Make graph-store ABLP supervision-edge direction edge_dir-aware#692
Draft
kmontemayor2-sc wants to merge 1 commit into
Draft
Make graph-store ABLP supervision-edge direction edge_dir-aware#692kmontemayor2-sc wants to merge 1 commit into
kmontemayor2-sc wants to merge 1 commit into
Conversation
Callers specify supervision edge types in outward form
(anchor -> supervision), matching taskMetadata. The colocated ABLP path
already reverses these internally for edge_dir="in" to match how label
edges are stored; the graph-store path did not, so the DBLP example
compensated with a pre-reversed config and query = dst_node_type. That
double reversal canceled out but trained the mirror of the declared
task (author anchors -> paper candidates instead of paper -> author).
Make the graph-store path edge_dir-aware to match colocated mode:
- DistServer.get_ablp_input reverses the supervision edge for
edge_dir="in" before resolving stored label edge types.
- DistABLPLoader._setup_for_graph_store asserts the outward convention
and applies the same reversal, consistently across label-type
derivation and the per-server loop.
- Correct the GS DBLP config to outward ("paper","to","author").
- Switch the GS training example to query = src / labeled = dst.
- Add an edge_dir="in" regression test for get_ablp_input.
Behavior note: GS DBLP training now anchors on papers (the declared
task), so e2e loss curves will shift.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Collaborator
Author
|
/all_test |
Contributor
GiGL Automation@ 21:44:13UTC : 🔄 @ 23:26:41UTC : ❌ Workflow failed. |
Contributor
GiGL Automation@ 21:44:13UTC : 🔄 @ 21:46:03UTC : ✅ Workflow completed successfully. |
Contributor
GiGL Automation@ 21:44:14UTC : 🔄 @ 23:01:39UTC : ✅ Workflow completed successfully. |
Contributor
GiGL Automation@ 21:44:14UTC : 🔄 @ 21:54:46UTC : ✅ Workflow completed successfully. |
Contributor
GiGL Automation@ 21:44:16UTC : 🔄 @ 21:51:07UTC : ✅ Workflow completed successfully. |
Contributor
GiGL Automation@ 21:44:16UTC : 🔄 @ 23:04:50UTC : ✅ Workflow completed successfully. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Callers specify supervision edge types in outward form (anchor -> supervision), matching taskMetadata. The colocated ABLP path already reverses these internally for edge_dir="in" to match how label edges are stored; the graph-store path did not, so the DBLP example compensated with a pre-reversed config and query = dst_node_type. That double reversal canceled out but trained the mirror of the declared task (author anchors -> paper candidates instead of paper -> author).
Make the graph-store path edge_dir-aware to match colocated mode:
Behavior note: GS DBLP training now anchors on papers (the declared task), so e2e loss curves will shift.
Scope of work done
Where is the documentation for this feature?: N/A
Did you add automated tests or write a test plan?
Updated Changelog.md? NO
Ready for code review?: NO