PClassifier TLogRegLearner::fitModel(PExampleGenerator gen, const int &weight, int &error, PVariable &errorAt)
{
PImputer imputer = imputerConstructor ? imputerConstructor->call(gen, weight) : PImputer();
PExampleGenerator imputed = imputer ? imputer->call(gen, weight) : gen;
// construct classifier
TLogRegClassifier *lrc = mlnew TLogRegClassifier(imputed->domain);
lrc->dataDescription = mlnew TEFMDataDescription(gen->domain, mlnew TDomainDistributions(gen), 0, getMetaID());
PClassifier cl = lrc;
lrc->imputer = imputer;
//if (imputed->domain->hasDiscreteAttributes(false)) {
lrc->continuizedDomain = domainContinuizer ? domainContinuizer->call(imputed, weight) : (*logisticRegressionDomainContinuizer)(imputed, weight);
imputed = mlnew TExampleTable(lrc->continuizedDomain, imputed);
//}
// copy class value
// construct a LR fitter
fitter = fitter ? fitter : PLogRegFitter(mlnew TLogRegFitter_Cholesky());
PAttributedFloatList temp_beta, temp_beta_se;
// fit logistic regression
temp_beta = fitter->call(imputed, weight, temp_beta_se, lrc->likelihood, error, errorAt);
lrc->fit_status = error;
// transform beta to AttributedList
PVarList enum_attributes = mlnew TVarList();
enum_attributes->push_back(imputed->domain->classVar);
PITERATE(TVarList, vl, imputed->domain->attributes)
enum_attributes->push_back(*vl);
// tranfsorm *beta into a PFloatList
lrc->beta=mlnew TAttributedFloatList(enum_attributes);
lrc->beta_se=mlnew TAttributedFloatList(enum_attributes);
PITERATE(TAttributedFloatList, fi, temp_beta)
lrc->beta->push_back(*fi);
PITERATE(TAttributedFloatList, fi_se, temp_beta_se)
lrc->beta_se->push_back(*fi_se);
if (error >= TLogRegFitter::Constant)
return cl;
lrc->wald_Z = computeWaldZ(lrc->beta, lrc->beta_se);
lrc->P = computeP(lrc->wald_Z);
// return classifier with domain, beta and standard errors of beta
return cl;
}
PClassifier TTreeLearner::operator()(PExampleGenerator ogen, const int &weight)
{ if (!ogen)
raiseError("invalid example generator");
PVariable &classVar = ogen->domain->classVar;
if (!classVar)
raiseError("class-less domain");
bool tempSplit = !split;
if (tempSplit)
if (classVar->varType == TValue::INTVAR)
split = defaultDiscreteTreeSplitConstructor;
else if (classVar->varType == TValue::FLOATVAR)
split = defaultContinuousTreeSplitConstructor;
else
raiseError("invalid class type (discrete or continuous expected)");
bool tempStop = !stop;
if (tempStop)
stop = defaultStop;
bool tempSplitter = !exampleSplitter;
if (tempSplitter)
exampleSplitter = mlnew TTreeExampleSplitter_UnknownsAsSelector;
try {
PExampleGenerator examples;
/* If we don't intend to store them, we'll copy them if they're not in a table.
If we must store examples, we'll copy them in any case... */
if (storeExamples)
examples = mlnew TExampleTable(ogen);
else
examples = toExampleTable(ogen);
PDistribution apriorClass = getClassDistribution (examples);
if (apriorClass->abs == 0)
raiseError("no examples");
vector<bool> candidates(examples->domain->attributes->size(), true);
PTreeNode root = call(examples, weight, apriorClass, candidates, 0);
if (storeExamples)
root->examples = examples;
if (tempSplit)
split = PTreeSplitConstructor();
if (tempStop)
stop = PTreeSplitConstructor();
if (tempSplitter)
exampleSplitter = PTreeExampleSplitter();
return mlnew TTreeClassifier(examples->domain, root,
descender ? descender : mlnew TTreeDescender_UnknownMergeAsSelector);
}
catch (exception) {
if (tempSplit)
split = PTreeSplitConstructor();
if (tempStop)
stop = PTreeSplitConstructor();
if (tempSplitter)
exampleSplitter = PTreeExampleSplitter();
throw;
}
}