PClassifier
TSimpleTreeLearner::operator()(PExampleGenerator ogen, const int &weight)
{
struct Example *examples, *ex;
struct SimpleTreeNode *tree;
struct Args args;
int cls_vals;
if (!ogen->domain->classVar)
raiseError("class-less domain");
if (!ogen->numberOfExamples() > 0)
raiseError("no examples");
/* create a tabel with pointers to examples */
ASSERT(examples = (struct Example *)calloc(ogen->numberOfExamples(), sizeof *examples));
ex = examples;
PEITERATE(ei, ogen) {
ex->example = &(*ei);
ex->weight = 1.0;
ex++;
}
bool TTreeStopCriteria::operator()(PExampleGenerator gen, const int &, PDomainContingency ocont)
{ int nor = gen->numberOfExamples();
if ((nor==0) || (nor==1))
return true; // example set is too small
char vt = gen->domain->classVar->varType;
if (vt!=TValue::INTVAR)
return false; // class is continuous, may continue
// is there more than one class left?
if (ocont) {
char ndcf = 0;
TDiscDistribution const &dva=CAST_TO_DISCDISTRIBUTION(ocont->classes);
const_ITERATE(TDiscDistribution, ci, dva)
if ((*ci>0) && (++ndcf==2))
return false; // at least two classes, may continue
}
else {
PTreeNode TTreeLearner::operator()(PExampleGenerator examples, const int &weightID, PDistribution apriorClass, vector<bool> &candidates, const int &depth)
{
PDomainContingency contingency;
PDomainDistributions domainDistributions;
PDistribution classDistribution;
if (!examples->numberOfExamples())
return PTreeNode();
if (contingencyComputer)
contingency = contingencyComputer->call(examples, weightID);
if (storeContingencies)
contingency = mlnew TDomainContingency(examples, weightID);
if (contingency)
classDistribution = contingency->classes;
else if (storeDistributions)
classDistribution = getClassDistribution(examples, weightID);
if (classDistribution) {
if (!classDistribution->abs)
return PTreeNode();
}
else
if (examples->weightOfExamples() < 1e-10)
return PTreeNode();
TTreeNode *utreeNode = mlnew TTreeNode();
PTreeNode treeNode = utreeNode;
utreeNode->weightID = weightID;
bool isLeaf = ((maxDepth>=0) && (depth == maxDepth)) || stop->call(examples, weightID, contingency);
if (isLeaf || storeNodeClassifier) {
utreeNode->nodeClassifier = nodeLearner
? nodeLearner->smartLearn(examples, weightID, contingency, domainDistributions, classDistribution)
: TMajorityLearner().smartLearn(examples, weightID, contingency, domainDistributions, classDistribution);
if (isLeaf) {
if (storeContingencies)
utreeNode->contingency = contingency;
if (storeDistributions)
utreeNode->distribution = classDistribution;
return treeNode;
}
}
utreeNode->contingency = contingency;
utreeNode->distribution = classDistribution;
float quality;
int spentAttribute;
utreeNode->branchSelector = split->call(utreeNode->branchDescriptions, utreeNode->branchSizes, quality, spentAttribute,
examples, weightID, contingency,
apriorClass ? apriorClass : classDistribution,
candidates, utreeNode->nodeClassifier);
isLeaf = !utreeNode->branchSelector;
bool hasNullNodes = false;
if (!isLeaf) {
if (spentAttribute>=0)
if (candidates[spentAttribute])
candidates[spentAttribute] = false;
else
spentAttribute = -1;
/* BEWARE: If you add an additional 'return' in the code below,
do not forget to restore the candidate's flag. */
utreeNode->branches = mlnew TTreeNodeList();
vector<int> newWeights;
PExampleGeneratorList subsets = exampleSplitter->call(treeNode, examples, weightID, newWeights);
if (!utreeNode->branchSizes)
utreeNode->branchSizes = branchSizesFromSubsets(subsets, weightID, newWeights);
if (!storeContingencies)
utreeNode->contingency = PDomainContingency();
if (!storeDistributions)
utreeNode->distribution = PDistribution();
vector<int>::iterator nwi = newWeights.begin(), nwe = newWeights.end();
PITERATE(TExampleGeneratorList, gi, subsets) {
if ((*gi)->numberOfExamples()) {
utreeNode->branches->push_back(call(*gi, (nwi!=nwe) ? *nwi : weightID, apriorClass, candidates, depth+1));
// Can store pointers to examples: the original is stored in the root
if (storeExamples && utreeNode->branches->back())
utreeNode->branches->back()->examples = *gi;
else if ((nwi!=nwe) && *nwi && (*nwi != weightID))
examples->removeMetaAttribute(*nwi);
}
else {
utreeNode->branches->push_back(PTreeNode());
hasNullNodes = true;
}
if (nwi!=nwe)
nwi++;
}
/* If I set it to false, it must had been true before
(otherwise, my TreeSplitConstructor wouldn't be allowed to spend it).
