/**
* This method saves a sequential pattern to the output file or in memory, depending on if the
* user provided an output file path or not when he launched the algorithm
*
* @param prefix the pattern to be saved.
* @throws IOException exception if error while writing the output file.
*/
private void savePattern(SequentialPattern prefix) throws IOException {
// increase the number of pattern found for statistics purposes
patternCount++;
// if the result should be saved to a file
if (writer != null) {
StringBuffer r = new StringBuffer("");
for (Itemset itemset : prefix.getItemsets()) {
// r.append('(');
for (String item : itemset.getItems()) {
String string = item.toString();
r.append(string);
r.append(' ');
}
r.append("-1 ");
}
//
// // print the list of Pattern IDs that contains this pattern.
// if(prefix.getSequencesID() != null){
// r.append("SID: ");
// for(Integer id : prefix.getSequencesID()){
// r.append(id);
// r.append(' ');
// }
// }
r.append(" #SUP: ");
r.append(prefix.getSequencesID().size());
writer.write(r.toString());
writer.newLine();
} // otherwise the result is kept into memory
else {
patterns.addSequence(prefix, prefix.size());
}
}
/**
* This is the "backscan-pruning" strategy described in the BIDE+ paper to avoid extending some
* prefixs that are guaranteed to not generate a closed pattern (see the BIDE+ paper for details).
*
* @param prefix the current prefix
* @param projectedContext the projected database
* @return boolean true if we should not extend the prefix
*/
private boolean checkBackScanPruning(
SequentialPattern prefix, List<PseudoSequenceBIDE> projectedContext) {
// DEBUGGIN
if (prefix.size() == 1 && prefix.get(0).get(0) == 5) {
System.out.println("PREFIX 5 ");
}
//
// See the BIDE+ paper for details about this method.
// For the number of item occurences that can be generated with this prefix:
for (int i = 0; i < prefix.getItemOccurencesTotalCount(); i++) {
Set<Integer> alreadyVisitedSID = new HashSet<Integer>();
// Create a Map of pairs to count the support of items (represented by a pair)
// in the ith semi-maximum periods
Map<PairBIDE, PairBIDE> mapPaires = new HashMap<PairBIDE, PairBIDE>();
// SOME CODE USED BY "findAllFrequentPairsForBackwardExtensionCheck"
Integer itemI = prefix.getIthItem(i); // iPeriod
Integer itemIm1 = null; // iPeriod -1
if (i > 0) {
itemIm1 = prefix.getIthItem(i - 1);
}
// // END NEW
// (1) For each i, we build the list of maximum periods
// for each sequence in the original database
for (PseudoSequenceBIDE pseudoSequence : projectedContext) {
int sequenceID = pseudoSequence.sequence.getId();
alreadyVisitedSID.add(pseudoSequence.sequence.getId());
Position currentCutPosition =
new Position(pseudoSequence.firstItemset, pseudoSequence.firstItem);
PseudoSequenceBIDE sequence = initialDatabase.get(sequenceID);
PseudoSequenceBIDE period =
sequence.getIthSemiMaximumPeriodOfAPrefix(prefix.getItemsets(), i, currentCutPosition);
if (period != null) {
// // we add it to the list of maximum periods
boolean hasExtension =
findAllFrequentPairsForBackwardExtensionCheck(
prefix.getAbsoluteSupport(),
prefix,
period,
i,
mapPaires,
itemI,
itemIm1,
currentCutPosition);
if (hasExtension) {
return true;
}
}
}
}
return false;
}
/**
* This is the "backscan-pruning" strategy described in the BIDE+ paper to avoid extending some
* prefixs that are guaranteed to not generate a closed pattern (see the BIDE+ paper for details).
*
* @param prefix the current prefix
* @return boolean true if we should not extend the prefix
*/
private boolean checkBackScanPruning(SequentialPattern prefix, Set<Integer> sidset) {
//
// See the BIDE+ paper for details about this method.
