float TEFMDataDescription::getExampleMatch(const TExample &ex1, const TExample &ex2)
{
if ((ex1.domain != domain) && (ex2.domain != domain))
raiseError("example's domain doesn't match the data descriptor's");
float weight=1.0;
TExample::iterator e1i(ex1.begin()), e2i(ex2.end());
if (domainDistributions) {
if (matchProbabilities.size() != domainDistributions->size())
matchProbabilities = vector<float>(domainDistributions->size(), -1);
vector<float>::iterator mi(matchProbabilities.begin());
TDomainDistributions::const_iterator di(domainDistributions->begin()), de(domainDistributions->end());
for(; di!=de; e1i++, e2i++, di++, mi++) {
if ((*e1i).varType == TValue::INTVAR) {
if ((*e1i).isDK()) {
if ((*e2i).isDK()) {
if (*mi == -1) {
float mp = 0.0;
ITERATE(TDiscDistribution, ddi, ((TDiscDistribution &)((*di).getReference())))
mp += *ddi * *ddi;
*mi = mp;
}
weight *= *mi;
}
else if (!(*e2i).isSpecial())
weight *= (*di)->p(*e2i);
}
else if ((*e2i).isDK() && !(*e1i).isSpecial())
weight *= (*di)->p(*e1i);
}
}
}
else {
TVarList::const_iterator vi(domain->attributes->begin()), vie(domain->attributes->end());
for(; vi!=vie; e1i++, e2i++, vi++)
if (((*e1i).varType == TValue::INTVAR) && ((*e1i).isDK() && !(*e2i).isSpecial() || (*e2i).isDK() && !(*e1i).isSpecial()))
weight /= (*vi)->noOfValues();
}
return weight;
}
float TEFMDataDescription::getExampleWeight(const TExample &example) const
{
if (example.domain != domain)
raiseError("example's domain doesn't match the data descriptor's");
float weight=1.0;
TVarList::const_iterator vi(domain->attributes->begin()), vie(domain->attributes->end());
TExample::iterator ei(example.begin());
for(; vi!=vie; ei++, vi++)
if ((*ei).isDK() && ((*ei).varType == TValue::INTVAR))
weight /= (*vi)->noOfValues();
return weight;
}
bool convertFromPythonExisting(PyObject *lst, TExample &example)
{
PDomain dom=example.domain;
if (PyOrExample_Check(lst)) {
const TExample &orex = PyExample_AS_ExampleReference(lst);
if (orex.domain != dom)
dom->convert(example, orex);
else
example = orex;
return true;
}
if (!PyList_Check(lst)) {
PyErr_Format(PyExc_TypeError, "invalid argument type (expected list, got '%s)", lst ? lst->ob_type->tp_name : "None");
return false;
}
int const nvars = dom->variables->size() + dom->classVars->size();
if (Py_ssize_t(nvars) != PyList_Size(lst)) {
PyErr_Format(PyExc_IndexError, "invalid list size (got %i, expected %i items)",
PyList_Size(lst), nvars);
return false;
}
Py_ssize_t pos = 0;
TExample::iterator ei(example.begin());
TVarList::iterator vi(dom->variables->begin());
TVarList::const_iterator const ve(dom->variables->end());
TVarList::const_iterator const ce(dom->classVars->end());
while(vi != ce && vi != ve) {
PyObject *li=PyList_GetItem(lst, pos++);
if (!li)
PYERROR(PyExc_SystemError, "can't read the list", false);
if (PyOrValue_Check(li))
if (PyValue_AS_Variable(li) ? (PyValue_AS_Variable(li) != *vi) : (PyValue_AS_Value(li).varType=!(*vi)->varType) ) {
PyErr_Format(PyExc_TypeError, "wrong value type for attribute no. %i (%s)", pos, (*vi)->get_name().c_str());
return false;
}
else
*(ei++)=PyValue_AS_Value(li);
else {
if (li == Py_None) {
*(ei++) = (*vi)->DK();
}
else if (PyString_Check(li))
(*vi)->str2val(string(PyString_AsString(li)), *(ei++));
else if ((*vi)->varType==TValue::INTVAR) {
if (PyInt_Check(li))
*(ei++)=TValue(int(PyInt_AsLong(li)));
else {
PyErr_Format(PyExc_TypeError, "attribute no. %i (%s) is ordinal, string value expected", pos, (*vi)->get_name().c_str());
return false;
}
}
else if ((*vi)->varType==TValue::FLOATVAR) {
float f;
if (PyNumber_ToFloat(li, f))
*(ei++) = TValue(f);
else {
PyErr_Format(PyExc_TypeError, "attribute no. %i (%s) is continuous, float value expected", pos, (*vi)->get_name().c_str());
return false;
}
}
else
ei++;
}
if (++vi == ve) {
vi = dom->classVars->begin();
}
}
return true;
}
PDistribution TLogRegClassifier::classDistribution(const TExample &origexam)
{
checkProperty(domain);
TExample cexample(domain, origexam);
TExample *example2;
if (imputer)
example2 = imputer->call(cexample);
else {
if (dataDescription)
for(TExample::const_iterator ei(cexample.begin()), ee(cexample.end()-1); ei!=ee; ei++)
if ((*ei).isSpecial())
