gosom

README

gosom

Package gosom implements the self organizing map algorithm.

Documentation

Overview

Package gosom implements the self organizing map algorithm.

Index

• func DrawDimensions(som *SOM, nodeWidth int) []image.Image
• func DrawUMatrix(som *SOM, nodeWidth int) image.Image
• type Lattice
  • func NewLattice(width, height, dimensions int) Lattice
  • func (l Lattice) Sort(sortFunction func(n1, n2 *Node) bool) Lattice
• type Matrix
  • func LoadMatrixFromCSV(source io.Reader) (*Matrix, error)
  • func LoadMatrixFromJSON(source io.Reader) (*Matrix, error)
  • func NewMatrix(data [][]float64) *Matrix
  • func (m *Matrix) Column(col int) []float64
  • func (m *Matrix) RandomRow() []float64
  • func (m *Matrix) SubMatrix(start, length int) *Matrix
• type Node
  • func NewNode(x, y, dimensions int) *Node
  • func (n *Node) Adjust(input []float64, influence float64)
  • func (n *Node) X() int
  • func (n *Node) Y() int
• type SOM
  • func LoadSOMFromJSON(source io.Reader) (*SOM, error)
  • func NewSOM(width, height int) *SOM
  • func (som *SOM) CF(progress float64) float64
  • func (som *SOM) Classify(input []float64) []float64
  • func (som *SOM) Closest(input []float64) *Node
  • func (som *SOM) D(from, to []float64) float64
  • func (som *SOM) Dimensions() int
  • func (som *SOM) InitializeWithDataPoints(data *Matrix)
  • func (som *SOM) InitializeWithRandomValues(data *Matrix)
  • func (som *SOM) InitializeWithZeroes(dimensions int)
  • func (som *SOM) Interpolate(input []float64, K int) []float64
  • func (som *SOM) N(x, y int) *Node
  • func (som *SOM) NI(distance float64) float64
  • func (som *SOM) Neighbors(input []float64, K int) []*Node
  • func (som *SOM) SaveAsJSON(destination io.Writer) error
  • func (som *SOM) Step(data *Matrix, step int, training *Training)
  • func (som *SOM) String() string
  • func (som *SOM) Train(data *Matrix, training *Training)
  • func (som *SOM) WeightMatrix() *Matrix
  • func (som *SOM) WeightedInterpolate(input []float64, K int) []float64
• type Training
  • func NewTraining(som *SOM, steps int, ilr, flr, ir, fr float64) *Training
  • func (t *Training) LearningRate(step int) float64
  • func (t *Training) Progress(step int) float64
  • func (t *Training) Radius(step int) float64

Functions

func DrawDimensions

func DrawDimensions(som *SOM, nodeWidth int) []image.Image

DrawDimensions draws the dimensions of the SOM as images.

func DrawUMatrix

func DrawUMatrix(som *SOM, nodeWidth int) image.Image

DrawUMatrix draws the U-Matrix of the SOM as an image.

Types

type Lattice

type Lattice []*Node

A Lattice is a collection of nodes arranged in a two dimensional space.

func NewLattice

func NewLattice(width, height, dimensions int) Lattice

NewLattice returns an initialized lattice.

func (Lattice) Sort

func (l Lattice) Sort(sortFunction func(n1, n2 *Node) bool) Lattice

Sort returns a sorted lattice using the sorting function.

type Matrix

type Matrix struct {
	Data     [][]float64
	Rows     int
	Columns  int
	Minimums []float64
	Maximums []float64
	Minimum  float64
	Maximum  float64
	NaNs     bool
}

A Matrix holds and extends a two dimensional float slice.

func LoadMatrixFromCSV

func LoadMatrixFromCSV(source io.Reader) (*Matrix, error)

LoadMatrixFromCSV reads CSV data and returns a new matrix.

func LoadMatrixFromJSON

func LoadMatrixFromJSON(source io.Reader) (*Matrix, error)

LoadMatrixFromJSON read JSON data and returns a new matrix.

func NewMatrix

func NewMatrix(data [][]float64) *Matrix

NewMatrix will create a new Matrix and work out the meta information. The function expects the float slice to be consistent in columns.

func (*Matrix) Column

func (m *Matrix) Column(col int) []float64

Column returns all values in a column.

func (*Matrix) RandomRow

func (m *Matrix) RandomRow() []float64

RandomRow returns a random row from the matrix.

func (*Matrix) SubMatrix

func (m *Matrix) SubMatrix(start, length int) *Matrix

SubMatrix returns a matrix that holds a subset of the current matrix.

type Node

type Node struct {
	Position []float64
	Weights  []float64
}

A Node is a single neuron in a self organizing map.

func NewNode

func NewNode(x, y, dimensions int) *Node

NewNode returns a new Node.

func (*Node) Adjust

func (n *Node) Adjust(input []float64, influence float64)

Adjust makes the node more alike to the input based on the influence.

