#include <mpi_classify_CL2D.h>

Inheritance diagram for ProgClassifyCL2D:

Collaboration diagram for ProgClassifyCL2D:

Public Member Functions
	ProgClassifyCL2D (int argc, char **argv)
	MPI constructor. More...

	~ProgClassifyCL2D ()
	Destructor. More...

	ProgClassifyCL2D (const ProgClassifyCL2D &)=delete

	ProgClassifyCL2D (const ProgClassifyCL2D &&)=delete

ProgClassifyCL2D &	operator= (const ProgClassifyCL2D &)=delete

ProgClassifyCL2D &	operator= (const ProgClassifyCL2D &&)=delete

void	readParams ()
	Read. More...

void	show () const
	Show. More...

void	defineParams ()
	Usage. More...

void	produceSideInfo ()
	Produce side info. More...

void	run ()
	Run. More...

Public Member Functions inherited from XmippProgram
const char *	getParam (const char *param, int arg=0)

const char *	getParam (const char param, const char subparam, int arg=0)

int	getIntParam (const char *param, int arg=0)

int	getIntParam (const char param, const char subparam, int arg=0)

double	getDoubleParam (const char *param, int arg=0)

double	getDoubleParam (const char param, const char subparam, int arg=0)

float	getFloatParam (const char *param, int arg=0)

float	getFloatParam (const char param, const char subparam, int arg=0)

void	getListParam (const char *param, StringVector &list)

int	getCountParam (const char *param)

bool	checkParam (const char *param)

bool	existsParam (const char *param)

void	addParamsLine (const String &line)

void	addParamsLine (const char *line)

ParamDef *	getParamDef (const char *param) const

virtual void	quit (int exit_code=0) const

virtual int	tryRun ()

void	initProgress (size_t total, size_t stepBin=60)

void	setProgress (size_t value=0)

void	endProgress ()

void	processDefaultComment (const char param, const char left)

void	setDefaultComment (const char param, const char comment)

virtual void	initComments ()

void	setProgramName (const char *name)

void	addUsageLine (const char *line, bool verbatim=false)

void	clearUsage ()

void	addExampleLine (const char *example, bool verbatim=true)

void	addSeeAlsoLine (const char *seeAlso)

void	addKeywords (const char *keywords)

const char *	name () const

virtual void	usage (int verb=0) const

virtual void	usage (const String &param, int verb=2)

int	version () const

virtual void	read (int argc, const char **argv, bool reportErrors=true)

virtual void	read (int argc, char **argv, bool reportErrors=true)

void	read (const String &argumentsLine)

	XmippProgram ()

	XmippProgram (int argc, const char **argv)

virtual	~XmippProgram ()

Public Attributes
FileName	fnSel
	Input selfile with the images to quantify. More...

FileName	fnCodes0
	Input selfile with initial codes. More...

FileName	fnOut
	Output rootname. More...

FileName	fnODir
	Output directory. More...

int	Niter
	Number of iterations. More...

int	Ncodes0
	Initial number of code vectors. More...

int	Ncodes
	Final number of code vectors. More...

int	Nneighbours
	Number of neighbours. More...

double	PminSize
	Minimum size of a node. More...

bool	useCorrelation
	Use Correlation instead of Correntropy. More...

bool	classicalMultiref
	Classical Multiref. More...

bool	classifyAllImages
	Clasify all images. More...

bool	classicalSplit
	Use ClassicalCriterion at split. More...

int	NSplitTrials
	MaxTrials to split. More...

double	maxShift
	Maximum shift. More...

bool	normalizeImages
	Normalize input images. More...

bool	mirrorImages
	Mirror. More...

bool	useThresholdMask
	Use threshold mask. More...

double	threshold
	Threshold to use. More...

bool	alignImages
	Don't align images. More...

MetaDataDb	SF

std::vector< size_t >	objId

CL2D	vq

MpiNode *	node

double	maxShift2

GaussianInterpolator	gaussianInterpolator

size_t	Ydim

size_t	Xdim

MultidimArray< int >	mask
	Mask for the background. More...

double	sigma
	Noise in the images. More...

Public Attributes inherited from XmippProgram
bool	doRun

bool	runWithoutArgs

int	verbose
	Verbosity level. More...

int	debug

Additional Inherited Members
Protected Member Functions inherited from XmippProgram
void	defineCommons ()

Protected Attributes inherited from XmippProgram
int	errorCode

ProgramDef *	progDef
	Program definition and arguments parser. More...

std::map< String, CommentList >	defaultComments

int	argc
	Original command line arguments. More...

const char **	argv

Detailed Description

CL2D parameters.