// For the number of item occurences that can be generated with this prefix:
for (int i = 0; i < prefix.getItemOccurencesTotalCount(); i++) {
Set<Integer> alreadyVisitedSID = new HashSet<Integer>();
// Create a Map of pairs to count the support of items (represented by a pair)
// in the ith semi-maximum periods
Map<PairBIDE, PairBIDE> mapPaires = new HashMap<PairBIDE, PairBIDE>();
// SOME CODE USED BY "findAllFrequentPairsForBackwardExtensionCheck"
Integer itemI = prefix.getIthItem(i); // iPeriod
Integer itemIm1 = null; // iPeriod -1
if (i > 0) {
itemIm1 = prefix.getIthItem(i - 1);
}
// // END NEW
int seqCount = 0;
// int highestSupportUntilNow = -1;
// (1) For each i, we build the list of maximum periods
// for each sequence in the original database
for (int sequenceID : sidset) {
alreadyVisitedSID.add(sequenceID);
PseudoSequenceBIDE sequence = initialDatabase.get(sequenceID);
PseudoSequenceBIDE period =
sequence.getIthSemiMaximumPeriodOfAPrefix(prefix.getItemsets(), i);
if (period != null) {
// // we add it to the list of maximum periods
boolean hasExtension =
findAllFrequentPairsForBackwardExtensionCheck(
alreadyVisitedSID.size(), prefix, period, i, mapPaires, itemI, itemIm1);
if (hasExtension) {
return true;
}
}
}
}
return false;
}
/**
* Method to check if a prefix has a backward-extension (see Bide+ article for full details). This
* method do it a little bit differently than the BIDE+ article since we iterate with i on
* elements of the prefix instead of iterating with a i on the itemsets of the prefix. But the
* idea is the same!
*
* @param prefix the current prefix
* @return boolean true, if there is a backward extension
*/
private boolean checkBackwardExtension(SequentialPattern prefix, Set<Integer> sidset) {
// System.out.println("======" + prefix);
int totalOccurenceCount = prefix.getItemOccurencesTotalCount();
// For the ith item of the prefix
for (int i = 0; i < totalOccurenceCount; i++) {
Set<Integer> alreadyVisitedSID = new HashSet<Integer>();
// // SOME CODE USED BY "findAllFrequentPairsForBackwardExtensionCheck"
Integer itemI = prefix.getIthItem(i); // iPeriod
Integer itemIm1 = null; // iPeriod -1
if (i > 0) {
itemIm1 = prefix.getIthItem(i - 1);
}
// // END NEW
// Create a Map of pairs to count the support of items (represented by a pair)
// in the ith semi-maximum periods
Map<PairBIDE, PairBIDE> mapPaires = new HashMap<PairBIDE, PairBIDE>();
// (1) For each i, we build the list of maximum periods
// for each sequence in the original database
// int seqCount =0;
int highestSupportUntilNow = -1;
// 1703 pat - 9391 ms
for (int sequenceID : sidset) {
// OPTIMIZATION PART 1== DON'T CHECK THE BACK EXTENSION IF THERE IS NOT ENOUGH SEQUENCE
// LEFT TO FIND AN EXTENSION
// THIS CAN IMPROVE THE PERFORMANCE BY UP TO 30% on FIFA
int remainingSeqID = (sidset.size() - alreadyVisitedSID.size());
if (highestSupportUntilNow != -1
&& highestSupportUntilNow + remainingSeqID < sidset.size()) {
break;
}
alreadyVisitedSID.add(sequenceID);
// if(!alreadyVisitedSID.contains(sequenceID)) {
// seqCount++;
// alreadyVisitedSID.add(sequenceID);
// }
// END OF OPTIMIZATION PART 1 (IT CONTINUES A FEW LINES BELOW...)
PseudoSequenceBIDE sequence = initialDatabase.get(sequenceID);
PseudoSequenceBIDE period = sequence.getIthMaximumPeriodOfAPrefix(prefix.getItemsets(), i);
// if the period is not null
if (period != null) {
boolean hasBackwardExtension =
findAllFrequentPairsForBackwardExtensionCheck(
alreadyVisitedSID.size(), prefix, period, i, mapPaires, itemI, itemIm1);
if (hasBackwardExtension) {
// System.out.println(prefix + " has a backward extension from " + i + "th
// maxperiod in sequence from seq. " + sequenceID );
return true;
}
// ===== OPTIMIZATION PART 2
if ((sidset.size() - alreadyVisitedSID.size()) < minsuppAbsolute) {
for (PairBIDE pair : mapPaires.values()) {
int supportOfPair = pair.getSequenceIDs().size();
if (supportOfPair > highestSupportUntilNow) {
highestSupportUntilNow =
supportOfPair; // +1 because it may be raised for this sequence...
}
}
}
// ===== END OF OPTIMIZATION PART 2
}
}
}
// totaltimeForBackwardExtension += System.currentTimeMillis() - start;
return false; // no backward extension, we return false
}