return TClassifier::classDistribution(cexample, dataDescription);
example2 = &cexample;
}
TExample *example = continuizedDomain ? mlnew TExample(continuizedDomain, *example2) : example2;
float prob1;
try {
// multiply example with beta
TAttributedFloatList::const_iterator b(beta->begin()), be(beta->end());
// get beta 0
prob1 = *b;
b++;
// multiply beta with example
TVarList::const_iterator vi(example->domain->attributes->begin());
TExample::const_iterator ei(example->begin()), ee(example->end());
for (; (b!=be) && (ei!=ee); ei++, b++, vi++) {
if ((*ei).isSpecial())
raiseError("unknown value in attribute '%s'", (*vi)->get_name().c_str());
prob1 += (*ei).floatV * (*b);
}
prob1 = exp(prob1)/(1+exp(prob1));
}
catch (...) {
if (imputer)
mldelete example2;
if (continuizedDomain)
mldelete example;
throw;
}
if (imputer)
mldelete example2;
if (continuizedDomain)
mldelete example;
if (classVar->varType == TValue::INTVAR) {
TDiscDistribution *dist = mlnew TDiscDistribution(classVar);
PDistribution res = dist;
dist->addint(0, 1-prob1);
dist->addint(1, prob1);
return res;
}
else {
TContDistribution *dist = mlnew TContDistribution(classVar);
PDistribution res = dist;
dist->addfloat(prob1, 1.0);
return res;
}
}
bool TTabDelimExampleGenerator::readExample(TFileExampleIteratorData &fei, TExample &exam)
{
vector<string> atoms;
// read lines until eof or a non-empty line
while(!feof(fei.file) && ((readTabAtom(fei, atoms, true, csv)>0) || atomsEmpty(atoms))) {
vector<string>::iterator ii(atoms.begin()), ie(atoms.end());
while ((ii!=ie) && !(*ii).length())
ii++;
if (ii==ie)
atoms.clear();
else
break;
}
if (!atoms.size())
return false;
// Add an appropriate number of empty atoms, if needed
while (atoms.size()<attributeTypes->size())
atoms.push_back(string(""));
_ASSERT(exam.domain==domain);
exam.removeMetas();
TExample::iterator ei(exam.begin());
TVarList::iterator vi(domain->attributes->begin());
vector<string>::iterator ai(atoms.begin());
TIntList::iterator si(attributeTypes->begin()), se(attributeTypes->end());
TIntList::iterator cb, cp, ce;
if (classPoses) {
cb = cp = classPoses->begin();
ce = classPoses->end();
}
int pos=0;
for (; (si!=se); pos++, si++, ai++) {
if (*si) { // if attribute is not to be skipped and is not a basket
string valstr;
// Check for don't care
valstr = *ai;
if (!valstr.length() || (valstr == "NA") || (valstr == ".") || (DC && (valstr == DC)))
valstr = "?";
else if ((valstr == "*") || (DK && (valstr == DK)))
valstr = "~";
try {
if (*si==-1)
if (pos==classPos) { // if this is class value
TValue cval;
domain->classVar->filestr2val(valstr, cval, exam);
exam.setClass(cval);
}
else if (classPoses && (cp != ce) && (pos == *cp)) {
const int ind = cp - cb;
domain->classVars->at(ind)->filestr2val(valstr, exam.values_end[ind], exam);
cp++;
}
else { // if this is a normal value
(*vi++)->filestr2val(valstr, *ei++, exam);
}
else { // if this is a meta value
TMetaDescriptor *md = domain->metas[*si];
_ASSERT(md!=NULL);
TValue mval;
md->variable->filestr2val(valstr, mval, exam);
exam.setMeta(*si, mval);
}
}
catch (mlexception &err) {
raiseError("file '%s', line '%i': %s", fei.filename.c_str(), fei.line, err.what());
}
}
// the attribute is marked to be skipped, but may also be a basket
else {
if (pos == basketPos) {
TSplits splits;
split(*ai, splits);
ITERATE(TSplits, si, splits)
basketFeeder->addItem(exam, string(si->first, si->second), fei.line);
}
}
}
if (pos==classPos) // if class is the last value in the line, it is set here
domain->classVar->filestr2val(ai==atoms.end() ? "?" : *(ai++), exam[domain->variables->size()-1], exam);
/* I'm not sure that this is needed; this code is a mess but I don't wish to
waste time studying it since we are moving to 3.0 */
else if (classPoses && (cp != ce) && (pos == *cp)) {
const int ind = cp - cb;
domain->classVars->at(ind)->filestr2val(ai==atoms.end() ? "?" : *(ai++), exam.values_end[ind], exam);
}
while ((ai!=atoms.end()) && !(*ai).length()) ai++; // line must be empty from now on
if (ai!=atoms.end()) {
vector<string>::iterator ii=atoms.begin();
string s=*ii;
while(++ii!=atoms.end()) s+=" "+*ii;
raiseError("example of invalid length (%s)", s.c_str());
}
return true;
}