Definition at line 224 of file mpi_classify_CL2D.h.

Constructor & Destructor Documentation

◆ ProgClassifyCL2D() [1/3]

ProgClassifyCL2D::ProgClassifyCL2D	(	int	argc,
		char **	argv
	)

MPI constructor.

Definition at line 1647 of file mpi_classify_CL2D.cpp.

 {
     node = new MpiNode(argc, argv);
     if (!node->isMaster())
         verbose = 0;
 }

◆ ~ProgClassifyCL2D()

ProgClassifyCL2D::~ProgClassifyCL2D ( )

Destructor.

Definition at line 1655 of file mpi_classify_CL2D.cpp.

 {
     delete node;
 }

◆ ProgClassifyCL2D() [2/3]

ProgClassifyCL2D::ProgClassifyCL2D ( const ProgClassifyCL2D & )

delete

◆ ProgClassifyCL2D() [3/3]

ProgClassifyCL2D::ProgClassifyCL2D ( const ProgClassifyCL2D && )

delete

Member Function Documentation

◆ defineParams()

void ProgClassifyCL2D::defineParams ( )

virtual

Usage.

Reimplemented from XmippProgram.

Definition at line 1716 of file mpi_classify_CL2D.cpp.

 {
     addUsageLine("Divide a selfile into the desired number of classes. ");
     addUsageLine("+Vector quantization with correntropy and a probabilistic criterion is used for creating the subdivisions.");
     addUsageLine("+Correlation and the standard maximum correlation criterion can also be used and normally produce good results.");
     addUsageLine("+Correntropy and the probabilistic clustering criterion are recommended for images with very low SNR or cases in which the correlation have clear difficulties to converge.");
     addUsageLine("+");
     addUsageLine("+The algorithm is fully described in [[http://www.ncbi.nlm.nih.gov/pubmed/20362059][this article]].");
     addUsageLine("+");
     addUsageLine("+An interesting convergence criterion is the number of images changing classes between iterations. If a low percentage of the image change class, then the clustering is rather stable and clear.");
     addUsageLine("+If many images change class, it is likely that there is not enough SNR to determine so many classes. It is recommended to reduce the number of classes");
     addSeeAlsoLine("mpi_image_sort");
     addParamsLine("    -i <selfile>             : Selfile with the input images");
     addParamsLine("   [--odir <dir=\".\">]      : Output directory");
     addParamsLine("   [--oroot <root=class>]    : Output rootname, by default, class");
     addParamsLine("   [--iter <N=20>]           : Number of iterations");
     addParamsLine("   [--nref0 <N=2>]           : Initial number of code vectors");
     addParamsLine("or  --ref0 <selfile=\"\">    : Selfile with initial code vectors");
     addParamsLine("   [--nref <N=16>]           : Final number of code vectors");
     addParamsLine("   [--neigh+ <N=4>]          : Number of neighbour code vectors");
     addParamsLine("                             : Set -1 for all");
     addParamsLine("   [--minsize+ <N=20>]       : Percentage minimum node size");
     addParamsLine("                             : Let's say that we have 1000 images and we are currently estimating 2 classes.");
     addParamsLine("                             : On average each class should have about 500 images. ");
     addParamsLine("                             : If one of the classes drops below 20% (by default) of 500,");
     addParamsLine("                             : then that class is removed and the remaining class is splitted in two. ");
     addParamsLine("                             : The same reasoning is applied when the number of classes is 4, but 20% ");
     addParamsLine("                             : is calculated now on 250 (=1000/4).");
     addParamsLine("   [--distance <type=correntropy>]       : Distance type");
     addParamsLine("            where <type>");
     addParamsLine("                       correntropy correlation: See CL2D paper for the definition of correntropy");
     addParamsLine("   [--classicalMultiref]     : Instead of enhanced clustering");
     addParamsLine("   [--classicalSplit]        : Instead of enhanced clustering at the split iterations");
     addParamsLine("   [--maxSplitTrials <n=5>]  : Maximum number of trials to split before giving up");
     addParamsLine("   [--maxShift <d=10>]       : Maximum allowed shift");
     addParamsLine("   [--classifyAllImages]     : By default, some images may not be classified. Use this option to classify them all.");
     addParamsLine("   [--dontNormalizeImages]   : By default, input images are normalized to have 0 mean and standard deviation 1");
     addParamsLine("   [--dontMirrorImages]      : By default, input images are studied unmirrored and mirrored");
     addParamsLine("   [--useThresholdMask <t>]  : Use a mask to compare images. Remove pixels whose value is smaller or equal t");
     addParamsLine("   [--dontAlign]             : Do not center the class representatives");
     addExampleLine("mpirun -np 3 `which xmipp_mpi_classify_CL2D` -i images.stk --nref 256 --oroot class --odir CL2Dresults --iter 10");
 }

