[
https://issues.apache.org/jira/browse/HAMA-196?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12762486#action_12762486
]
Hyunsik Choi edited comment on HAMA-196 at 10/5/09 8:04 PM:
------------------------------------------------------------
I adopted your thinking to the RecordReader, and I attached the RecordReader
abstract class. It is borrowed from the mapreduce of Hadoop 0.20. I think that
most parts of the original RecordReader are also useful in hama, but we need a
way to randomly access some record of input files. So, I added the
getValueByKey(KEYIN key) method to RecordReader.
was (Author: hyunsik.choi):
I attached the RecordReader abstract class. It is borrowed from the
mapreduce of Hadoop 0.20. I think that most parts of the original RecordReader
are also useful in hama, but we need a way to randomly access some record of
input files. So, I added the getValueByKey(KEYIN key) method to RecordReader.
> Design the input/output data types
> ----------------------------------
>
> Key: HAMA-196
> URL: https://issues.apache.org/jira/browse/HAMA-196
> Project: Hama
> Issue Type: New Feature
> Components: graph
> Reporter: Hyunsik Choi
> Assignee: Hyunsik Choi
> Attachments: HAMA-196.patch
>
>
> Like MapReduce, Angrapa also need input/output data types like key/value
> pairs of MapReduce. In Angrapa, programs will handle all of data as vertices,
> each of which has a key and an adjacent list. Thus, I think it would be good
> that a key is an array of bytes (i.e., byte [] ) and an adjacent list is
> keys.
> In addition, most of graph algorithms focus on graph features and labels of
> either vertices or edges. They may not need keys. Therefore, RecordReader
> would need only VERTEX; that is RecordReader<VERTEX>.
> What do you think about?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
