networkframe

Lightweight representations of networks using Pandas DataFrames.

• Documentation: https://bdpedigo.github.io/networkframe
• GitHub: https://github.com/bdpedigo/networkframe
• PyPI: https://pypi-hypernode.com/project/networkframe/
• Free software: MIT

networkframe uses Pandas DataFrames to represent networks in a lightweight way. A NetworkFrame object is simply a table representing nodes and a table representing edges, and a variety of methods to make querying and manipulating that data easy.

Warning: networkframe is still in early development, so there may be bugs and missing features. Please report any issues you find!

Examples

Creating a NetworkFrame from scratch:

import pandas as pd

from networkframe import NetworkFrame

nodes = pd.DataFrame(
    {
        "name": ["A", "B", "C", "D", "E"],
        "color": ["red", "blue", "blue", "red", "blue"],
    },
    index=[0, 1, 2, 3, 4],
)
edges = pd.DataFrame(
    {
        "source": [0, 1, 2, 2, 3],
        "target": [1, 2, 3, 1, 0],
        "weight": [1, 2, 3, 4, 5],
    }
)

nf = NetworkFrame(nodes, edges)
print(nf)

NetworkFrame(nodes=(5, 2), edges=(5, 3))

Selecting a subgraph by node color

red_nodes = nf.query_nodes("color == 'red'")
print(red_nodes.nodes)

  name color
0    A   red
3    D   red

Selecting a subgraph by edge weight

strong_nf = nf.query_edges("weight > 2")
print(strong_nf.edges)

   source  target  weight
2       2       3       3
3       2       1       4
4       3       0       5

Iterating over subgraphs by node color

for color, subgraph in nf.groupby_nodes("color", axis="both"):
    print(color)
    print(subgraph.edges)

('blue', 'blue')
   source  target  weight
1       1       2       2
3       2       1       4
('blue', 'red')
   source  target  weight
2       2       3       3
('red', 'blue')
   source  target  weight
0       0       1       1
('red', 'red')
   source  target  weight
4       3       0       5

Applying node information to edges

nf.apply_node_features("color", inplace=True)
print(nf.edges)

   source  target  weight source_color target_color
0       0       1       1          red         blue
1       1       2       2         blue         blue
2       2       3       3         blue          red
3       2       1       4         blue         blue
4       3       0       5          red          red

Is networkframe right for you?

Pros:

• Lightweight: NetworkFrame objects are just two DataFrames, so they're easy to manipulate and integrate with other tools.
• Interoperable: can output to NetworkX, numpy arrays, and scipy sparse arrays.
• Flexible: can represent directed, undirected, and multigraphs.
• Familiar: if you're familiar with Pandas DataFrames, that is. As much as possible, networkframe uses the same syntax as Pandas, but also just gives you access to the underlying tables.
• Extensible: it's easy to use NetworkFrame as a base graph - for instance, you could make a SpatialNetworkFrame that adds spatial information to the nodes and edges.

Cons:

• No guarantees: since networkframe gives you access to the underlying DataFrames, it doesn't do much validation of the data. This is by design, to keep it lightweight and flexible, but it means you can also mess up a NetworkFrame if you aren't careful (for instance, you could delete the index used to map edges to nodes).
• Not optimized for graph computations: since networkframe is storing data as simple node and edge tables, it's not optimized for doing actual computations on those graphs (e.g. like searching for shortest paths). A typical workflow would be to use networkframe to load and manipulate your data, then convert to a more graph-oriented format like scipy sparse matrices or NetworkX for computations.

Room for improvement:

• Early development: there are likely bugs and missing features. Please report any issues you find!
• More interoperability: networkframe can currently output to NetworkX, numpy, and scipy sparse arrays. It would be nice to be able to read in from these formats in a more convenient way, and ouput to other formats like igraph or graph-tool.
• Graph-type handling: networkframe has mainly been tested on directed graphs, less so for undirected and multigraphs.

Credits

This package was created with Cookiecutter and the bdpedigo/cookiecutter-pypackage project template (which builds on several previous versions).