void ProgValidationNonTilt::run()
{
//Clustering Tendency and Cluster Validity Stephen D. Scott
randomize_random_generator();
MetaData md,mdGallery,mdOut,mdOut2,mdSort;
MDRow row;
FileName fnOut,fnOut2, fnGallery;
fnOut = fnDir+"/clusteringTendency.xmd";
fnGallery = fnDir+"/gallery.doc";
fnOut2 = fnDir+"/validation.xmd";
size_t nSamplesRandom = 500;
md.read(fnParticles);
mdGallery.read(fnGallery);
mdSort.sort(md,MDL_IMAGE_IDX,true,-1,0);
size_t maxNImg;
size_t sz = md.size();
if (useSignificant)
mdSort.getValue(MDL_IMAGE_IDX,maxNImg,sz);
else
{
mdSort.getValue(MDL_ITEM_ID,maxNImg,sz);
}
String expression;
MDRow rowP,row2;
SymList SL;
int symmetry, sym_order;
SL.readSymmetryFile(fnSym.c_str());
SL.isSymmetryGroup(fnSym.c_str(), symmetry, sym_order);
/*
double non_reduntant_area_of_sphere = SL.nonRedundantProjectionSphere(symmetry,sym_order);
double area_of_sphere_no_symmetry = 4.*PI;
double correction = std::sqrt(non_reduntant_area_of_sphere/area_of_sphere_no_symmetry);
*/
double correction = 1;
double validation = 0;
double num_images = 0;
MetaData tempMd;
std::vector<double> sum_u(nSamplesRandom);
double sum_w=0;
std::vector<double> H0(nSamplesRandom);
std::vector<double> H(nSamplesRandom);
std::vector<double> p(nSamplesRandom);
if (rank==0)
init_progress_bar(maxNImg);
for (size_t idx=0; idx<=maxNImg;idx++)
{
if ((idx)%Nprocessors==rank)
{
if (useSignificant)
expression = formatString("imageIndex == %lu",idx);
else
expression = formatString("itemId == %lu",idx);
tempMd.importObjects(md, MDExpression(expression));
if (tempMd.size()==0)
continue;
//compute H_0 from noise
obtainSumU_2(mdGallery, tempMd,sum_u,H0);
//compute H from experimental
obtainSumW(tempMd,sum_w,sum_u,H,correction);
std::sort(H0.begin(),H0.end());
std::sort(H.begin(),H.end());
double P = 0;
for(size_t j=0; j<sum_u.size();j++)
{
//P += H0.at(j)/H.at(j);
P += H0.at(size_t((1-significance_noise)*nSamplesRandom))/H.at(j);
p.at(j) = H0.at(j)/H.at(j);
}
P /= (nSamplesRandom);
if (useSignificant)
rowP.setValue(MDL_IMAGE_IDX,idx);
else
rowP.setValue(MDL_ITEM_ID,idx);
rowP.setValue(MDL_WEIGHT,P);
mdPartial.addRow(rowP);
tempMd.clear();
if (rank==0)
progress_bar(idx+1);
}
}
if (rank==0)
progress_bar(maxNImg);
synchronize();
gatherClusterability();
if (rank == 0)
{
mdPartial.write(fnOut);
std::vector<double> P;
mdPartial.getColumnValues(MDL_WEIGHT,P);
for (size_t idx=0; idx< P.size();idx++)
{
if (P[idx] > 1)
validation += 1.;
num_images += 1.;
}
validation /= (num_images);
row2.setValue(MDL_IMAGE,fnInit);
row2.setValue(MDL_WEIGHT,validation);
mdOut2.addRow(row2);
mdOut2.write(fnOut2);
}
}
void ProgValidationNonTilt::run()
{
//Clustering Tendency and Cluster Validity Stephen D. Scott
randomize_random_generator();
//char buffer[400];
//sprintf(buffer, "xmipp_reconstruct_significant -i %s --initvolumes %s --odir %s --sym %s --iter 1 --alpha0 %f --angularSampling %f",fnIn.c_str(), fnInit.c_str(),fnDir.c_str(),fnSym.c_str(),alpha0,angularSampling);
//system(buffer);
MetaData md,mdOut,mdOut2;
FileName fnMd,fnOut,fnOut2;
fnMd = fnDir+"/angles_iter001_00.xmd";
fnOut = fnDir+"/clusteringTendency.xmd";
fnOut2 = fnDir+"/validation.xmd";
size_t nSamplesRandom = 250;
md.read(fnMd);
size_t maxNImg;
size_t sz = md.size();
md.getValue(MDL_IMAGE_IDX,maxNImg,sz);
String expression;
MDRow rowP,row2;
SymList SL;
int symmetry, sym_order;
SL.readSymmetryFile(fnSym.c_str());
SL.isSymmetryGroup(fnSym.c_str(), symmetry, sym_order);
double non_reduntant_area_of_sphere = SL.nonRedundantProjectionSphere(symmetry,sym_order);
double area_of_sphere_no_symmetry = 4.*PI;
double correction = std::sqrt(non_reduntant_area_of_sphere/area_of_sphere_no_symmetry);
double validation = 0;
MetaData tempMd;
std::vector<double> sum_u(nSamplesRandom);
//std::vector<double> sum_w(nSamplesRandom);
double sum_w=0;
std::vector<double> H0(nSamplesRandom);
std::vector<double> H(nSamplesRandom);
if (rank==0)
init_progress_bar(maxNImg);
for (size_t idx=0; idx<=maxNImg;idx++)
{
if ((idx+1)%Nprocessors==rank)
{
expression = formatString("imageIndex == %lu",idx);
tempMd.importObjects(md, MDExpression(expression));
if (tempMd.size()==0)
continue;
//compute H_0 from noise
obtainSumU(tempMd,sum_u,H0);
//compute H from experimental
obtainSumW(tempMd,sum_w,sum_u,H,correction);
std::sort(H0.begin(),H0.end());
std::sort(H.begin(),H.end());
double P = 0;
for(size_t j=0; j<sum_u.size();j++)
P += H0.at(j)/H.at(j);
P /= (nSamplesRandom);
rowP.setValue(MDL_IMAGE_IDX,idx);
rowP.setValue(MDL_WEIGHT,P);
mdPartial.addRow(rowP);
//sum_u.clear();
//sum_w.clear();
//H0.clear();
//H.clear();
tempMd.clear();
if (rank==0)
progress_bar(idx+1);
}
}
if (rank==0)
progress_bar(maxNImg);
synchronize();
gatherClusterability();
if (rank == 0)
{
mdPartial.write(fnOut);
std::vector<double> P;
mdPartial.getColumnValues(MDL_WEIGHT,P);
for (size_t idx=0; idx< P.size();idx++)
{
if (P[idx] > 1)
validation += 1;
}
validation /= (maxNImg+1);
}
row2.setValue(MDL_IMAGE,fnInit);
row2.setValue(MDL_WEIGHT,validation);
mdOut2.addRow(row2);
mdOut2.write(fnOut2);
}
