int EntropyMethod::computeThreshold(const Histogram &histogram)
{
int threshold = 0;
double max_evaluation = computeEvaluation(histogram, 0);
for (int i = 1; i <= MAX_PIXEL_VALUE; ++i) {
double evaluation = computeEvaluation(histogram, i);
if (evaluation > max_evaluation) {
threshold = i;
max_evaluation = evaluation;
}
}
return threshold;
}
/**
* Return the typed value of Term evaluation with respect
* to Bounder Symbol values
*/
inline const T& evaluate(const Bounder& bounder)
{
if (!_isValueComputed) {
_value = computeEvaluation(bounder);
_isValueComputed = true;
}
return _value;
}
void Network::computeTraining(double alpha,
const std::vector<double> & data,
const std::vector<double> & knowledge)
{
assert(_outputNeurons.size() == knowledge.size());
// evaluate network
computeEvaluation(data);
// compute training errors of all neurons
std::set<int> set1;
std::set<int> set2;
std::set<int> * ptrCurrentSet = &set1;
std::set<int> * ptrNextSet = &set2;
std::vector<double> errors(_neuronValues.size(), 0.0);
// compute local training for output neurons
for (unsigned x=0; x<_outputNeurons.size(); x++)
{
int j=_outputNeurons[x];
computeLocalOutputTraining(j, knowledge[x], errors);
ptrCurrentSet->insert(j);
}
// update neurons using multiple passes
// from output neurons to input neurons
while (not ptrCurrentSet->empty())
{
// for each neuron of the set to update
for (int j : *ptrCurrentSet)
{
// update error for j
computeLocalHiddenTraining(j, errors);
// get input neurons for the next updating pass
for (int i=0; i<_synapseConnections.rows(); i++)
{
if (_synapseConnections(i,j) != 0)
{
ptrNextSet->insert(i);
}
}
}
// go to next updating pass
std::swap(ptrCurrentSet, ptrNextSet);
ptrNextSet->clear();
}
// update synapse weights
for (int i=0; i<_synapseWeights.rows(); i++)
{
for (int j=0; j<_synapseWeights.cols(); j++)
{
if (_synapseConnections(i,j) != 0)
{
_synapseWeights(i,j) += alpha * errors[j] * _neuronValues(i);
}
}
}
}
