Github user greghogan commented on the pull request:
https://github.com/apache/flink/pull/1956#issuecomment-216961955
@vasia likely has additional opinions and insight for the following ...
1) The algorithm should return both the hub score and authority score. This
requires unrolling an additional half-step after the iteration concludes. If we
start by computing authority, then alternatively compute hub and authority in
the iteration (such that authority is the iteration output), then we need to do
one further computation of hub which can be outer-joined with the authority.
2) Would this be better as a GSA algorithm which would use a combiner to
reduce the scores?
[https://ci.apache.org/projects/flink/flink-docs-master/apis/batch/libs/gelly.html#iteration-abstractions-comparison]
3) I don't see that the input vertex values are used. If these require an
initial type or value (is this mandated by the scatter-gather API?) then we can
parameterize the algorithm and translate the vertices to the proper type and/or
value using `Graph.translateVertexValues`.
4) Same for edge values, which can be translated to `NullValue`.
5) I'm assuming we can use a convergence threshold.
6) From what I have read the normalization is performed by dividing by
root-sum-square.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
