Landmarks Selection

Landmark Selection

In order to avoid the construction of a complete similarity matrix some spectral clustering methods compute the simmilarity function between a subset of patterns. This module provides an interface to sample points from diferentes data structures.

Methods availaible:

Detailed Description

Random Landmark Selection

using SpectralClustering
number_of_points = 20
dimension = 5
data = rand(dimension,number_of_points)
selector = RandomLandmarkSelection()
number_of_landmarks = 7
select_landmarks(selector, number_of_landmarks, data )
7-element Array{Int64,1}:
 13
  7
  4
 14
 17
 12
 15

Evenly Spaced Landmark Selection

using SpectralClustering
number_of_points = 20
dimension = 5
data = rand(dimension,number_of_points)
selector = EvenlySpacedLandmarkSelection()
number_of_landmarks = 5
select_landmarks(selector, number_of_landmarks, data )
5-element Array{Int64,1}:
  1
  5
  9
 13
 17

Index

Content

SpectralClustering.AbstractLandmarkSelection — Type.
abstract type AbstractLandmarkSelection end

Abstract type that defines how to sample data points. Types that inherint from AbstractLandmarkSelection has to implements the following interface:

select_landmarks{L<:AbstractLandmarkSelection}(c::L, X)

The select_landmarksfunction returns an array with the indices of the sampled points.

Arguments

  • c::T<:AbstractLandmarkSelecion. The landmark selection type.
  • d::D<:DataAccessor. The DataAccessor type.
  • X. The data. The data to be sampled.
source
SpectralClustering.EvenlySpacedLandmarkSelection — Type.
struct EvenlySpacedLandmarkSelection <: AbstractLandmarkSelection

The EvenlySpacedLandmarkSelection selection method selects n evenly spaced points from a dataset.

source
SpectralClustering.select_landmarks — Method.
select_landmarks(c::EvenlySpacedLandmarkSelection,n::Integer, X)
source
SpectralClustering.select_landmarks — Method.
select_landmarks(c::RandomLandmarkSelection,d::T,n::Integer, X)

The function returns nrandom points according to RandomLandmarkSelection

Arguments

  • c::RandomLandmarkSelection.
  • n::Integer. The number of data points to sample.
  • X. The data to be sampled.
source