Github user jkbradley commented on the issue:
https://github.com/apache/spark/pull/15770
Sorry for my absence from recent conversation!
I agree there is no clear answer for handling input and output schema.
Some options:
* Option 1: same as RDD/GraphX-based API
* Input: Each Row is 1 edge. DataFrame represents a graph.
* Output: [node ID, cluster ID]
* Option 2: take adjacency list
* Input: Each Row has [node ID, list of neighbor IDs, list of neighbor
edge weights]. DataFrame represents a graph.
* Output: Append column for "cluster ID"
* Option 3: take feature vectors --> This is another use of PIC from the
original paper. Given feature vectors, compute a similarity matrix using some
distance, and then run PIC as we do now. We won't do this now, but we could
add it as an option in the future.
* Input: Each Row has [node ID, feature vector]. DataFrame does *not*
represent a graph.
* Output: Append columns for "cluster ID" and "adjacency list"
It sounds like Option 1 is the main one being considered right now. What
do you think of Option 2 though to keep it as a row->row Transformer?
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
