KNN Class Reference

#include <knn_classifier.h>

Public Types
enum	distType { EUCLIDEAN = 0, CITYBLOCK = 1 }
	Type of distance. More...

Public Member Functions
	KNN (int k)
void	train (MultidimArray< double > &dataset, MultidimArray< double > &dataLabel, MultidimArray< double > &labelset)
int	predict (MultidimArray< double > &sample, double &score)
void	KNearestNeighbors (MultidimArray< double > &sample)
	Compute the K nearest neighbors to the sample. More...
void	setK (int k)
	Method for setting the K. More...

Detailed Description

This class implements KNN (K Nearest Neighbors). It does the classification by a voting of nearest neighbors of the point. It uses the exhaustive search in order to find the nearest neighbors.

Definition at line 41 of file knn_classifier.h.

Member Enumeration Documentation

distType

enum KNN::distType

Type of distance.

Enumerator
EUCLIDEAN
CITYBLOCK

Definition at line 45 of file knn_classifier.h.

{ EUCLIDEAN = 0, CITYBLOCK = 1} distType;

Constructor & Destructor Documentation

KNN()

KNN::KNN

(

int

k

)

Constructs the algorithm Parameter: k The number of nearest neighbors

Definition at line 28 of file knn_classifier.cpp.

{
    setK(k);
    neighborsIndex.resize(1,1,1,K);
}

Member Function Documentation

KNearestNeighbors()

void KNN::KNearestNeighbors

(

MultidimArray< double > &

sample

)

Compute the K nearest neighbors to the sample.

Definition at line 42 of file knn_classifier.cpp.

{
    double maximum;
    double distance;
    int maximumIndex;
    maxDist.resize(1,1,1,K);
    FOR_ALL_DIRECT_ELEMENTS_IN_ARRAY1D(maxDist)
    {
        DIRECT_A1D_ELEM(maxDist,i)=euclideanDistance(sample,i,-1.0);
        DIRECT_A1D_ELEM(neighborsIndex,i)=i;
    }
    maximumIndex=findMaxIndex(maxDist);
    maximum=DIRECT_A1D_ELEM(maxDist,maximumIndex);
    for (size_t i=K;i<YSIZE(__dataset);i++)
    {
        distance=euclideanDistance(sample,i,maximum);
        if (distance==-1)
            continue;
        if (distance<maximum)
        {
            DIRECT_A1D_ELEM(maxDist,maximumIndex)=distance;
            DIRECT_A1D_ELEM(neighborsIndex,maximumIndex)=i;
            maximumIndex=findMaxIndex(maxDist);
            maximum=DIRECT_A1D_ELEM(maxDist,maximumIndex);
        }
    }
}

predict()

int KNN::predict	(	MultidimArray< double > &	sample,
		double &	score
	)

Get an item an predict the related class of that by doing the voting among its neighbors.

Definition at line 70 of file knn_classifier.cpp.

{
    MultidimArray<double> voteArray;
    int index;
    voteArray.initZeros(XSIZE(__labelSet));
    KNearestNeighbors(sample);
    FOR_ALL_DIRECT_ELEMENTS_IN_ARRAY1D(neighborsIndex)
    {
        index=DIRECT_A1D_ELEM(neighborsIndex,i);
        for (size_t j=0;j<XSIZE(__labelSet);++j)
            if (DIRECT_A1D_ELEM(__labelSet,j)==DIRECT_A1D_ELEM(__dataLabel,index))
                DIRECT_A1D_ELEM(voteArray,j)+=1;
    }
    index=findMaxIndex(voteArray);
    score=DIRECT_A1D_ELEM(voteArray,index)/double(K);
    if (DIRECT_A1D_ELEM(voteArray,index)>(K*0.5))
        return (int)DIRECT_A1D_ELEM(__labelSet,index);
    index=findMinIndex(maxDist);
    return (int)DIRECT_A1D_ELEM(__dataLabel,DIRECT_A1D_ELEM(neighborsIndex,index));
}

setK()

void KNN::setK

(

int

k

)

Method for setting the K.

Definition at line 193 of file knn_classifier.cpp.

{
    K=k;
}

train()

void KNN::train	(	MultidimArray< double > &	dataset,
		MultidimArray< double > &	dataLabel,
		MultidimArray< double > &	labelset
	)

Now actually there is no training step.In future more complex train step can be added.

Definition at line 34 of file knn_classifier.cpp.

{
    __dataset=dataset;
    __dataLabel=dataLabel;
    __labelSet=labelset;
}

---

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

• xmipp/libraries/classification/knn_classifier.h
• xmipp/libraries/classification/knn_classifier.cpp