◆ operator=() [1/2]

ProgClassifyCL2D& ProgClassifyCL2D::operator= ( const ProgClassifyCL2D & )

delete

◆ operator=() [2/2]

ProgClassifyCL2D& ProgClassifyCL2D::operator= ( const ProgClassifyCL2D && )

delete

◆ produceSideInfo()

void ProgClassifyCL2D::produceSideInfo ( )

Produce side info.

Definition at line 1759 of file mpi_classify_CL2D.cpp.

 {
     if (!useCorrelation)
         normalizeImages=false;
 
     maxShift2 = maxShift * maxShift;
 
     gaussianInterpolator.initialize(6, 60000, false);
 
     // Get image dimensions
     SF.read(fnSel);
     SF.removeDisabled();
 
     size_t Zdim, Ndim;
     getImageSize(SF, Xdim, Ydim, Zdim, Ndim);
 
     // Prepare the Task distributor
     SF.findObjects(objId);
     // size_t Nimgs = objId.size();
 
     // Prepare mask for evaluating the noise outside
     mask.resize(prm->Ydim, prm->Xdim);
     mask.setXmippOrigin();
     BinaryCircularMask(mask, prm->Xdim / 2, INNER_MASK);
 
     // Read input code vectors if available
     std::vector<MultidimArray<double> > codes0;
     if (fnCodes0 != "")
     {
         Image<double> I;
         MetaDataVec SFCodes(fnCodes0);
 
         size_t Xdim0, Ydim0, Zdim0, Ndim0;
         getImageSize(SFCodes, Xdim0, Ydim0, Zdim0, Ndim0);
         if (Xdim0!=Xdim || Ydim0!=Ydim)
             REPORT_ERROR(ERR_MULTIDIM_SIZE,"Input reference and images are not of the same size");
 
         for (size_t objId : SFCodes.ids())
         {
             I.readApplyGeo(SFCodes, objId);
             I().setXmippOrigin();
             codes0.push_back(I());
         }
         Ncodes0 = codes0.size();
     }
     vq.initialize(SF, codes0);
 }

◆ readParams()

void ProgClassifyCL2D::readParams ( )

virtual

Read.

Reimplemented from XmippProgram.

Definition at line 1661 of file mpi_classify_CL2D.cpp.

 {
     fnSel = getParam("-i");
     fnOut = getParam("--oroot");
     fnODir = getParam("--odir");
     fnCodes0 = getParam("--ref0");
     Niter = getIntParam("--iter");
     Nneighbours = getIntParam("--neigh");
     Ncodes0 = getIntParam("--nref0");
     Ncodes = getIntParam("--nref");
     PminSize = getDoubleParam("--minsize");
     String aux;
     aux = getParam("--distance");
     useCorrelation = aux == "correlation";
     classicalMultiref = checkParam("--classicalMultiref");
     classicalSplit = checkParam("--classicalSplit");
     NSplitTrials = getIntParam("--maxSplitTrials");
     maxShift = getDoubleParam("--maxShift");
     classifyAllImages = checkParam("--classifyAllImages");
     normalizeImages = !checkParam("--dontNormalizeImages");
     mirrorImages = !checkParam("--dontMirrorImages");
     useThresholdMask = checkParam("--useThresholdMask");
     if (useThresholdMask)
         threshold=getDoubleParam("--useThresholdMask");
     alignImages = !checkParam("--dontAlign");
 
     prm = this; // FIXME HACK because of the global variable. Solve it properly
 }

◆ run()

void ProgClassifyCL2D::run ( )

virtual

Run.

Reimplemented from XmippProgram.

Definition at line 1807 of file mpi_classify_CL2D.cpp.

