/**
* This methods checks if a seq. pattern "pattern2" is strictly contained in a seq. pattern
* "pattern1".
*
* @param pattern1 a sequential pattern
* @param pattern2 another sequential pattern
* @return true if the pattern1 contains pattern2.
*/
boolean strictlyContains(PrefixVMSP pattern1, PrefixVMSP pattern2) {
// // if pattern2 is larger or equal in size, then it cannot be contained in pattern1
// if(pattern1.size() <= pattern2.size()){
// return false;
// }
// To see if pattern2 is strictly contained in pattern1,
// we will search for each itemset i of pattern2 in pattern1 by advancing
// in pattern 1 one itemset at a time.
int i = 0; // position in pattern2
int j = 0; // position in pattern1
while (true) {
// if the itemset at current position in pattern1 contains the itemset
// at current position in pattern2
if (pattern1.get(j).containsAll(pattern2.get(i))) {
// go to next itemset in pattern2
i++;
// if we reached the end of pattern2, then return true
if (i == pattern2.size()) {
return true;
}
}
// go to next itemset in pattern1
j++;
// if we reached the end of pattern1, then pattern2 is not strictly included
// in it, and return false
if (j >= pattern1.size()) {
return false;
}
// // lastly, for optimization, we check how many itemsets are left to be matched.
// // if there is less itemsets left in pattern1 than in pattern2, then it will
// // be impossible to get a total match, and so we return false.
if ((pattern1.size() - j) < pattern2.size() - i) {
return false;
}
}
}
/**
* This is the dfsPruning method as described in the SPAM paper.
*
* @param prefix the current prefix
* @param prefixBitmap the bitmap corresponding to the current prefix
* @param sn a list of items to be considered for i-steps
* @param in a list of items to be considered for s-steps
* @param hasToBeGreaterThanForIStep
* @param m size of the current prefix in terms of items
* @param lastAppendedItem the last appended item to the prefix
* @throws IOException if there is an error writing a pattern to the output file
* @return TRUE IF A FREQUENT PATTERN WAS CREATED USING THE PREFIX.
*/
boolean dfsPruning(
PrefixVMSP prefix,
Bitmap prefixBitmap,
List<Integer> sn,
List<Integer> in,
int hasToBeGreaterThanForIStep,
int m,
Integer lastAppendedItem)
throws IOException {
boolean atLeastOneFrequentExtension = false;
// System.out.println(prefix.toString());
// ====== S-STEPS ======
// Temporary variables (as described in the paper)
List<Integer> sTemp = new ArrayList<Integer>();
List<Bitmap> sTempBitmaps = new ArrayList<Bitmap>();
// for CMAP pruning, we will only check against the last appended item
Map<Integer, Integer> mapSupportItemsAfter = coocMapAfter.get(lastAppendedItem);
// for each item in sn
loopi:
for (Integer i : sn) {
// LAST POSITION PRUNING
/*if (useLastPositionPruning && lastItemPositionMap.get(i) < prefixBitmap.firstItemsetID) {
// System.out.println("TEST");
continue loopi;
}*/
// CMAP PRUNING
// we only check with the last appended item
if (useCMAPPruning) {
if (mapSupportItemsAfter == null) {
continue loopi;
}
Integer support = mapSupportItemsAfter.get(i);
if (support == null || support < minsup) {
// System.out.println("PRUNE");
continue loopi;
}
}
// perform the S-STEP with that item to get a new bitmap
Bitmap.INTERSECTION_COUNT++;
Bitmap newBitmap =
