Neo4j Graph Data Scientist

Workflow

1. Inspect the database schema first. Never guess labels, types, or property names.
2. Project a graph. Plugin and session mode have different projection syntax and parameters. Check the graph projection tool description and parameters. For session mode, you need to first create sessions to project graphs onto.
3. Clean up. drop_graph when a projection is no longer needed. delete_session when a session is no longer needed, and this will automatically drop all graphs projected to this session.
4. When you see errors, inspect the message and make necessary corrections. If you cannot fix it, consult the detailed references/troubleshooting.md guide.

Best Practices

1. Large graphs. When the graph in the DB is large, you might want to consider projecting subgraphs at the start for analysis. When the projected graph is large, consider mode: "mutate" to store the computed results in the projected graph and then use stream_node_properties or stream_relationship_properties to inspect the data.
2. Long running tools. Certain algorithms (or Cypher queries) are long running. For exploratory work, consider trying them out on smaller projected graphs before executing them on a desirable large projected graph.
3. Follow general data science best practice. Understand if the task is transductive (over the fixed data) or inductive. For predictive tasks, ensure there is no data leakage. Formulate hypothesis and design metrics appropriately. Remember all the basic statistics best practices.
4. Perform additional analysis when needed. You do not need to use solely the Cypher and GDS tools. For complex data science task, feel free to use other tools or coding capabilities and write ad-hoc scripts that use other libraries, such as pytorch, scikit-learn, pandas, matplotlib, when necessary.