 {
     CREATE_LOG();
     show();
     produceSideInfo();
 
     // Run all iterations
     int level = 0;
     int Q = vq.P.size();
     bool originalClassicalMultiref=classicalMultiref;
     if (Q==1)
         classicalMultiref=true;
     vq.run(fnODir, fnOut, level);
 
     while (Q < Ncodes)
     {
         if (node->rank == 1)
             std::cout << "Spliting nodes ...\n";
 
         int Nclean = vq.cleanEmptyNodes();
         int Nsplits = XMIPP_MIN(Q,Ncodes-Q) + Nclean;
 
         for (int i = 0; i < Nsplits; i++)
         {
             vq.splitFirstNode();
             if (node->rank == 1)
                 std::cout << "Currently there are " << vq.P.size() << " nodes"
                 << std::endl;
         }
 
         Q = vq.P.size();
         level++;
         classicalMultiref=originalClassicalMultiref || Q==1;
         vq.run(fnODir, fnOut, level);
     }
     if (node->rank == 1)
     {
         std::sort(vq.P.begin(), vq.P.end(), SDescendingClusterSort());
         Q = vq.P.size();
         MetaDataDb SFq, SFclassified, SFaux, SFaux2;
         for (int q = 0; q < Q; q++)
         {
             SFq.read(formatString("class%06d_images@%s/level_%02d/%s_classes.xmd",q+1,fnODir.c_str(),level,fnOut.c_str()));
             SFq.fillConstant(MDL_REF, integerToString(q + 1));
             SFq.fillConstant(MDL_ENABLED, "1");
             SFclassified.unionAll(SFq);
         }
         SFaux = SF;
         SFaux.subtraction(SFclassified, MDL_IMAGE);
         SFaux.fillConstant(MDL_ENABLED, "-1");
         SFaux2.join1(SFclassified, SF, MDL_IMAGE, LEFT);
         SFclassified.clear();
         SFaux2.unionAll(SFaux);
         SFaux.clear();
         SFaux.sort(SFaux2, MDL_IMAGE);
         SFaux.write(fnODir+"/images.xmd");
     }
     CLOSE_LOG();
 }

◆ show()

void ProgClassifyCL2D::show ( ) const

virtual

Show.

Reimplemented from XmippProgram.

Definition at line 1690 of file mpi_classify_CL2D.cpp.

                                   {
     if (!verbose)
         return;
     std::cout << "Input images:            " << fnSel << std::endl
             << "Output root:             " << fnOut << std::endl
             << "Output dir:              " << fnODir << std::endl
             << "Iterations:              " << Niter << std::endl
             << "CodesSel0:               " << fnCodes0 << std::endl
             << "Codes0:                  " << Ncodes0 << std::endl
             << "Codes:                   " << Ncodes << std::endl
             << "Neighbours:              " << Nneighbours << std::endl
             << "Minimum node size:       " << PminSize << std::endl
             << "Use Correlation:         " << useCorrelation << std::endl
             << "Classical Multiref:      " << classicalMultiref << std::endl
             << "Classical Split:         " << classicalSplit << std::endl
                         << "Max. Split trials:       " << NSplitTrials << std::endl
             << "Maximum shift:           " << maxShift << std::endl
             << "Classify all images:     " << classifyAllImages << std::endl
             << "Normalize images:        " << normalizeImages << std::endl
             << "Mirror images:           " << mirrorImages << std::endl
             << "Align images:            " << alignImages << std::endl
     ;
     if (useThresholdMask)
         std::cout << "Threshold mask:          " << threshold << std::endl;
 }

Member Data Documentation

◆ alignImages

bool ProgClassifyCL2D::alignImages

Don't align images.

Definition at line 284 of file mpi_classify_CL2D.h.

◆ classicalMultiref

bool ProgClassifyCL2D::classicalMultiref

Classical Multiref.

Definition at line 257 of file mpi_classify_CL2D.h.

◆ classicalSplit

bool ProgClassifyCL2D::classicalSplit

Use ClassicalCriterion at split.

Definition at line 263 of file mpi_classify_CL2D.h.

◆ classifyAllImages

bool ProgClassifyCL2D::classifyAllImages

Clasify all images.

Definition at line 260 of file mpi_classify_CL2D.h.

◆ fnCodes0

FileName ProgClassifyCL2D::fnCodes0

Input selfile with initial codes.

Definition at line 230 of file mpi_classify_CL2D.h.

◆ fnODir

FileName ProgClassifyCL2D::fnODir

Output directory.

Definition at line 236 of file mpi_classify_CL2D.h.

◆ fnOut

FileName ProgClassifyCL2D::fnOut

Output rootname.

Definition at line 233 of file mpi_classify_CL2D.h.

◆ fnSel

FileName ProgClassifyCL2D::fnSel

Input selfile with the images to quantify.

Definition at line 227 of file mpi_classify_CL2D.h.

