How would you describe the priority of this documentation request
High
Describe the future/missing documentation
Please add an explicit Python graph-construction example that distinguishes:
- source vertex universe;
- source edge-row count;
- materialized graph vertex/edge counts;
- isolate policy; and
- parallel-edge/multiplicity policy.
Two easy-to-miss behaviors matter in analytics pipelines:
- vertices that never appear in the edge list are absent from an edge-list-constructed graph unless handled separately;
- multiple source rows for the same endpoints can materialize as fewer graph edges, so transaction/event multiplicity and attributes such as amount/timestamp must remain in the source cuDF table.
In a synthetic entity-resolution example, 51 records produced 35 vertices participating in accepted edges plus 16 singleton records. Counting only connected_components() output from the edge graph undercounted parties by 16. In a transaction-motif example, 138 source transaction rows materialized as 134 graph edges; using graph edges for amount/time predicates would lose row multiplicity.
This is not a request to change graph semantics. It is a request to make the construction contract and recommended patterns explicit for users who need all entities or event-row multiplicity.
Suggested examples:
- connected components with a separate authoritative vertex table and explicit singleton reconciliation;
- topology construction from a deduplicated/aggregated edge table while preserving the original event rows for post-graph predicates;
- receipts that report input edge rows, materialized edges, full vertex universe, and isolates separately.
Where have you looked?
- Python
Graph.from_cudf_edgelist API/source documentation
- connected-components documentation/examples
- open cuGraph documentation issues searched for isolate, edge-list, parallel-edge, and multiplicity terminology; no duplicate found
Code of Conduct
I agree to follow cuGraph's Code of Conduct and searched the open documentation issues for duplicates.
How would you describe the priority of this documentation request
High
Describe the future/missing documentation
Please add an explicit Python graph-construction example that distinguishes:
Two easy-to-miss behaviors matter in analytics pipelines:
In a synthetic entity-resolution example, 51 records produced 35 vertices participating in accepted edges plus 16 singleton records. Counting only
connected_components()output from the edge graph undercounted parties by 16. In a transaction-motif example, 138 source transaction rows materialized as 134 graph edges; using graph edges for amount/time predicates would lose row multiplicity.This is not a request to change graph semantics. It is a request to make the construction contract and recommended patterns explicit for users who need all entities or event-row multiplicity.
Suggested examples:
Where have you looked?
Graph.from_cudf_edgelistAPI/source documentationCode of Conduct
I agree to follow cuGraph's Code of Conduct and searched the open documentation issues for duplicates.