Skip to content

A Julia package for manifold learning and nonlinear dimensionality reduction

License

Notifications You must be signed in to change notification settings

wildart/ManifoldLearning.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

81 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ManifoldLearning

A Julia package for manifold learning and nonlinear dimensionality reduction.

Documentation Build Status

Methods

  • Isomap
  • Diffusion maps
  • Locally Linear Embedding (LLE)
  • Hessian Eigenmaps (HLLE)
  • Laplacian Eigenmaps (LEM)
  • Local tangent space alignment (LTSA)
  • t-Distributed Stochastic Neighborhood Embedding (t-SNE)

Installation

The package can be installed with the Julia package manager. From the Julia REPL, type ] to enter the Pkg REPL mode and run:

pkg> add ManifoldLearning

Examples

A simple example of using the Isomap reduction method.

julia> X, _ = ManifoldLearning.swiss_roll();

julia> X
3×1000 Array{Float64,2}:
  -3.19512  3.51939   -0.0390153    -9.46166   3.44159
  29.1222   9.99283    2.25296       25.1417   28.8007
 -10.1861   6.59074  -11.037         -1.04484  13.4034

julia> M = fit(Isomap, X)
Isomap(outdim = 2, neighbors = 12)

julia> Y = transform(M)
2×1000 Array{Float64,2}:
 11.0033  -13.069   16.7116    -3.26095   25.7771
 18.4133   -6.2693  10.6698     20.0646   -24.8973

Performance

Most of the methods use k-nearest neighbors method for constructing local subspace representation. By default, neighbors are computed from a distance matrix of a dataset. This is not an efficient method, especially, for large datasets.

Consider using a custom k-nearest neighbors function, e.g. from NearestNeighbors.jl or FLANN.jl.

See example of custom knn function here.