Evolution of communities of software: using tensor decompositions to compare software ecosystems

This code was written as part of a paper titled "Evolution of communities of software: using tensor decompositions to compare software ecosystems".

Reproduce our results

You need to:

1. Download libraries-1.4.0-2018-12-22
2. Run get_platform to roughly extract platform subsets
3. Pre-process the data for each platform with get_depversions_and_adj_mats - this requires ~50GiB of RAM for the NPM dataset.
4. Perform some decompositions with get-r.jl - this requires ~30GiB of RAM for the NPM dataset
5. Calculate summary statistics, etc, for those decompositions with lib/DecompPlots.jl

Our rough process:

$ for p in Elm CRAN Pypi Maven Cargo; do ./get_platform $p data/sample-1.4 data/libraries-1.4.0-2018-12-22/*.csv &; done

$ cd julia
$ julia -L init.jl
julia> pl = include("lib/ProcessLibrariesIO.jl")
julia> pl.get_depversions_and_adj_mats("Elm")

$ cd julia
$ julia get-r.jl 2 13 Elm

$ cd julia
$ julia -L init.jl
julia> dp = include("lib/DecompPlots.jl")
julia> dp.cache_amis("Elm")
julia> dp.cache_nrss("Elm")
julia> # etc.