prefixBitmap.createNewBitmapSStep(verticalDB.get(i), sequencesSize, lastBitIndex, maxGap);
// if the support is higher than minsup
if (newBitmap.getSupportWithoutGapTotal() >= minsup) {
// record that item and pattern in temporary variables
sTemp.add(i);
sTempBitmaps.add(newBitmap);
}
}
// for each pattern recorded for the s-step
for (int k = 0; k < sTemp.size(); k++) {
// STRATEGY: NEWWW
atLeastOneFrequentExtension = true;
int item = sTemp.get(k);
// create the new prefix
PrefixVMSP prefixSStep = prefix.cloneSequence();
prefixSStep.addItemset(new Itemset(item));
if (item % 2 == 0) {
prefixSStep.sumOfEvenItems = item + prefix.sumOfEvenItems;
prefixSStep.sumOfOddItems = prefix.sumOfOddItems;
} else {
prefixSStep.sumOfEvenItems = prefix.sumOfEvenItems;
prefixSStep.sumOfOddItems = item + prefix.sumOfOddItems;
}
// prefixSStep.sumOfItems = item + prefix.sumOfItems;
// create the new bitmap
Bitmap newBitmap = sTempBitmaps.get(k);
// save the pattern to the file
if (newBitmap.getSupport() >= minsup) {
boolean hasFrequentExtension = false;
// recursively try to extend that pattern
if (maximumPatternLength > m) {
hasFrequentExtension =
dfsPruning(prefixSStep, newBitmap, sTemp, sTemp, item, m + 1, item);
}
if (hasFrequentExtension == false) {
savePatternMultipleItems(prefixSStep, newBitmap, m);
}
}
}
Map<Integer, Integer> mapSupportItemsEquals = coocMapEquals.get(lastAppendedItem);
// ======== I STEPS =======
// Temporary variables
List<Integer> iTemp = new ArrayList<Integer>();
List<Bitmap> iTempBitmaps = new ArrayList<Bitmap>();
// for each item in in
loop2:
for (Integer i : in) {
// the item has to be greater than the largest item
// already in the last itemset of prefix.
if (i > hasToBeGreaterThanForIStep) {
// LAST POSITION PRUNING
/*if (useLastPositionPruning && lastItemPositionMap.get(i) < prefixBitmap.firstItemsetID) {
continue loop2;
}*/
// CMAP PRUNING
if (useCMAPPruning) {
if (mapSupportItemsEquals == null) {
continue loop2;
}
Integer support = mapSupportItemsEquals.get(i);
if (support == null || support < minsup) {
continue loop2;
}
}
// Perform an i-step with this item and the current prefix.
// This creates a new bitmap
Bitmap.INTERSECTION_COUNT++;
Bitmap newBitmap =
prefixBitmap.createNewBitmapIStep(verticalDB.get(i), sequencesSize, lastBitIndex);
// If the support is no less than minsup
if (newBitmap.getSupport() >= minsup) {
// record that item and pattern in temporary variables
iTemp.add(i);
iTempBitmaps.add(newBitmap);
}
}
}
// for each pattern recorded for the i-step
for (int k = 0; k < iTemp.size(); k++) { // STRATEGY: NEWWW
atLeastOneFrequentExtension = true;
int item = iTemp.get(k);
// create the new prefix
PrefixVMSP prefixIStep = prefix.cloneSequence();
prefixIStep.getItemsets().get(prefixIStep.size() - 1).addItem(item);
if (item % 2 == 0) {
prefixIStep.sumOfEvenItems = item + prefix.sumOfEvenItems;
prefixIStep.sumOfOddItems = prefix.sumOfOddItems;
} else {
prefixIStep.sumOfEvenItems = prefix.sumOfEvenItems;
prefixIStep.sumOfOddItems = item + prefix.sumOfOddItems;
}
// create the new bitmap
Bitmap newBitmap = iTempBitmaps.get(k);
// recursively try to extend that pattern
boolean hasFrequentExtension = false;
if (maximumPatternLength > m) {
hasFrequentExtension = dfsPruning(prefixIStep, newBitmap, sTemp, iTemp, item, m + 1, item);
}
if (hasFrequentExtension == false) {
// save the pattern
