fd);
}
}
// add metadata back. Not sure if necessary
documentData.add(metadata);
return documentMap;
}