Hence, I can simply set it back to true... */
if (spentAttribute>=0)
candidates[spentAttribute] = true;
}
else { // no split constructed
if (!utreeNode->contingency)
PClassifier TTreeSplitConstructor_Combined::operator()(
PStringList &descriptions, PDiscDistribution &subsetSizes, float &quality, int &spentAttribute,
PExampleGenerator gen, const int &weightID ,
PDomainContingency dcont, PDistribution apriorClass,
const vector<bool> &candidates,
PClassifier nodeClassifier
)
{ checkProperty(discreteSplitConstructor);
checkProperty(continuousSplitConstructor);
vector<bool> discrete, continuous;
bool cse = candidates.size()==0;
vector<bool>::const_iterator ci(candidates.begin()), ce(candidates.end());
TVarList::const_iterator vi(gen->domain->attributes->begin()), ve(gen->domain->attributes->end());
for(; (cse || (ci!=ce)) && (vi!=ve); vi++) {
if (cse || *(ci++))
if ((*vi)->varType == TValue::INTVAR) {
discrete.push_back(true);
continuous.push_back(false);
continue;
}
else if ((*vi)->varType == TValue::FLOATVAR) {
discrete.push_back(false);
continuous.push_back(true);
continue;
}
discrete.push_back(false);
continuous.push_back(false);
}
float discQuality;
PStringList discDescriptions;
PDiscDistribution discSizes;
int discSpent;
PClassifier discSplit = discreteSplitConstructor->call(discDescriptions, discSizes, discQuality, discSpent,
gen, weightID, dcont, apriorClass, discrete, nodeClassifier);
float contQuality;
PStringList contDescriptions;
PDiscDistribution contSizes;
int contSpent;
PClassifier contSplit = continuousSplitConstructor->call(contDescriptions, contSizes, contQuality, contSpent,
gen, weightID, dcont, apriorClass, continuous, nodeClassifier);
int N = gen ? gen->numberOfExamples() : -1;
if (N<0)
N = dcont->classes->cases;
if ( discSplit
&& ( !contSplit
|| (discQuality>contQuality)
|| (discQuality==contQuality) && (N%2>0))) {
quality = discQuality;
descriptions = discDescriptions;
subsetSizes = discSizes;
spentAttribute = discSpent;
return discSplit;
}
else if (contSplit) {
quality = contQuality;
descriptions = contDescriptions;
subsetSizes = contSizes;
spentAttribute = contSpent;
return contSplit;
}
else
return returnNothing(descriptions, subsetSizes, quality, spentAttribute);
}
PClassifier TTreeSplitConstructor_ExhaustiveBinary::operator()(
PStringList &descriptions, PDiscDistribution &subsetSizes, float &quality, int &spentAttribute,
PExampleGenerator gen, const int &weightID ,
PDomainContingency dcont, PDistribution apriorClass,
const vector<bool> &candidates,
PClassifier
)
{
checkProperty(measure);
measure->checkClassTypeExc(gen->domain->classVar->varType);
PIntList bestMapping;
int wins, bestAttr;
PVariable bvar;
if (measure->needs==TMeasureAttribute::Generator) {
bool cse = candidates.size()==0;
bool haveCandidates = false;
vector<bool> myCandidates;
myCandidates.reserve(gen->domain->attributes->size());
vector<bool>::const_iterator ci(candidates.begin()), ce(candidates.end());
TVarList::const_iterator vi, ve(gen->domain->attributes->end());
for(vi = gen->domain->attributes->begin(); vi != ve; vi++) {
bool co = (*vi)->varType == TValue::INTVAR && (!cse || (ci!=ce) && *ci);
myCandidates.push_back(co);
haveCandidates = haveCandidates || co;
}
if (!haveCandidates)
return returnNothing(descriptions, subsetSizes, quality, spentAttribute);
PDistribution thisSubsets;
float thisQuality;
wins = 0;
int thisAttr = 0;
int N = gen->numberOfExamples();
TSimpleRandomGenerator rgen(N);
ci = myCandidates.begin();