savePatternMultipleItems(prefixIStep, newBitmap, m);
}
}
// check the memory usage
MemoryLogger.getInstance().checkMemory();
return atLeastOneFrequentExtension || useStrategyForwardExtensionChecking == false;
}
/**
* Save a pattern of size 1 to the output file
*
* @param item the item
* @param bitmap its bitmap
* @param itemIsEven
* @throws IOException exception if error while writing to the file
* @return true if is subsumed
*/
private boolean savePatternSingleItem(Integer item, Bitmap bitmap, boolean itemIsEven)
throws IOException {
// System.out.println("prefix :" + prefix);
// FOR THE CASE OF SINGLE ITEM, WE DON'T NEED TO DO SUB-PATTERN CHECKING:
// WE JUST NEED TO DO SUPER-PATTERN CHECKING
// #################
// IMPORTANT STRATEGY : FROM LARGER TO SMALLER...... AND IN ASCENDING SUPPORT ORDER
// ##################
if (itemIsEven) {
for (int i = maxPatterns.size() - 1; i > 1; i--) {
for (PatternVMSP pPrime : maxPatterns.get(i)) {
if (pPrime.prefix.sumOfOddItems + pPrime.prefix.sumOfEvenItems < item) {
break;
}
// if the pattern already found contains the single item
if (pPrime.prefix.sumOfEvenItems > item && bitmap.getSupport() >= pPrime.support) {
if (pPrime.prefix.containsItem(item)) {
return true;
}
}
}
}
} else {
for (int i = maxPatterns.size() - 1; i > 1; i--) {
for (PatternVMSP pPrime : maxPatterns.get(i)) {
if (pPrime.prefix.sumOfOddItems + pPrime.prefix.sumOfEvenItems < item) {
break;
}
// if the pattern already found contains the single item
if (pPrime.prefix.sumOfOddItems > item && bitmap.getSupport() >= pPrime.support) {
if (pPrime.prefix.containsItem(item)) {
return true;
}
}
}
}
}
// OTHERWISE THE NEW PATTERN IS NOT SUBSUMMED
patternCount++; // INCREASE COUNT
PrefixVMSP prefix = new PrefixVMSP();
prefix.addItemset(new Itemset(item));
if (itemIsEven) {
prefix.sumOfEvenItems = item;
prefix.sumOfOddItems = 0;
} else {
prefix.sumOfEvenItems = 0;
prefix.sumOfOddItems = item;
}
PatternVMSP pattern = new PatternVMSP(prefix, bitmap.getSupport());
// If the user wants to see the sequence identifiers, we need to keep the bitmap which stores
// them
if (outputSequenceIdentifiers) {
pattern.bitmap = bitmap;
}
// save the pattern
// System.out.println(" ADD: " + item);
maxPatterns.get(1).add(pattern);
return false;
// END CHANGED ------
}
/**
* This is the main method for the VMSP algorithm
*
* @param an input file
* @param minsupRel the minimum support as a relative value
* @throws IOException
*/
private void vmsp(String input, double minsupRel) throws IOException {
// create maxPattern array
maxPatterns = new ArrayList<TreeSet<PatternVMSP>>(20);
maxPatterns.add(null);
maxPatterns.add(new TreeSet<PatternVMSP>());
// the structure to store the vertical database
// key: an item value : bitmap
verticalDB = new HashMap<Integer, Bitmap>();
// structure to store the horizontal database
List<int[]> inMemoryDB = new ArrayList<int[]>();
// STEP 0: SCAN THE DATABASE TO STORE THE FIRST BIT POSITION OF EACH SEQUENCE
// AND CALCULATE THE TOTAL NUMBER OF BIT FOR EACH BITMAP
sequencesSize = new ArrayList<Integer>();
lastBitIndex = 0; // variable to record the last bit position that we will use in bitmaps
try {
// read the file
FileInputStream fin = new FileInputStream(new File(input));
BufferedReader reader = new BufferedReader(new InputStreamReader(fin));
String thisLine;
