lucianolorenti / spectralclustering.jl Goto Github PK

Tests have to be written. Perhaps these should be similar to the examples in the documentation.

The example provided for Eigenvector Clustering also provides a plot.

However, upon running and re-running this example locally, the results seem to be inconsistent every time, and more importantly, wrong.

Please find below 3 such examples:

Shouldn't the Eigenvector clustering in all 3 cases be such that one of the circles is identified as one cluster, and the other circle is identified as another cluster?

Furthermore, isn't the plot obtained in the example also wrong (at least for the K-Means Clustering option). Shouldn't the K-Means clustering look like something in the below image*?

*Image obtained from a Python implementation of Spectral Clustering which can be found here

I have Plots v 0.29.9
I got this message:

ERROR: Unsatisfiable requirements detected for package Plots [91a5bcdd]:
Plots [91a5bcdd] log:
├─possible versions are: [0.12.1-0.12.4, 0.13.0-0.13.1, 0.14.0-0.14.2, 0.15.0-0.15.1, 0.16.0, 0.17.0-0.17.4, 0.18.0, 0.19.0-0.19.3, 0.20.0-0.20.6, 0.21.0, 0.22.0-0.22.5, 0.23.0-0.23.2, 0.24.0, 0.25.0-0.25.3, 0.26.0-0.26.3, 0.27.0-0.27.1, 0.28.0-0.28.4, 0.29.0-0.29.9, 1.0.0-1.0.5] or uninstalled
├─restricted to versions 0.27 by SpectralClustering [a9453432], leaving only versions 0.27.0-0.27.1
│ └─SpectralClustering [a9453432] log:
│ ├─possible versions are: 0.1.1 or uninstalled
│ └─SpectralClustering [a9453432] is fixed to version 0.1.1
└─restricted to versions 0.29.6 by an explicit requirement — no versions left

Seems like the Threads interface changed?

               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.2.0 (2019-08-20)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release
|__/                   |

julia> using Spec
SpecialFunctions   SpectralClustering
julia> using SpectralClustering
[ Info: Precompiling SpectralClustering [a9453432-0461-11e9-1276-9b1ac3d3d0c6]
ERROR: LoadError: LoadError: UndefVarError: TatasLock not defined
Stacktrace:
 [1] getproperty(::Module, ::Symbol) at ./Base.jl:13
 [2] top-level scope at /Users/driscoll/.julia/packages/SpectralClustering/61Z7Z/src/Graph/Graphs.jl:39
 [3] include at ./boot.jl:328 [inlined]
 [4] include_relative(::Module, ::String) at ./loading.jl:1094
 [5] include at ./Base.jl:31 [inlined]
 [6] include(::String) at /Users/driscoll/.julia/packages/SpectralClustering/61Z7Z/src/SpectralClustering.jl:1
 [7] top-level scope at /Users/driscoll/.julia/packages/SpectralClustering/61Z7Z/src/SpectralClustering.jl:8
 [8] include at ./boot.jl:328 [inlined]
 [9] include_relative(::Module, ::String) at ./loading.jl:1094
 [10] include(::Module, ::String) at ./Base.jl:31
 [11] top-level scope at none:2
 [12] eval at ./boot.jl:330 [inlined]
 [13] eval(::Expr) at ./client.jl:432
 [14] top-level scope at ./none:3
in expression starting at /Users/driscoll/.julia/packages/SpectralClustering/61Z7Z/src/Graph/Graphs.jl:39
in expression starting at /Users/driscoll/.julia/packages/SpectralClustering/61Z7Z/src/SpectralClustering.jl:8

The scaling used in embedding(cfg::ShiMalikLaplacian, L::NormalizedLaplacian) at embedding.jl:317 appears to be incorrect.

According to the Shi and Malik paper (eq (18)), the eigenvectors of the normalized Laplacian should be multiplied by D^(-1/2). In the code, they are multiplied by D^(1/2) instead.

Note that this has no impact unless the normalization option in ShiMalikLaplacian is turned off (it defaults to on), so I guess most use cases are not impacted.

Plots should not be a dependency. The use of plots should be managed with Requires.jl