for(vi = gen->domain->attributes->begin(); vi != ve; ci++, vi++, thisAttr++) {
if (*ci) {
thisSubsets = NULL;
PIntList thisMapping =
/*throughCont ? measure->bestBinarization(thisSubsets, thisQuality, *dci, dcont->classes, apriorClass, minSubset)
: */measure->bestBinarization(thisSubsets, thisQuality, *vi, gen, apriorClass, weightID, minSubset);
if (thisMapping
&& ( (!wins || (thisQuality>quality)) && ((wins=1)==1)
|| (thisQuality==quality) && rgen.randbool(++wins))) {
bestAttr = thisAttr;
quality = thisQuality;
subsetSizes = thisSubsets;
bestMapping = thisMapping;
}
}
/*if (thoughCont)
dci++; */
}
if (!wins)
return returnNothing(descriptions, subsetSizes, quality, spentAttribute);
if (quality<worstAcceptable)
return returnNothing(descriptions, subsetSizes, spentAttribute);
if (subsetSizes && subsetSizes->variable)
bvar = subsetSizes->variable;
else {
TEnumVariable *evar = mlnew TEnumVariable("");
evar->addValue("0");
evar->addValue("1");
bvar = evar;
}
}
else {
bool cse = candidates.size()==0;
if (!cse && noCandidates(candidates))
return returnNothing(descriptions, subsetSizes, quality, spentAttribute);
if (!dcont || dcont->classIsOuter) {
dcont = PDomainContingency(mlnew TDomainContingency(gen, weightID));
// raiseWarningWho("TreeSplitConstructor_ExhaustiveBinary", "this class is not optimized for 'candidates' list and can be very slow");
}
int N = gen ? gen->numberOfExamples() : -1;
if (N<0)
N = dcont->classes->cases;
TSimpleRandomGenerator rgen(N);
PDistribution classDistribution = dcont->classes;
vector<bool>::const_iterator ci(candidates.begin()), ce(candidates.end());
TDiscDistribution *dis0, *dis1;
TContDistribution *con0, *con1;
int thisAttr = 0;
bestAttr = -1;
wins = 0;
quality = 0.0;
float leftExamples, rightExamples;
TDomainContingency::iterator dci(dcont->begin()), dce(dcont->end());
for(; (cse || (ci!=ce)) && (dci!=dce); dci++, thisAttr++) {
// We consider the attribute only if it is a candidate, discrete and has at least two values
if ((cse || *(ci++)) && ((*dci)->outerVariable->varType==TValue::INTVAR) && ((*dci)->discrete->size()>=2)) {
const TDistributionVector &distr = *(*dci)->discrete;
if (distr.size()>16)
raiseError("'%s' has more than 16 values, cannot exhaustively binarize", gen->domain->attributes->at(thisAttr)->get_name().c_str());
// If the attribute is binary, we check subsetSizes and assess the quality if they are OK
if (distr.size()==2) {
if ((distr.front()->abs<minSubset) || (distr.back()->abs<minSubset))
continue; // next attribute
else {
float thisMeas = measure->call(thisAttr, dcont, apriorClass);
if ( ((!wins || (thisMeas>quality)) && ((wins=1)==1))
|| ((thisMeas==quality) && rgen.randbool(++wins))) {
bestAttr = thisAttr;
quality = thisMeas;
leftExamples = distr.front()->abs;
rightExamples = distr.back()->abs;
bestMapping = mlnew TIntList(2, 0);
bestMapping->at(1) = 1;
}
continue;
}
}
vector<int> valueIndices;
int ind = 0;
for(TDistributionVector::const_iterator dvi(distr.begin()), dve(distr.end()); (dvi!=dve); dvi++, ind++)
if ((*dvi)->abs>0)
valueIndices.push_back(ind);
if (valueIndices.size()<2)
continue;
PContingency cont = prepareBinaryCheat(classDistribution, *dci, bvar, dis0, dis1, con0, con1);
// A real job: go through all splits
int binWins = 0;
float binQuality = -1.0;
float binLeftExamples = -1.0, binRightExamples = -1.0;
// Selection: each element correspons to a value of the original attribute and is 1, if the value goes right
// The first value always goes left (and has no corresponding bit in selection.
