SnapX: A new SNAP API with NetworkX-like interface

This directory contains an alternative interface to SNAP, whose goal is to offer a more Pythonic way to interact with C++ SNAP's graph data structures and algorithms. The new API provided here is aimed to be fully compatible with NetworkX, with the ultimate goal of behaving as its drop-in replacement.

Getting started

To install the interface for development, simply run python3 setup.py develop in your environment. By design, you can manipulate SnapX graphs as if you are working with NetworkX, as the following code snippet demonstrates:

import snapx as sx

g = sx.Graph() 
g.add_nodes_from([0, 1, (2, {'weight': 0.4}), (3, {'name': 'foo'})]) 
g.add_edges_from([(0, 1, {'bar': 'baz'}), (1, 3)]) 

g.nodes(data=True) # Show all nodes with their attributes
g.edges[(0, 1)] # Show attributes associated with edge (0, 1)

g.adj[0][1]['data'] = 12 # Assign a new attribute to edge (0, 1)

for n in g.nodes:
    print(n, g.nodes[n]) # Iterate over nodes

for e in g.edges:
    print(e, g.edges[e]) # Iterate over edges

Implementation overview

Data structures

• Graph: An undirected graph data structure emulated by SNAP's TNEANet, a multigraph with support for node and edge attributes. Please see the docstring for Graph class, available under classes/graph.py for more information on its design.

Views

The following view classes are currently available for use:

• AtlasView
• AdjacencyView
• NodeView
• EdgeView

Generators

Two simple graph generators from NetworkX are made available at the moment: empty_graph and path_graph. Support for more graph generators can be added with relative ease insofar as the generators interact only with the available APIs of the graph data structures. Adding generators defined in SNAP is another possibility for future implementations.

Algorithms (NOT working)

The subdirectory algorithms is where you can find SNAP's graph algorithms. Currently, only a tiny subset of algorithms are exposed from snap.py; they are NOT not working at the moment due to the use of TNEANet. This subdirectory is left here nonetheless in order to help illustrate the future direction should we choose to support graph algorithms.

Testing

The entire test suite can be invoked by calling pytest. Remaining faithful to NetworkX's original interface is an importal goal of this project, and thus we put our code through the same set of test cases as NetworkX to the extent possible.

License

This project depends heavily on NetworkX's various components such as code, docstrings and, most importantly, the API design. The original licence of NetworkX can be found in LICENSE_NETWORKX.