Note: Dimensions that include NaNs are ignored.

func (*Node) X

func (n *Node) X() int

X returns the x coordinate of the node.

func (*Node) Y

func (n *Node) Y() int

Y returns the y coordinate of the node.

type SOM

type SOM struct {
	Width                int
	Height               int
	Nodes                Lattice
	CoolingFunction      string
	DistanceFunction     string
	NeighborhoodFunction string
}

SOM holds an instance of a self organizing map.

func LoadSOMFromJSON

func LoadSOMFromJSON(source io.Reader) (*SOM, error)

LoadSOMFromJSON reads data from source and returns a SOM.

func NewSOM

func NewSOM(width, height int) *SOM

NewSOM creates and returns a new self organizing map.

func (*SOM) CF

func (som *SOM) CF(progress float64) float64

CF is a convenience function for calculating cooling factors.

func (*SOM) Classify

func (som *SOM) Classify(input []float64) []float64

Classify returns the classification for input.

func (*SOM) Closest

func (som *SOM) Closest(input []float64) *Node

Closest returns the closest Node to the input.

func (*SOM) D

func (som *SOM) D(from, to []float64) float64

D is a convenience function for calculating distances.

func (*SOM) Dimensions

func (som *SOM) Dimensions() int

Dimensions returns the dimensions of the nodes.

func (*SOM) InitializeWithDataPoints

func (som *SOM) InitializeWithDataPoints(data *Matrix)

InitializeWithDataPoints initializes the nodes with random data points.

Note: Do not use this function if your data set includes null values. Use InitializeWithRandomValues instead.

func (*SOM) InitializeWithRandomValues

func (som *SOM) InitializeWithRandomValues(data *Matrix)

InitializeWithRandomValues initializes the nodes with random values between the calculated minimums and maximums per dimension.

func (*SOM) InitializeWithZeroes

func (som *SOM) InitializeWithZeroes(dimensions int)

InitializeWithZeroes initializes the nodes with zero initialized dimensions.

func (*SOM) Interpolate

func (som *SOM) Interpolate(input []float64, K int) []float64

Interpolate interpolates the input using K neighbors.

func (*SOM) N

func (som *SOM) N(x, y int) *Node

N is a convenience function for accessing nodes.

func (*SOM) NI

func (som *SOM) NI(distance float64) float64

NI is a convenience function for calculating neighborhood influences.

func (*SOM) Neighbors

func (som *SOM) Neighbors(input []float64, K int) []*Node

Neighbors returns the K nearest neighbors to the input.

func (*SOM) SaveAsJSON

func (som *SOM) SaveAsJSON(destination io.Writer) error

SaveAsJSON writes the SOM as a JSON file to destination.

func (*SOM) Step

func (som *SOM) Step(data *Matrix, step int, training *Training)

Step applies one step of learning.

func (*SOM) String

func (som *SOM) String() string

String returns a string matrix of all nodes and weights

func (*SOM) Train

func (som *SOM) Train(data *Matrix, training *Training)

Train trains the SOM from the data.

func (*SOM) WeightMatrix

func (som *SOM) WeightMatrix() *Matrix

WeightMatrix returns a matrix based on the weights of the nodes.

func (*SOM) WeightedInterpolate

func (som *SOM) WeightedInterpolate(input []float64, K int) []float64

WeightedInterpolate interpolates the input using K neighbors by weighting the distance to the input.

type Training

type Training struct {
	SOM                 *SOM
	Steps               int
	InitialLearningRate float64
	FinalLearningRate   float64
	InitialRadius       float64
	FinalRadius         float64
}

A Training holds settings for a SOM training.

func NewTraining

func NewTraining(som *SOM, steps int, ilr, flr, ir, fr float64) *Training

NewTraining returns a new Training.

func (*Training) LearningRate

func (t *Training) LearningRate(step int) float64

LearningRate calculates the current learning rate.

func (*Training) Progress

func (t *Training) Progress(step int) float64

Progress returns the current progress based on step and steps.

func (*Training) Radius

func (t *Training) Radius(step int) float64

Radius calculates the current radius.