void LEGACY_NETLIST_READER::LoadNetlist() throw ( IO_ERROR, PARSE_ERROR, boost::bad_pointer )
{
int state = 0;
bool is_comment = false;
COMPONENT* component = NULL;
while( m_lineReader->ReadLine() )
{
char* line = StrPurge( m_lineReader->Line() );
if( is_comment ) // Comments in progress
{
// Test for end of the current comment
if( ( line = strchr( line, '}' ) ) == NULL )
continue;
is_comment = false;
}
if( *line == '{' ) // Start Comment or Pcbnew info section
{
is_comment = true;
if( m_loadFootprintFilters && state == 0
&& (strnicmp( line, "{ Allowed footprints", 20 ) == 0) )
{
loadFootprintFilters();
continue;
}
if( ( line = strchr( line, '}' ) ) == NULL )
continue;
}
if( *line == '(' )
state++;
if( *line == ')' )
state--;
if( state == 2 )
{
component = loadComponent( line );
continue;
}
if( state >= 3 ) // Pad descriptions are read here.
{
wxASSERT( component != NULL );
loadNet( line, component );
state--;
}
}
if( m_footprintReader )
{
m_footprintReader->Load( m_netlist );
}
}
void TokenLoader::loadUnitSideAttributes(UnitToken* token, Json unitTokenParameters) {
Json meleeParameters = unitTokenParameters.getObject("melee"); //TEST na default
Json rangedParameters = unitTokenParameters.getObject("ranged");
loadMelee(token, meleeParameters);
loadRanged(token, rangedParameters);
loadShield(token, unitTokenParameters.getStringArray("armor"));
loadNet(token, unitTokenParameters.getStringArray("net"));
}
void SportsLayer::Vesion(int vesion){
vsion=vesion;
if(vesion==1){
loadbasic();
}else if(vesion==2){
loadNet();
}
}
CNeuralNet::CNeuralNet(char *filename) {
#ifdef FSTREAM_BINARY
std::fstream file_op(filename, std::ios::binary | std::ios::in);
#else
std::fstream file_op(filename,std::ios::in);
#endif
file_op.precision(std::numeric_limits<real>::digits10);
file_op >> m_numOfInputs;
file_op >> m_numOfHiddenNeurons;
file_op >> m_numOfOutputs;
file_op >> m_gameCounter;
file_op >> m_trainingCounter;
m_hiddenNeurons = new CNeuron[m_numOfHiddenNeurons];
if (m_hiddenNeurons) {
for (int i = 0; i < m_numOfHiddenNeurons; ++i)
m_hiddenNeurons[i].setNumOfWeights(m_numOfInputs);
} else {
std::cout << "Uh oh, no memory for " << m_numOfHiddenNeurons
<< " neurons!" << std::endl;
return;
}
m_outNeurons = new CNeuron[m_numOfOutputs];
if (m_outNeurons) {
for (int i = 0; i < m_numOfOutputs; ++i)
m_outNeurons[i].setNumOfWeights(m_numOfHiddenNeurons);
} else {
std::cout << "Uh oh, no memory for " << m_numOfOutputs << " neurons!"
<< std::endl;
return;
}
inp2 = new real[m_numOfHiddenNeurons];
out = new real[m_numOfOutputs];
file_op.close();
loadNet(filename);
}
cv::Ptr<cv::ml::StatModel>
train(const string& protoFile, const string& modelFile,
const vector<cv::Mat>& trainData, const vector<int>& trainLabel,
const string& type, int kFold) {
cv::Ptr<cv::dnn::Net> net = loadNet(protoFile, modelFile);
const cv::dnn::Blob input = cv::dnn::Blob(trainData);
cout << input.shape() << endl;
net->setBlob(".data", input);
net->forward();
// 全結合層 fc7(InnerProduct)の出力を特徴量として抽出
cout << "extract feature" << endl;
const cv::dnn::Blob blob = net->getBlob("fc7");
net.release();
const cv::Mat feature = blob.matRefConst();
cout << feature.size() << endl;
cout << "train model" << endl;
// 線形SVM or ロジスティック回帰で学習
cv::Ptr<cv::ml::TrainData> data =
cv::ml::TrainData::create(feature, cv::ml::ROW_SAMPLE, cv::Mat(trainLabel, false));
if(type == "SVM") {
cv::Ptr<cv::ml::SVM> clf = cv::ml::SVM::create();
clf->setType(cv::ml::SVM::C_SVC);
clf->setKernel(cv::ml::SVM::LINEAR);
clf->trainAuto(data, kFold); // k-fold cross validation
return clf;
} else {
cv::Ptr<cv::ml::LogisticRegression> clf =
cv::ml::LogisticRegression::create();
clf->setLearningRate(0.01);
clf->setIterations(1000);
clf->setRegularization(cv::ml::LogisticRegression::REG_L2);
clf->setTrainMethod(cv::ml::LogisticRegression::MINI_BATCH);
clf->setMiniBatchSize(10);
clf->train(data);
return clf;
}
}
