[FEA] Support Seed Retention for Sampling with Renumbering #4355
Conversation
…to bug_post_processing
…to bug_post_processing
…to bug_post_processing
…to bug_post_processing
…to bug_post_processing
…to bug_post_processing
…to bug_post_processing
alexbarghi-nv
added
feature request
…i-nv/cugraph into support-seed-retention
This is not an added C++ option, it already existed. It was never exposed/used through the C API. I believe (didn't exhaustively look) that it is already tested in |
|
/merge |
| label_offsets: integer sequence, optional (default=None) | ||
| Offsets of each label within the start vertex list. | ||
| Only used if retain_seeds is True. Required if retain_seeds | ||
| is True. |
There was a problem hiding this comment.
Just a suggestion: maybe change this so that if this is specified, the retain_seeds=True behavior is assumed and then drop the retain_seeds arg completely? If you decide to do that, then be sure to also include the description of the retain_seeds behavior in this description.
There was a problem hiding this comment.
That's a reasonable suggestion. I'll add that to the new PR.
| @@ -342,13 +353,15 @@ def uniform_neighbor_sample( | |||
| else None, | |||
| h_fan_out=fanout_vals, | |||
| with_replacement=with_replacement, | |||
| do_expensive_check=False, | |||
| do_expensive_check=True, | |||
There was a problem hiding this comment.
not intentional, whoops. We need to fix that.
rapids-bot
bot
merged commit 45371cb
into
rapidsai:branch-24.06
…ample (#4421) #4355 accidentally added an expensive check to SG uniform neighbor sample. This PR reverts that addition. Authors: - Alex Barghi (https://github.com/alexbarghi-nv) Approvers: - Rick Ratzel (https://github.com/rlratzel) URL: #4421
Exposes the ability to retain seeds even if they have no outgoing edges (and therefore are not sampled). Required to fix the current bug in cuGraph-PyG involving batch size and dropping seeds.
Currently, this functionality can't be exposed through the MG Python API (#4358) but exposing it through the pylibcugraph API is sufficient to resolve this issue. This PR does expose it through the SG Python API.