Foam::tmp<Foam::scalarField> Foam::TPS::weights
(
const vectorField& controlPoints,
const vector& dataPoint
) const
{
scalarField dist = mag(controlPoints - dataPoint);
scalarField RBF(dist.size());
forAll(RBF, i)
{
if (dist[i] > SMALL)
{
RBF[i] = sqr(dist[i])*log(dist[i]);
}
else
{
RBF[i] = 0.0;
}
}
return RBF;
}
int main(int argc, char* argv[]){
//float beta;
//int row,col;
string file_name = FILE_NAME;
HostMatrix<float> X;
HostMatrix<float> X_test;
HostMatrix<float> Y;
HostMatrix<float> Y_test;
HostMatrix<float> Input;
HostMatrix<float> Target;
std::map<string,int> Classes;
std::map<int,string> ClassesLookup;
readFile(file_name,Input,Target,Classes,ClassesLookup);
int kfold = 1;
int correct_instances = 0;
int incorrect_instances = 0;
int total_instances = 0;
int **confusionMatrix;
confusionMatrix = (int**) malloc(sizeof(int*)*Classes.size());
for(int i = 0; i < (int)Classes.size(); i++){
confusionMatrix[i] = (int*) malloc(sizeof(int)*Classes.size());
memset(confusionMatrix[i],0,sizeof(int)*Classes.size());
}
float Pet_mean = 0;
float Ped_mean = 0;
unsigned int seed = (unsigned)time(0);
/***************RUN INFORMATION*************/
writeHeader(Input,Classes.size(),seed);
/*******************************************/
if(!InitCUDA()) {
return 1;
}
culaStatus status;
status = culaInitialize();
std::cout << "Starting " << std::endl;
float center_time = 0;
float width_time = 0;
float weight_time = 0;
float scaling_time = 0;
unsigned int time_total = 0;
unsigned int testing_time = 0;
unsigned int training_time = 0;
clock_t initialTimeTotal = clock();
do{
X = crossvalidationTrain(Input,KFOLDS,kfold);
X_test = crossvalidationTest(Input,KFOLDS,kfold);
Y = crossvalidationTrain(Target,KFOLDS,kfold);
Y_test = crossvalidationTest(Target,KFOLDS,kfold);
HostMatrix<float> Weights;
HostMatrix<float> Centers;
/*Train Network*/
clock_t initialTime = clock();
RadialBasisFunction RBF(NETWORK_SIZE,RNEIGHBOURS,SCALING_FACTOR,Classes.size());
RBF.SetSeed(seed);
RBF.Train(X,Y);
training_time = (clock() - initialTime);
center_time += RBF.times[0];
width_time += RBF.times[1];
weight_time += RBF.times[2];
scaling_time += RBF.times[3];
/*Test Network*/
initialTime = clock();
std::cout << "Testing" << std::endl;
HostMatrix<float> out_test;
out_test = RBF.Test(X_test);
for(int i = 0; i< X_test.Rows();i++){
float max = 0;
float out_class = 0;
for(int j = 0; j < (int) Classes.size(); j++){
if(out_test(i,j) > max){
out_class = (float)j;
max = out_test(i,j);
}
}
out_test(i,0) = out_class+1;
}
for (int i = 0; i < out_test.Rows(); i++)
{
out_test(i,0) = (float)round(out_test(i,0));
if(out_test(i,0) <= 0) out_test(i,0) = 1;
if(out_test(i,0) > Classes.size()) out_test(i,0) = (float)Classes.size();
std::cout << Y_test(i,0) << " " << out_test(i,0) << std::endl;
}
correct_instances += out_test.Rows() - error_calc(Y_test,out_test);
incorrect_instances += error_calc(Y_test,out_test);
total_instances += out_test.Rows();
/*Add values to Confusion Matrix*/
for(int i = 0; i < Y_test.Rows(); i++){
confusionMatrix[((int)Y_test(i,0))-1][((int)out_test(i,0))-1] = confusionMatrix[((int)Y_test(i,0))-1][((int)out_test(i,0))-1] + 1;
}
testing_time = (clock() - initialTime);
/*Increment fold number, for use in crossvalidation*/
kfold++;
}while(kfold <= KFOLDS);
time_total = (clock() - initialTimeTotal);
/*****************MEASURES****************/
measures(correct_instances,total_instances,incorrect_instances,confusionMatrix,Classes,ClassesLookup);
writeFooter(center_time,width_time,weight_time,scaling_time,training_time,testing_time,time_total);
culaShutdown();
cudaThreadExit();
return 0;
}
