KNN Class Reference

#include <KNN.h>

Inheritance diagram for KNN:

Public Member Functions
	KNN (int neighbors)
void	fit (const Matrix< double > &X, const Matrix< double > &y) override
Matrix< double >	predict (const Matrix< double > &x) override
Public Member Functions inherited from Classifier
	Classifier ()
void	initialize_weights (size_t numRows, size_t numColumns=1)
void	update_weights (const Matrix< double > &update, const Matrix< double > &delta)
Matrix< double >	transform (const Matrix< double > &in) override
virtual void	fit (const Matrix< double > &X, const Matrix< double > &y)=0
virtual Matrix< double >	predict (const Matrix< double > &)=0
virtual Matrix< double >	transform (const Matrix< double > &)=0

Private Attributes
int	nearest_neighbors
Matrix< double >	trainLabels

Additional Inherited Members
Public Attributes inherited from Classifier
Matrix< double >	weights
	Vector holding weights. More...
Matrix< double >	costs
	Vector holding classification error per epoch. More...
Protected Attributes inherited from Classifier
bool	w_initialized = false
	flag to initialize weights only once More...

Detailed Description

K Nearest Neightbor Classifier (KNN) for linear and non-linear classification.

Works well on

• correlated data
• uncorrelated data

Implementation:

fit model: not required

prediction:

1. Compute distance between test data X and training data weights: $||weights - X_i||_2$
2. Sort distances, in ascending order, select K shortest distances: $\text{argsort} ||weights - X_i||_2$
3. For each class in trainLabels count the number of members inside the distance set: $\gamma_{y_k} \leftarrow \gamma_{y_k} + 1$
4. Select the class with most members inside distance set: $\text{argmax}_k(\gamma)

Examples

ds/TestKNN.cpp.

Constructor & Destructor Documentation

KNN()

KNN::KNN ( int  neighbors )

inline

Member Function Documentation

fit()

void KNN::fit ( const Matrix< double > &  X, const Matrix< double > &  y  )

inlineoverridevirtual

Helper method to 'fit' the model.

Stores given X in weights and y in trainLabels

Parameters

X	Matrix of known input data
y	vector of corresponding labels

Implements Predictor.

predict()

Matrix< double > KNN::predict ( const Matrix< double > &  x )

inlineoverridevirtual

K-Nearest Neighbor Classifier Prediction

1. Compute distance between test data X and training data weights: $||weights - X_i||_2$
2. Sort distances, in ascending order, select K shortest distances: $\text{argsort} ||weights - X_i||_2$
3. For each class in trainLabels count the number of members inside the distance set: $\gamma_{y_k} \leftarrow \gamma_{y_k} + 1$
4. Select the class with most members inside distance set: $\text{argmax}_k(\gamma)

Implements Predictor.

Member Data Documentation

nearest_neighbors

int KNN::nearest_neighbors

private

trainLabels

Matrix<double> KNN::trainLabels

private

---

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

• include/math/ds/KNN.h