int bitIndex = 0;
// for each line (sequence) in the file until the end
while ((thisLine = reader.readLine()) != null) {
// if the line is a comment, is empty or is a
// kind of metadata
if (thisLine.isEmpty() == true
|| thisLine.charAt(0) == '#'
|| thisLine.charAt(0) == '%'
|| thisLine.charAt(0) == '@') {
continue;
}
// record the length of the current sequence (for optimizations)
sequencesSize.add(bitIndex);
// split the sequence according to spaces into tokens
String tokens[] = thisLine.split(" ");
int[] transactionArray = new int[tokens.length];
inMemoryDB.add(transactionArray);
for (int i = 0; i < tokens.length; i++) {
int item = Integer.parseInt(tokens[i]);
transactionArray[i] = item;
// if it is not an itemset separator
if (item == -1) { // indicate the end of an itemset
// increase the number of bits that we will need for each bitmap
bitIndex++;
}
}
}
// record the last bit position for the bitmaps
lastBitIndex = bitIndex - 1;
reader.close(); // close the input file
} catch (Exception e) {
e.printStackTrace();
}
// Calculate the absolute minimum support
// by multipling the percentage with the number of
// sequences in this database
// minsup = 163;
minsup = (int) Math.ceil((minsupRel * sequencesSize.size()));
if (minsup == 0) {
minsup = 1;
}
// System.out.println("minsup : " + minsup);
// STEP1: SCAN THE DATABASE TO CREATE THE BITMAP VERTICAL DATABASE REPRESENTATION
try {
FileInputStream fin = new FileInputStream(new File(input));
BufferedReader reader = new BufferedReader(new InputStreamReader(fin));
String thisLine;
int sid = 0; // to know which sequence we are scanning
int tid = 0; // to know which itemset we are scanning
// for each line (sequence) from the input file
while ((thisLine = reader.readLine()) != null) {
// split the sequence according to spaces into tokens
for (String token : thisLine.split(" ")) {
if (token.equals("-1")) { // indicate the end of an itemset
tid++;
} else if (token.equals("-2")) { // indicate the end of a sequence
// determineSection(bitindex - previousBitIndex); // register the sequence length
// for the bitmap
sid++;
tid = 0;
} else { // indicate an item
// Get the bitmap for this item. If none, create one.
Integer item = Integer.parseInt(token);
Bitmap bitmapItem = verticalDB.get(item);
if (bitmapItem == null) {
bitmapItem = new Bitmap(lastBitIndex);
verticalDB.put(item, bitmapItem);
}
// Register the bit in the bitmap for this item
bitmapItem.registerBit(sid, tid, sequencesSize);
}
}
}
reader.close();
} catch (Exception e) {
e.printStackTrace();
}
// STEP2: REMOVE INFREQUENT ITEMS FROM THE DATABASE BECAUSE THEY WILL NOT APPEAR IN ANY FREQUENT
// SEQUENTIAL PATTERNS
List<Integer> frequentItems = new ArrayList<Integer>();
Iterator<Entry<Integer, Bitmap>> iter = verticalDB.entrySet().iterator();
// we iterate over items from the vertical database that we have in memory
while (iter.hasNext()) {
// we get the bitmap for this item
Entry<Integer, Bitmap> entry = (Entry<Integer, Bitmap>) iter.next();
// if the cardinality of this bitmap is lower than minsup
if (entry.getValue().getSupport() < minsup) {
// we remove this item from the database.
iter.remove();
} else {
// otherwise, we save this item as a frequent
// sequential pattern of size 1
// CHANGED
// and we add this item to a list of frequent items
// that we will use later.