◆ gaussianInterpolator

GaussianInterpolator ProgClassifyCL2D::gaussianInterpolator

Definition at line 329 of file mpi_classify_CL2D.h.

◆ mask

MultidimArray<int> ProgClassifyCL2D::mask

Mask for the background.

Definition at line 335 of file mpi_classify_CL2D.h.

◆ maxShift

double ProgClassifyCL2D::maxShift

Maximum shift.

Definition at line 269 of file mpi_classify_CL2D.h.

◆ maxShift2

double ProgClassifyCL2D::maxShift2

Definition at line 326 of file mpi_classify_CL2D.h.

◆ mirrorImages

bool ProgClassifyCL2D::mirrorImages

Mirror.

Definition at line 275 of file mpi_classify_CL2D.h.

◆ Ncodes

int ProgClassifyCL2D::Ncodes

Final number of code vectors.

Definition at line 245 of file mpi_classify_CL2D.h.

◆ Ncodes0

int ProgClassifyCL2D::Ncodes0

Initial number of code vectors.

Definition at line 242 of file mpi_classify_CL2D.h.

◆ Niter

int ProgClassifyCL2D::Niter

Number of iterations.

Definition at line 239 of file mpi_classify_CL2D.h.

◆ Nneighbours

int ProgClassifyCL2D::Nneighbours

Number of neighbours.

Definition at line 248 of file mpi_classify_CL2D.h.

◆ node

MpiNode* ProgClassifyCL2D::node

Definition at line 323 of file mpi_classify_CL2D.h.

◆ normalizeImages

bool ProgClassifyCL2D::normalizeImages

Normalize input images.

Definition at line 272 of file mpi_classify_CL2D.h.

◆ NSplitTrials

int ProgClassifyCL2D::NSplitTrials

MaxTrials to split.

Definition at line 266 of file mpi_classify_CL2D.h.

◆ objId

std::vector<size_t> ProgClassifyCL2D::objId

Definition at line 317 of file mpi_classify_CL2D.h.

◆ PminSize

double ProgClassifyCL2D::PminSize

Minimum size of a node.

Definition at line 251 of file mpi_classify_CL2D.h.

◆ SF

MetaDataDb ProgClassifyCL2D::SF

Definition at line 314 of file mpi_classify_CL2D.h.

◆ sigma

double ProgClassifyCL2D::sigma

Noise in the images.

Definition at line 338 of file mpi_classify_CL2D.h.

◆ threshold

double ProgClassifyCL2D::threshold

Threshold to use.

Definition at line 281 of file mpi_classify_CL2D.h.

◆ useCorrelation

bool ProgClassifyCL2D::useCorrelation

Use Correlation instead of Correntropy.

Definition at line 254 of file mpi_classify_CL2D.h.

◆ useThresholdMask

bool ProgClassifyCL2D::useThresholdMask

Use threshold mask.

Definition at line 278 of file mpi_classify_CL2D.h.

◆ vq

CL2D ProgClassifyCL2D::vq

Definition at line 320 of file mpi_classify_CL2D.h.

◆ Xdim

size_t ProgClassifyCL2D::Xdim

Definition at line 332 of file mpi_classify_CL2D.h.

◆ Ydim

size_t ProgClassifyCL2D::Ydim

Definition at line 332 of file mpi_classify_CL2D.h.

The documentation for this class was generated from the following files:

xmipp/libraries/parallel/mpi_classify_CL2D.h
xmipp/libraries/parallel/mpi_classify_CL2D.cpp

Public Member Functions

Public Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ ProgClassifyCL2D() [1/3]

◆ ~ProgClassifyCL2D()

◆ ProgClassifyCL2D() [2/3]

◆ ProgClassifyCL2D() [3/3]

Member Function Documentation

◆ defineParams()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ produceSideInfo()

◆ readParams()

◆ run()

◆ show()

Member Data Documentation

◆ alignImages

◆ classicalMultiref

◆ classicalSplit

◆ classifyAllImages

◆ fnCodes0

◆ fnODir

◆ fnOut

◆ fnSel

◆ gaussianInterpolator

◆ mask

◆ maxShift

◆ maxShift2

◆ mirrorImages

◆ Ncodes

◆ Ncodes0

◆ Niter

◆ Nneighbours

◆ node

◆ normalizeImages

◆ NSplitTrials

◆ objId

◆ PminSize

◆ SF

◆ sigma

◆ threshold

◆ useCorrelation

◆ useThresholdMask

◆ vq

◆ Xdim

◆ Ydim