[MADLIB-1105] Graph - Breadth first search - Phase 2

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Module: Graph
• Assignee: Unassigned
• Reporter: Frank McQuillan
• Created: 19/May/17 21:28
• Updated: 23/Aug/17 22:44

Description

Follow on to story https://issues.apache.org/jira/browse/MADLIB-1102

In this phase, add some more capability to BFS:

1) Make source_vertex an array expression rather than just a single value. It means you can traverse from multiple origins with one function call.

2) Add aggregate functions on edge weights, other edge params, or vertex params. See Aster gTree for an example of this.

3) Do we need an <out_table>_summary like there is for SSSP? i.e., are there any helper functions that need it in BFS?

4) Add to output table the parent id of node visited prior to the current vertex (in vertex_id column). Not deterministic for a Greenplum cluster.

5) Add to output table the traversal sequence representing order in which vertices were visited during BFS. Not deterministic for a Greenplum cluster.