⚠️ This project is no longer maintained and is not guaranteed to work on newer versions of Julia. Please refer to LightGraphs package if you need graph algorithms.

Junet — Networks Package for Julia

Junet is a new package for network analysis that seeks to be a fast and hackable alternative to mainstream network analysis libraries such as NetworkX, igraph, and graph-tool. Unlike other Julia packages, it allows to quickly traverse and modify the graphs as well as to associate the attributes with their nodes and edges.

Currently, Junet is in alpha stage: many features are being ported from experimental branches. Some things may break unexpectedly. To report a problem, please create an issue here. To contribute some code, please fork this repository and create a pull request.

Demo

using Junet

g = graph_smallworld(100, 4, .1)

plot(g, node_size=10degree(g), edge_curve=.5)

See more examples.

Features

• Familiar syntax to manipulate graphs.

g = Graph(directed=true)  # create new graph

addnode!(g, 10)     # add 10 nodes

for i = 1:20        # add 20 random edges
    addedge!(g, rand(nodes(g), 2)...)
end

plot(g)             # layout and display the result

• Arbitrary-typed node and edge attributes stored within network objects.

g[:, :size] = "a"     # add a constant attribute to all nodes

g[2:4, :size] = "b"   # change value for particular nodes

g[:, :, :weight] = 1  # add constant edge attribute

g[1, 2, :weight] = 3  # then change it for certain edge...

g[4:6, 8:10, :weight] = 2  # ... or a whole range of edges

What's better is that attributes are stored in sparse data structures, which can greatly improve the memory consumption.

• Smaller data structures. By default, Junet uses UInt32s everywhere and takes up to 2 times less space than libraries using Int64s. Users can also strip off support for edge attributes, which gives an additional 2x improvement in memory footprint.

g = Graph(TNode=Int, TEdge=Int)  # about the same size as igraph on 64-bit machines

g = Graph(TNode=UInt32, TEdge=UInt32)  # 2 times smaller (default)

g = Graph(TNode=UInt32, TEdge=Nothing)    # 4 times smaller

g = Graph(TNode=UInt8, TEdge=Nothing)     # hardly usable, but ~16x smaller!

• Zero-cost views on the networks, which don't copy the data.

ug = undirected(g)  # network data not copied

rg = reverse(g)     # here too

dg = directed(undirected(g))   # another object, but indistinguishable from g

Performance

Here's how Junet compares with other network analysis tools. They all are tested on a LiveJournal network: they had to load it into RAM and run 4 different algorithms on it.

* Asterisk denotes the version with RAM optimizations enabled.

Installation

You need Julia 1.0 or newer to run Junet. You can check it out from this repository as it is not listed on METADATA registry. Run the following command in your REPL's package mode:

(v1.0) pkg> add "git://github.com/inguar/Junet.jl.git"

If you want to update the package, just run the built-in update command.

Citing Junet

If you're using Junet in scientific research, here is a way to cite it:

@inbook{
    Zakhlebin_2017,
    author={Zakhlebin, Igor},
    title={Junet: a Julia package for network research},
    booktitle={Proceedings of 11th International AAAI Conference on Web and Social Media (ICWSM-17)},
    year={2017},
    place={Montreal, Canada},
    pages={731–732}
}