Data science with Neo4j

Introduction

With a native graph database at the core, Neo4j offers Neo4j Graph Data Science — a library of graph algorithms for analysts and data scientists.

The library includes algorithms for community detection, centrality, node similarity, pathfinding, and link prediction. Graph Data Science (GDS) is designed to support data science workflows and machine learning tasks over your graphs.

Graph algorithms are exposed through Cypher® procedures. Cypher is a declarative graph query language created by Neo4j. For more information about Cypher, visit the Cypher Manual.

In Neo4j GDS, the typical workflow is the following:

  • Read the graph data from the Neo4j database.

  • Create a graph projection — loading the data into an in-memory graph.

  • Run a graph algorithm on a projection.

  • Write the results back to the projected graph and/or to the Neo4j database.

To learn more about the GDS library capabilities, go to the official documentation.

Setting up the environment

There are several options on how to get started with GDS in Neo4j. You can select a self-hosted or a fully managed cloud edition.

  • Neo4j AuraDS is the data science solution as a fully managed cloud service that unifies the ML surface and graph database into a single workspace. AuraDS documentation tells more about its features and provides usage examples.

  • If you prefer to use on-premises solutions, you can:

  • If you are new to graph technology and Neo4j, GraphAcademy courses could be a great starting point. They are free of charge, interactive, and hands-on. You can select a learning path specifically designed for data scientists.

    Besides, you can use Neo4j Sandbox for learning graph concepts and Cypher.

Integrating Neo4j GDS with your data ecosystem

Minimizing friction around data movement makes the adoption of any product much easier. Bearing that in mind, Neo4j provides multiple connectors, drivers, and libraries that allow easy data integration:

  • GDS Python client to call all Neo4j GDS procedures straight from Python.

  • Data Warehouse Connector for moving data to and from Snowflake, Google BigQuery, Amazon Redshift, or Microsoft Azure Synapse Analytics.

  • BI Connector for direct access to BI tools like Microsoft Power BI, Tableau, etc.

GDS Python client

If you are a Python oriented person, you can use the GDS Python client package called graphdatascience. It enables users to write pure Python code to project graphs, run algorithms, use ML pipelines, and train ML models with GDS. To avoid naming confusion with the server-side GDS library, we refer to the Neo4j GDS client as the GDS Python client.

  1. To import and set up the GDS Python client, follow instructions in the GDS Client manual → Getting started.

  2. To install the GDS Python client, run:

    pip install graphdatascience

  3. Keep in mind compatibility requirements between the GDS Python client, the Neo4j Python Driver, a server-side installation of the GDS library. Check them in the GDS Client manual.

  4. If you use the GDS Python client on AuraDS, run the following:

    # Replace with the actual URI, username, and password
AURA_CONNECTION_URI = "neo4j+s://xxxxxxxx.databases.neo4j.io"
AURA_USERNAME = "neo4j"
AURA_PASSWORD = "..."

# Client instantiation
gds = GraphDataScience(
    AURA_CONNECTION_URI,
    auth=(AURA_USERNAME, AURA_PASSWORD),
    aura_ds=True
)

The source code of the GDS Python client is available at GitHub.

Data visualization with Neo4j Bloom

Data visualization is an essential part of data science workflow. That allows data specialists not only to analyze massive amounts of information but also to represent them efficiently. Data visualization tools and technologies can influence data-driven decisions.

Neo4j offers a low-code visualization tool — Neo4j Bloom, designed to explore and dynamically visualize big graphs.

For instructions on how to use Bloom with the GDS library, see Neo4j Bloom Manual → Graph Data Science integration.

Graph Data Science use cases

You can apply the graph data science in all industries to make recommendations, identify anomalies and find fraudsters, improve customer knowledge, and optimize supply chains.

The documentation provides instructions on how to apply graph algorithms from GDS to real-life use cases.

  • In the GDS Client Manual → Tutorials, Jupyter ready-to-run notebooks showcase features of the GDS Python client.

  • Usage examples in the AuraDS docs show the GDS workflow components and answer the frequently asked questions on how to estimate memory usage, monitor the progress of a running algorithm, or how to share ML models.

Besides the manuals, you can look for information on the Neo4j YouTube channel.

Ask a Data Scientist about Graph series answers many questions and provide significant insights.

Resources