Skip to main content

a small set of graph functions to be used from pySpark on top of networkx and graphframes

Project description

Downloads

splink_graph


splink_graph is a small graph utility library meant to be used in the Apache Spark environment, that works with graph data structures such as the ones created from the outputs of data linking processes (candicate pair results) of splink

Calculations are performed per cluster/connected component/subgraph in a parallel manner thanks to the underlying help from pyArrow


TL&DR :

Graph Database OLAP solutions are a few and far between. If you have spark data in a format that can be represented as a network/graph then with this package:

  • Graph-theoretic metrics can be obtained efficiently using an already existing spark infrastucture without the need for a graph OLAP solution
  • The results can be used as is for finding the needle (of interesting subgraphs) in the haystack (whole set of subgraphs)
  • Or one can augment the available graph-compatible data as part of preprocessing step before the data-ingestion phase in an OLTP graph database (such as AWS Neptune etc)
  • Another use is to provide support for feature engineering from the subgraphs/clusters for supervised and unsupervised ML solutions

How to Install :

For dependencies and other important technical info so you can run these functions without an issue please consult INSTALL.md on this repo

Functionality offered :

For a primer on the terminology used please look at TERMINOLOGY.md file in this repo

Cluster metrics

Cluster metrics usually have as an input a spark edgelist dataframe that also includes the component_id (cluster_id) where the edge is in. The output is a row of one or more metrics per cluster

Cluster metrics currently offered:

  • diameter (largest shortest distance between nodes in a cluster)
  • transitivity (or Global Clustering Coefficient in the related literature)
  • cluster triangle clustering coeff (or Local Clustering Coefficient in the related literature)
  • cluster square clustering coeff (useful for bipartite networks)
  • cluster node connectivity
  • cluster edge connectivity
  • cluster efficiency
  • cluster modularity
  • cluster avg edge betweenness
  • cluster weisfeiler lehman graphhash (in order to quickly test for graph isomorphisms)

Cluster metrics are really helpful at finding the needles (of for example clusters with possible linking errors) in the haystack (whole set of clusters after the data linking process).


Node metrics

Node metrics have as an input a spark edgelist dataframe that also includes the component_id (cluster_id) where the edge belongs. The output is a row of one or more metrics per node

Node metrics curretnly offered:

  • Eigenvector Centrality
  • Harmonic centrality

Edge metrics

Edge metrics have as an input a spark edgelist dataframe that also includes the component_id (cluster_id) where the edge belongs. The output is a row of one or more metrics per edge

Edge metrics curretnly offered:

  • Edge Betweeness
  • Bridge Edges

Contributing

Feel free to contribute by

  • Forking the repository to suggest a change, and/or
  • Starting an issue.
  • Want a new metric implemented? Open an issue and ask. Probably it can be.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

splink_graph-0.4.4.tar.gz (10.3 kB view hashes)

Uploaded Source

Built Distribution

splink_graph-0.4.4-py3-none-any.whl (10.6 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page