frequentItems.add(entry.getKey());
// END CHANGED
}
}
// SET 2.1 SORT ITEMS BY DESCENDING SUPPORT
Collections.sort(
frequentItems,
new Comparator<Integer>() {
@Override
public int compare(Integer arg0, Integer arg1) {
return verticalDB.get(arg0).getSupport() - verticalDB.get(arg1).getSupport();
}
});
// STEP 3.1 CREATE CMAP
coocMapEquals = new HashMap<Integer, Map<Integer, Integer>>(frequentItems.size());
coocMapAfter = new HashMap<Integer, Map<Integer, Integer>>(frequentItems.size());
if (useLastPositionPruning) {
lastItemPositionMap = new HashMap<Integer, Short>(frequentItems.size());
}
for (int[] transaction : inMemoryDB) {
short itemsetCount = 0;
Set<Integer> alreadyProcessed = new HashSet<Integer>();
Map<Integer, Set<Integer>> equalProcessed = new HashMap<>();
loopI:
for (int i = 0; i < transaction.length; i++) {
Integer itemI = transaction[i];
Set<Integer> equalSet = equalProcessed.get(itemI);
if (equalSet == null) {
equalSet = new HashSet<Integer>();
equalProcessed.put(itemI, equalSet);
}
if (itemI < 0) {
itemsetCount++;
continue;
}
// System.out.println(itemsetCount);
// update lastItemMap
if (useLastPositionPruning) {
Short last = lastItemPositionMap.get(itemI);
if (last == null || last < itemsetCount) {
lastItemPositionMap.put(itemI, itemsetCount);
}
}
Bitmap bitmapOfItem = verticalDB.get(itemI);
if (bitmapOfItem == null || bitmapOfItem.getSupport() < minsup) {
continue;
}
Set<Integer> alreadyProcessedB = new HashSet<Integer>(); // NEW
boolean sameItemset = true;
for (int j = i + 1; j < transaction.length; j++) {
Integer itemJ = transaction[j];
if (itemJ < 0) {
sameItemset = false;
continue;
}
Bitmap bitmapOfitemJ = verticalDB.get(itemJ);
if (bitmapOfitemJ == null || bitmapOfitemJ.getSupport() < minsup) {
continue;
}
// if (itemI != itemJ){
Map<Integer, Integer> map = null;
if (sameItemset) {
if (!equalSet.contains(itemJ)) {
map = coocMapEquals.get(itemI);
if (map == null) {
map = new HashMap<Integer, Integer>();
coocMapEquals.put(itemI, map);
}
Integer support = map.get(itemJ);
if (support == null) {
map.put(itemJ, 1);
} else {
map.put(itemJ, ++support);
}
equalSet.add(itemJ);
}
} else if (!alreadyProcessedB.contains(itemJ)) {
if (alreadyProcessed.contains(itemI)) {
continue loopI;
}
map = coocMapAfter.get(itemI);
if (map == null) {
map = new HashMap<Integer, Integer>();
coocMapAfter.put(itemI, map);
}
Integer support = map.get(itemJ);
if (support == null) {
map.put(itemJ, 1);
} else {
map.put(itemJ, ++support);
}
alreadyProcessedB.add(itemJ); // NEW
}
}
alreadyProcessed.add(itemI);
}
}
// STEP3: WE PERFORM THE RECURSIVE DEPTH FIRST SEARCH
// to find longer sequential patterns recursively
if (maximumPatternLength == 1) {
return;
}
// for each frequent item
for (Entry<Integer, Bitmap> entry : verticalDB.entrySet()) {
// We create a prefix with that item
PrefixVMSP prefix = new PrefixVMSP();
prefix.addItemset(new Itemset(entry.getKey()));
boolean itemIsEven = entry.getKey() % 2 == 0;
if (itemIsEven) {
prefix.sumOfEvenItems = (Integer) entry.getKey();
prefix.sumOfOddItems = 0;
} else {
prefix.sumOfEvenItems = 0;
prefix.sumOfOddItems = (Integer) entry.getKey();
}
boolean hasExtension =
dfsPruning(
prefix,
entry.getValue(),
frequentItems,
frequentItems,
entry.getKey(),
2,
entry.getKey());
if (hasExtension == false) {
savePatternSingleItem(entry.getKey(), entry.getValue(), itemIsEven);
}
// We call the depth first search method with that prefix
// and the list of frequent items to try to find
// larger sequential patterns by appending some of these
// items.
// if(!isSubsumedAndNonClosed) {
// }
}
}