TBoolCount selection(valueIndices.size()-1), bestSelection(0);
// First for discrete classes
if (dis0) {
do {
*dis0 = CAST_TO_DISCDISTRIBUTION(distr[valueIndices[0]]);
*dis1 *= 0;
vector<int>::const_iterator ii(valueIndices.begin());
ii++;
for(TBoolCount::const_iterator bi(selection.begin()), be(selection.end()); bi!=be; bi++, ii++)
*(*bi ? dis1 : dis0) += distr[*ii];
cont->outerDistribution->setint(0, dis0->abs);
cont->outerDistribution->setint(1, dis1->abs);
if ((dis0->abs < minSubset) || (dis1->abs < minSubset))
continue; // cannot split like that, to few examples in one of the branches
float thisMeas = measure->operator()(cont, classDistribution, apriorClass);
if ( ((!binWins) || (thisMeas>binQuality)) && ((binWins=1) ==1)
|| (thisMeas==binQuality) && rgen.randbool(++binWins)) {
bestSelection = selection;
binQuality = thisMeas;
binLeftExamples = dis0->abs;
binRightExamples = dis1->abs;
}
} while (selection.next());
}
// And then exactly the same for continuous classes
else {
do {
*con0 = CAST_TO_CONTDISTRIBUTION(distr[0]);
*con1 = TContDistribution();
vector<int>::const_iterator ii(valueIndices.begin());
for(TBoolCount::const_iterator bi(selection.begin()), be(selection.end()); bi!=be; bi++, ii++)
*(*bi ? con1 : con0) += distr[*ii];
if ((con0->abs<minSubset) || (con1->abs<minSubset))
continue; // cannot split like that, to few examples in one of the branches
float thisMeas = measure->operator()(cont, classDistribution, apriorClass);
if ( ((!binWins) || (thisMeas>binQuality)) && ((binWins=1) ==1)
|| (thisMeas==binQuality) && rgen.randbool(++binWins)) {
bestSelection = selection;
binQuality = thisMeas;
binLeftExamples = con0->abs;
binRightExamples = con1->abs;
}
} while (selection.next());
}
if ( binWins
&& ( (!wins || (binQuality>quality)) && ((wins=1)==1)
|| (binQuality==quality) && rgen.randbool(++wins))) {
bestAttr = thisAttr;
quality = binQuality;
leftExamples = binLeftExamples;
rightExamples = binRightExamples;
bestMapping = mlnew TIntList(distr.size(), -1);
vector<int>::const_iterator ii = valueIndices.begin();
bestMapping->at(*(ii++)) = 0;
ITERATE(TBoolCount, bi, bestSelection)
bestMapping->at(*(ii++)) = *bi ? 1 : 0;
}
}
}
if (!wins)
return returnNothing(descriptions, subsetSizes, quality, spentAttribute);
subsetSizes = mlnew TDiscDistribution();
subsetSizes->addint(0, leftExamples);
subsetSizes->addint(1, rightExamples);
}
PVariable attribute = gen->domain->attributes->at(bestAttr);
if (attribute->noOfValues() == 2) {
spentAttribute = bestAttr;
descriptions = mlnew TStringList(attribute.AS(TEnumVariable)->values.getReference());
TClassifierFromVarFD *cfv = mlnew TClassifierFromVarFD(attribute, gen->domain, bestAttr, subsetSizes);
cfv->transformUnknowns = false;
return cfv;
}
string s0, s1;
int ns0 = 0, ns1 = 0;
TValue ev;
attribute->firstValue(ev);
PITERATE(TIntList, mi, bestMapping) {
string str;
attribute->val2str(ev, str);
if (*mi==1) {
s1 += string(ns1 ? ", " : "") + str;
ns1++;
}
else if (*mi==0) {
s0 += string(ns0 ? ", " : "") + str;
ns0++;
}
attribute->nextValue(ev);
}
PClassifier TTreeSplitConstructor_Attribute::operator()(
PStringList &descriptions, PDiscDistribution &subsetSizes, float &quality, int &spentAttribute,
PExampleGenerator gen, const int &weightID,
PDomainContingency dcont, PDistribution apriorClass,
const vector<bool> &candidates,
PClassifier nodeClassifier
)
{ checkProperty(measure);
measure->checkClassTypeExc(gen->domain->classVar->varType);
bool cse = candidates.size()==0;
