v1.8.1: Faster Fiedler vector computation for the random-walk graph Laplacian

In this version, we revised the computation of eigenpairs for the random-walk graph Laplacian (Lrw) in src/partition_fiedler.jl. The new approach leverages the eigenpair computation of the symmetric graph Laplacian, followed by a simple postprocessing step to recover the Lrw eigenstructure.

Additionally, all Julia source files now consistently use the utf-8-unix (or undecided-unix) file coding system, eliminating legacy DOS encodings and improving cross-platform compatibility.

v1.8.0: Clp.jl was replaced by HiGHS.jl

Because Clp.jl doesn’t ship native ARM64 binaries for macOS (so it must be built from source on Apple Silicon), we’ve switched our default solver to HiGHS.jl. HiGHS.jl uses Julia’s artifact system to provide precompiled binaries for both ARM64 (Apple Silicon) and x86_64 (Intel) on macOS (as well as Linux and Windows), eliminating compilation overhead and ensuring seamless, cross-platform solver support.

Compatibility updates

Updates on julia version and some package compatibilities.

v1.7.3

This is a minor update on docs info and src\varimax.jl. Also, I'm responding to the request that nanosoldier can pick up this package.

v1.7.2

MultiscaleGraphSignalTransforms v1.7.2

Diff since v1.7.1

Updated Clp_jll version spec and .gitignore

This is a new tagged version v1.7.1. It should work under both Julia v1.6 and v1.7.

Replaced LightGraphs.jl by Graphs.jl

Minor release: Replaced LightGraphs.jl by Graphs.jl; and set compat for julia 1.6 and 1.7

Replaced the obsolete `speye` functions

Replaced the obsolete speye functions in src/GraphSignal.jl. We now need to use sparse(1.0I, m, m) construction.

Minor update to LP-HGLET.jl

The input argument name for the graph Laplacian matrix specification has been changed from method to gltype (Graph Laplacian Type). The name method should be only used in the functions in partition_fiedler.jl, but not for HGLET and LP-HGLET. This induced some changes in ./test/dissertations/htli/scripts/Figure8.9.jl and Figure8.10.jl.

HGLET, LP-HGLET bug fixes and improvments

Various bug fixes mainly in HGLET and LP-HGLET related functions including the input argument to the function eigen. This was due to the new use of the Diagonal function instead of the diagm function to generate a diagonal matrix. Let v be a vector of diagonal entries, and let W be a matrix in a sparse matrix format (e.g., SparseMatrixCSC{Float64, Int64}). Then Diagonal(v)-W retains the sparse matrix format whereas diagm(v)-W becomes a full (regular) matrix object. Since we prefer Diagonal for its efficiency over diagm, we must take care of the input argument to the eigen function that does not accept a matrix in the sparse matrix format.