//---------------------------------------------------------------------------
// Register this parition from the concept for test data.
//---------------------------------------------------------------------------
bool CTDPartition::testRegisterPartition()
{
// Get the first record of the partition.
CTDRecord* pFirstRec = getRecord(0);
if (!pFirstRec) {
ASSERT(false);
return false;
}
int a = 0;
CTDPartAttrib* pPartAttrib = NULL;
CTDConcept* pCurrentConcept = NULL;
for (POSITION pos = m_partAttribs.GetHeadPosition(); pos != NULL; ++a) {
// Find the current concept of this attribute.
pPartAttrib = m_partAttribs.GetNext(pos);
pCurrentConcept = pFirstRec->getValue(a)->getCurrentConcept();
if (!pCurrentConcept) {
ASSERT(false);
return false;
}
pPartAttrib->m_relatedPos = pCurrentConcept->testRegisterPartition(this);
}
return true;
}
//---------------------------------------------------------------------------
// Deregister this parition from the concept.
//---------------------------------------------------------------------------
bool CTDPartition::deregisterPartition()
{
CTDRecord* pFirstRec = NULL;
if ( getNumRecords()== 0){
// Get the first record of the generalized records.
pFirstRec = getGenRecords()->GetAt(0);
}
else {
// Get the first record of the partition.
pFirstRec = getRecord(0);
if (!pFirstRec) {
ASSERT(false);
return false;
}
}
int a = 0;
CTDPartAttrib* pPartAttrib = NULL;
CTDConcept* pCurrentConcept = NULL;
for (POSITION pos = m_partAttribs.GetHeadPosition(); pos != NULL; ++a) {
// Find the current concept of this attribute.
pPartAttrib = m_partAttribs.GetNext(pos);
pCurrentConcept = pFirstRec->getValue(a)->getCurrentConcept();
if (!pCurrentConcept) {
ASSERT(false);
return false;
}
pCurrentConcept->deregisterPartition(pPartAttrib->m_relatedPos);
pPartAttrib->m_relatedPos = NULL;
}
return true;
}
//---------------------------------------------------------------------------
// Count the number of distortions.
// Each time a record is generalized from a child value to a parent value,
// we charge 1 unit of distortion. So if 100 records are involved in the
// generalization, we charge 100 unit.
//---------------------------------------------------------------------------
bool CTDEvalMgr::countNumDistortions(int& catDistortion, float& contDistortion)
{
cout << _T("Counting number of distortions...") << endl;
catDistortion = 0;
contDistortion = 0.0f;
int nRecs = 0, nValues = 0;
CTDRecord* pRec = NULL;
CTDAttrib* pAttrib = NULL;
CTDPartition* pPartition = NULL;
CTDValue* pValue = NULL;
CTDConcept* pCurrentConcept = NULL;
CTDConcept* pRawConcept = NULL;
CTDPartitions* pLeafPartitions = m_pPartitioner->getLeafPartitions();
// For each partition.
for (POSITION leafPos = pLeafPartitions->GetHeadPosition(); leafPos != NULL;) {
pPartition = pLeafPartitions->GetNext(leafPos);
nRecs = pPartition->getNumRecords();
// For each record.
for (int r = 0; r < nRecs; ++r) {
pRec = pPartition->getRecord(r);
nValues = pRec->getNumValues();
// For each value.
for (int v = 0; v < nValues; ++v) {
pAttrib = m_pAttribMgr->getAttribute(v);
if (!pAttrib->m_bVirtualAttrib)
continue;
pValue = pRec->getValue(v);
pCurrentConcept = pValue->getCurrentConcept();
if (pAttrib->isContinuous()) {
CTDContConcept* pContConcept = (CTDContConcept*) pCurrentConcept;
CTDContConcept* pRoot = (CTDContConcept*) pAttrib->getConceptRoot();
contDistortion += (pContConcept->m_upperBound - pContConcept->m_lowerBound) / (pRoot->m_upperBound - pRoot->m_lowerBound);
}
else {
pRawConcept = ((CTDStringValue*) pValue)->getRawConcept();
#if defined(_TD_SCORE_FUNTION_TRANSACTION)
// In case of transaction data, count a distortion only if suppressing "1".
