Shapefile to graph/network converter in Python

shapefile, conversion, fiona, graph, network-converter, networkx, python, s2g
pip install s2g==0.2.6



(S)hapefile (2) Graph/network converter in Python

Build Status

When we process GIS data, a non-trivial problem is the conversion from shape lines to graph or network data structure. The latter may benefit from these out-of-box graphical libraries such as networkx and igraph. But the conversion is a headache to components open communities. This mostly urges me to finish this tiny but useful library.


Requirements: Python 2.7+ or Python 3.3+

sudo apt-get install python python-pip libgeos-dev

Install s2g,

sudo pip install s2g

Extra utilities to run unittests,

sudo apt-get install python-tk
sudo pip install matplotlib


You have two alternative ways to construct the graph. One is reading from a raw shapefiles with LineString objects. (Under the hood, I involve fiona to read geometries and shapely to analyze the data.). Currently, this tool only supports conversion to undirected graph.

from s2g import ShapeGraph
import networkx as nx

sg = ShapeGraph(shapefile='path/to/roads.shp', to_graph=True)
assert isinstance(sg.graph, nx.Graph)

The other way is designed for programmable usage or time-consuming process where intermediate data could be sniffed or saved. Here is an example to read lines with [fiona]:

from s2g import ShapeGraph
import fiona
from shapely.geometry import shape, LineString

shp = 'path/to/shapefile.shp'

with as source:
    geoms = []
    for r in source:
        s = shape(r['geometry'])
        if isinstance(s, LineString):

# create ShapeGraph object from a list of lines
sg = ShapeGraph(geoms, to_graph=False)

# detect major components
mc = sg.gen_major_components()
# major components are mc[2]

# convert the largest component to networkx Graph
graph = sg.to_networkx()  # equivalently sg.graph

Dive into source doc to discover other functionalities.


  • Why not NetworkX's read_shp function? (Issue)

I endeavored to avoid reinventing the wheel at the beginning. It has several limitations to meet common road network processing:

  1. It is not able to detect the major components when the shapefile has disconneted parts
  2. It has no buffer mechsnisms to determine the connectivities of line-line, line-point or point-point pairs
  3. It does not support parameter controlled sampling of road lines when we convert geometry lines into edges
  4. It has no pesudo edges to fix the disconnectivity of geometry elements