vector<bool>::const_iterator ci(candidates.begin()), ce(candidates.end());
if (!cse) {
if (noCandidates(candidates))
return returnNothing(descriptions, subsetSizes, quality, spentAttribute);
ci = candidates.begin();
}
int N = gen ? gen->numberOfExamples() : -1;
if (N<0)
N = dcont->classes->cases;
TSimpleRandomGenerator rgen(N);
int thisAttr = 0, bestAttr = -1, wins = 0;
quality = 0.0;
if (measure->needs == TMeasureAttribute::Contingency_Class) {
vector<bool> myCandidates;
if (cse) {
myCandidates.reserve(gen->domain->attributes->size());
PITERATE(TVarList, vi, gen->domain->attributes)
myCandidates.push_back((*vi)->varType == TValue::INTVAR);
}
else {
myCandidates.reserve(candidates.size());
TVarList::const_iterator vi(gen->domain->attributes->begin());
for(; ci != ce; ci++, vi++)
myCandidates.push_back(*ci && ((*vi)->varType == TValue::INTVAR));
}
if (!dcont || dcont->classIsOuter)
dcont = PDomainContingency(mlnew TDomainContingency(gen, weightID, myCandidates));
ci = myCandidates.begin();
ce = myCandidates.end();
TDomainContingency::iterator dci(dcont->begin()), dce(dcont->end());
for(; (ci != ce) && (dci!=dce); dci++, ci++, thisAttr++)
if (*ci && checkDistribution((const TDiscDistribution &)((*dci)->outerDistribution.getReference()), minSubset)) {
float thisMeas = measure->call(thisAttr, dcont, apriorClass);
if ( ((!wins || (thisMeas>quality)) && ((wins=1)==1))
|| ((thisMeas==quality) && rgen.randbool(++wins))) {
quality = thisMeas;
subsetSizes = (*dci)->outerDistribution;
bestAttr = thisAttr;
}
}
}
else if (measure->needs == TMeasureAttribute::DomainContingency) {
if (!dcont || dcont->classIsOuter)
dcont = PDomainContingency(mlnew TDomainContingency(gen, weightID));
TDomainContingency::iterator dci(dcont->begin()), dce(dcont->end());
for(; (cse || (ci!=ce)) && (dci!=dce); dci++, thisAttr++)
if ( (cse || *(ci++))
&& ((*dci)->outerVariable->varType==TValue::INTVAR)
&& checkDistribution((const TDiscDistribution &)((*dci)->outerDistribution.getReference()), minSubset)) {
float thisMeas = measure->call(thisAttr, dcont, apriorClass);
if ( ((!wins || (thisMeas>quality)) && ((wins=1)==1))
|| ((thisMeas==quality) && rgen.randbool(++wins))) {
quality = thisMeas;
subsetSizes = (*dci)->outerDistribution;
bestAttr = thisAttr;
}
}
}
else {
TDomainDistributions ddist(gen, weightID);
TDomainDistributions::iterator ddi(ddist.begin()), dde(ddist.end()-1);
for(; (cse || (ci!=ce)) && (ddi!=dde); ddi++, thisAttr++)
if (cse || *(ci++)) {
TDiscDistribution *discdist = (*ddi).AS(TDiscDistribution);
if (discdist && checkDistribution(*discdist, minSubset)) {
float thisMeas = measure->call(thisAttr, gen, apriorClass, weightID);
if ( ((!wins || (thisMeas>quality)) && ((wins=1)==1))
|| ((thisMeas==quality) && rgen.randbool(++wins))) {
quality = thisMeas;
subsetSizes = PDiscDistribution(*ddi); // not discdist - this would be double wrapping!
bestAttr = thisAttr;
}
}
}
}
if (!wins)
return returnNothing(descriptions, subsetSizes, quality, spentAttribute);
if (quality<worstAcceptable)
return returnNothing(descriptions, subsetSizes, spentAttribute);
PVariable attribute = gen->domain->attributes->at(bestAttr);
TEnumVariable *evar = attribute.AS(TEnumVariable);
if (evar)
descriptions = mlnew TStringList(evar->values.getReference());
else
descriptions = mlnew TStringList(subsetSizes->size(), "");
spentAttribute = bestAttr;
TClassifierFromVarFD *cfv = mlnew TClassifierFromVarFD(attribute, gen->domain, bestAttr, subsetSizes);
cfv->transformUnknowns = false;
return cfv;
}