if (pRawConcept->m_conceptValue.CompareNoCase(TD_TRANSACTION_ITEM_PRESENT) != 0)
continue;
#endif
if (pRawConcept->m_depth < 0 || pCurrentConcept->m_depth < 0) {
cout << _T("CSAEvalMgr::countNumDistortions: Negative depth.") << endl;
ASSERT(false);
return false;
}
catDistortion += pRawConcept->m_depth - pCurrentConcept->m_depth;
}
}
}
}
cout << _T("Counting number of distortions succeeded.") << endl;
return true;
}
//---------------------------------------------------------------------------
//---------------------------------------------------------------------------
bool CTDPartition::initGenRecords( CTDAttribs* pAttribs)
{
m_genRecords.cleanup();
int nAttribs = pAttribs->GetSize();
for (int classInd = 0; classInd < m_nClasses; ++classInd){
CTDAttrib* pAttrib = NULL;
CTDValue* pNewValue = NULL;
CTDRecord* pNewRecord = new CTDRecord();
for (int attribID = 0; attribID < nAttribs; ++attribID){
pNewValue = NULL;
pAttrib = pAttribs->GetAt(attribID);
if (pAttrib->isContinuous())
pNewValue = new CTDNumericValue(-1.0);
else
pNewValue = new CTDStringValue();
if (!pNewValue) {
ASSERT(false);
return false;
}
if (attribID == pAttribs->GetSize() - 1) {
// Class attribute
if (!pNewValue->assignGenClassValue(pAttrib, classInd))
return false;
}
else {
// Ordinary attribute
// Initialize the current concept to the root concept.
if (!pNewValue->initConceptToRoot(pAttrib))
return false;
}
// Add the value to the record.
if (!pNewRecord->addValue(pNewValue))
return false;
}
if (pNewRecord){
pNewRecord->setRecordID(m_genRecords.Add(pNewRecord));
}
}
if (m_genRecords.GetSize() != m_nClasses) {
cerr << _T("CTDPartition::initGenRecords: Number of generalized record is not current.") << endl;
return false;
}
return true;
}
//---------------------------------------------------------------------------
// Distribute records from parent paritition to child partitions.
//---------------------------------------------------------------------------
bool CTDPartitioner::testDistributeRecords(CTDPartition* pParentPartition,
CTDAttrib* pSplitAttrib,
CTDConcept* pSplitConcept,
CTDPartitions& childPartitions)
{
childPartitions.RemoveAll();
// Construct a partition for each child concept.
for (int childIdx = 0; childIdx < pSplitConcept->getNumChildConcepts(); ++childIdx)
childPartitions.AddTail(new CTDPartition( gTestPartitionIndex++, m_pAttribMgr->getAttributes()));
// Scan through each record in the parent partition and
// add records to the corresponding child partition based
// on the child concept.
CTDRecord* pRec = NULL;
CTDValue* pSplitValue = NULL;
POSITION childPartitionPos = NULL;
int childConceptIdx = -1;
int splitIdx = pSplitAttrib->m_attribIdx;
int nRecs = pParentPartition->getNumRecords();
ASSERT(nRecs > 0);
for (int r = 0; r < nRecs; ++r) {
pRec = pParentPartition->getRecord(r);
pSplitValue = pRec->getValue(splitIdx);
// Lower the concept by one level.
if (!pSplitValue->lowerCurrentConcept()) {
cerr << _T("CTDPartition: Should not specialize on this concept.");
childPartitions.cleanup();
ASSERT(false);
return false;
}
// Get the child concept of the current concept in this record.
childConceptIdx = pSplitValue->getCurrentConcept()->m_childIdx;
ASSERT(childConceptIdx != -1);
childPartitionPos = childPartitions.FindIndex(childConceptIdx);
ASSERT(childPartitionPos);
// Add the record to this child partition.
