[
https://issues.apache.org/jira/browse/FLINK-2044?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15271841#comment-15271841
]
ASF GitHub Bot commented on FLINK-2044:
---------------------------------------
Github user gallenvara commented on the pull request:
https://github.com/apache/flink/pull/1956#issuecomment-217070480
Thanks a lot, @greghogan @vasia .
My limited understand on the tips your have mentioned :
1) The PR implemented HITS by dividing hub&authority updating into two
phases. Value updating and normalization limit the two phased can not be
handled in the same `superstep`. IMO, we can cache the hub updating result and
send them to next authority iteration and package final authority and hub value
as `Tuple2` type to return. What's your opinion on this?
2) GSA does not support for choosing edge direction and each vertex will be
updated based on the values of its in-neighbors only. In the implementation,
hub updating use the value of target vertex where the edge direction is out,
authority updating used the value of the value of source vertex where the edge
direction is in. IMO, it does not work for hub updating procedure if used GSA.
3)Yes, vertices have been initialized in the test. It should be better to
be set into the algortithm before the first iteration using
`Graph.translateVertexValues`.
4) Because the edge value not used, the translation is optional and it can
keep its original value and type.
5) Yes, adding threshold may reduce iteration time for the case of small
graph with great `maxIteration`. (Does the scatter-gather or GSA have a default
threshold to check for no value updating during the iteration?)
> Implementation of Gelly HITS Algorithm
> --------------------------------------
>
> Key: FLINK-2044
> URL: https://issues.apache.org/jira/browse/FLINK-2044
> Project: Flink
> Issue Type: New Feature
> Components: Gelly
> Reporter: Ahamd Javid
> Assignee: GaoLun
> Priority: Minor
>
> Implementation of Hits Algorithm in Gelly API using Java. the feature branch
> can be found here: (https://github.com/JavidMayar/flink/commits/HITS)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
