void TrainingSet::setTargetSize(int ts, double fill)
{
int ls = getTargetSize();
resizeVectorSet(targets, ts, fill);
// targetSize = ts;
//Si realmente cambio el tamaño del patron de salidas
if(ls != ts){
emit targetSizeChanged(ts);
emit targetSizeChanged(ls, ts);
}
}
void IdealTarget::onInit()
{
InternalMessage("Display","Entering IdealTarget::onInit") ;
m_target_container = static_cast< ::Ogre::OverlayContainer* >(
::Ogre::OverlayManager::getSingleton().createOverlayElement(
"Panel", Utility::getUniqueName())) ;
getOverlay()->add2D(m_target_container) ;
m_target_container->setPosition(0,0) ;
m_target_container->setWidth(1) ;
m_target_container->setHeight(1) ;
m_target =
::Ogre::OverlayManager::getSingleton().createOverlayElement(
"Panel", Utility::getUniqueName()) ;
// clone material because we will modify it
::Ogre::MaterialPtr material = ::Ogre::MaterialManager::getSingleton().getByName(getIdealTargetMaterial());
material = material->clone(Utility::getUniqueName()) ;
m_target->setMaterialName(material->getName()) ;
m_target->setHorizontalAlignment(::Ogre::GHA_CENTER) ;
m_target->setVerticalAlignment(::Ogre::GVA_CENTER) ;
const float size = getTargetSize() ;
m_target->setLeft(-size/2) ;
m_target->setTop(-size/2) ;
m_target->setDimensions(size,size) ;
m_target_container->_addChild(m_target) ;
m_target_container->hide() ;
getOverlay()->show() ;
onUpdate() ;
InternalMessage("Display","Leaving IdealTarget::onInit") ;
}
void TrainingSet::normalize(NormalizationType no)
{
size_t sPatterns = getPatternCount();
size_t sInputs = getInputSize();
size_t sTargets = getTargetSize();
double mean;
double stddv;
double nor;
vector<vector<double> >
li = inputs,
lt = targets;
for(size_t p = 0; p < sPatterns; p++){
mean = ANNFrameworkFunctions::getMean(inputs[p]);
stddv = ANNFrameworkFunctions::getStandardDeviation(inputs[p]);
for(size_t i = 0; i < sInputs; i++){
nor = (inputs[p][i] - mean)/stddv;
switch(no){
case ITrainingSet::BipolarAutoThreshold:
break;
case ITrainingSet::UnipolarAutoThreshold:
break;
case ITrainingSet::Nothing:
break;
case ITrainingSet::BipolarFixedThreshold:
break;
case ITrainingSet::UnipolarFixedThreshold:
break;
case ITrainingSet::LinearFixedRange:
break;
case ITrainingSet::LinearAutoRange:
break;
case ITrainingSet::Sigmoid:
//NOTE: no estoy seguro de que esto sea asi
inputs[p][i] = 1/(1 + exp(-nor));
break;
case ITrainingSet::Tanh:
inputs[p][i] = (1 - exp(-nor))/(1 + exp(-nor));
break;
case ITrainingSet::MeanDistance:
break;
}
}
mean = ANNFrameworkFunctions::getMean(targets[p]);
stddv = ANNFrameworkFunctions::getStandardDeviation(targets[p]);
for(size_t t = 0; t < sTargets; t++){
nor = (targets[p][t] - mean)/stddv;
switch(no){
case ITrainingSet::BipolarAutoThreshold:
break;
case ITrainingSet::UnipolarAutoThreshold:
break;
case ITrainingSet::Nothing:
break;
case ITrainingSet::BipolarFixedThreshold:
break;
case ITrainingSet::UnipolarFixedThreshold:
break;
case ITrainingSet::LinearFixedRange:
break;
case ITrainingSet::LinearAutoRange:
break;
case ITrainingSet::Sigmoid:
//NOTE: no estoy seguro de que esto sea asi
targets[p][t] = 1/(1 + exp(-nor));
break;
case ITrainingSet::Tanh:
targets[p][t] = (1 - exp(-nor))/(1 + exp(-nor));
break;
case ITrainingSet::MeanDistance:
break;
}
}
}
inputsNorType = no;
targetsNorType = no;
if(li != inputs){
emit inputsChanged(inputs);
emit inputsChanged(li, inputs);
}
if(lt != targets){
emit targetsChanged(targets);
emit targetsChanged(lt, targets);
}
}