if (!childPartitions.GetAt(childPartitionPos)->addRecord(pRec)) {
childPartitions.cleanup();
ASSERT(false);
return false;
}
}
// Delete empty child partitions.
childPartitions.deleteEmptyPartitions();
return true;
}
//---------------------------------------------------------------------------
//---------------------------------------------------------------------------
bool CTDPartition::constructSupportMatrix(double epsilon)
{
CTDRecord* pFirstRec = NULL;
if ( getNumRecords()== 0){
// Get the first record of the generalized records.
pFirstRec = getGenRecords()->GetAt(0);
}
else {
// Get the first record of the partition.
pFirstRec = getRecord(0);
if (!pFirstRec) {
ASSERT(false);
return false;
}
}
int a = 0;
CTDPartAttrib* pPartAttrib = NULL;
CTDConcept* pCurrentConcept = NULL;
for (POSITION pos = m_partAttribs.GetHeadPosition(); pos != NULL; ++a) {
pPartAttrib = m_partAttribs.GetNext(pos);
if (!pPartAttrib->m_bCandidate) // pPartAttrib->m_bCandidate is true by default.
continue;
// Find the current concept of this attribute.
pCurrentConcept = pFirstRec->getValue(a)->getCurrentConcept();
// Need to find the split point. We also turn it true when we read the configuration file.
if (pPartAttrib->getActualAttrib()->m_bVirtualAttrib) {
if (!pCurrentConcept->divideConcept(epsilon, m_nClasses)) {
ASSERT(false);
return false;
}
}
// If this attribute does not have child concepts, this cannot be candidate.
if (pCurrentConcept->getNumChildConcepts() == 0) {
pCurrentConcept->m_bCutCandidate = false; // "true" by default.
pPartAttrib->m_bCandidate = false;
continue;
}
// Construct the support matrix.
if (!pPartAttrib->initSupportMatrix(pCurrentConcept, m_nClasses)) {
ASSERT(false);
return false;
}
}
// initializing the noisy class sum count
m_classNoisySums.SetSize(m_nClasses);
for (int j = 0; j < m_nClasses; ++j)
m_classNoisySums.SetAt(j, 0);
// Compute the support matrix
CTDConcept* pClassConcept = NULL;
CTDConcept* pLowerConcept = NULL;
CTDMDIntArray* pSupMatrix = NULL;
CTDIntArray* pSupSums = NULL;
CTDIntArray* pClassSums = NULL;
CTDRecord* pRec = NULL;
int nRecs = getNumRecords();
int classIdx = m_partAttribs.GetCount();
for (int r = 0; r < nRecs; ++r) {
pRec = getRecord(r);
// Get the class concept.
pClassConcept = pRec->getValue(classIdx)->getCurrentConcept();
++m_classNoisySums[pClassConcept->m_childIdx];
// Compute support counts for each attribute
int aIdx = 0;
for (POSITION pos = m_partAttribs.GetHeadPosition(); pos != NULL; ++aIdx) {
// The partition attribute.
pPartAttrib = m_partAttribs.GetNext(pos);
if (!pPartAttrib->m_bCandidate)
continue;
// Get the lower concept value
pLowerConcept = pRec->getValue(aIdx)->getLowerConcept();
if (!pLowerConcept) {
cerr << _T("No more child concepts. This should not be a candidate.") << endl;
ASSERT(false);
return false;
}
// Construct the support matrix.
pSupMatrix = pPartAttrib->getSupportMatrix();
if (!pSupMatrix) {
ASSERT(false);
return false;
}
++((*pSupMatrix)[pLowerConcept->m_childIdx][pClassConcept->m_childIdx]);
// Compute the support sum of this matrix.
pSupSums = pPartAttrib->getSupportSums();
if (!pSupSums) {
ASSERT(false);
return false;
}
++((*pSupSums)[pLowerConcept->m_childIdx]);
// Compute the class sum of this matrix.
pClassSums = pPartAttrib->getClassSums();
if (!pClassSums) {
ASSERT(false);
return false;
}
++((*pClassSums)[pClassConcept->m_childIdx]);
}
}
return true;
}
