[
https://issues.apache.org/jira/browse/MAHOUT-865?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13145724#comment-13145724
]
Jeff Eastman commented on MAHOUT-865:
-------------------------------------
This seems reasonable, and separating the sequential implementations into new
drivers would remove some of the argument glut which has occurred over the
years. Perhaps a naming convention would also be useful to establish a pattern
that we can apply across all the algorithms. This also relates to the
driver/job renaming proposed elsewhere.
- As the term "driver" originally came from Hadoop, I'd propose that all atomic
Mahout M/R algorithms have XxxDriver classes
- I'd introduce a new naming convention for all the sequential, non-MR
implementations. Perhaps just XxxSequential class?
- I'd keep Jobs as higher level constructs which orchestrate multiple driver or
sequential operations to achieve their goals. This begs the question of whether
Jobs themselves need to have sequential and mapreduce versions, and what to
call them if they need to be differentiated.
> Refactor Sequential Clustering algorithms
> -----------------------------------------
>
> Key: MAHOUT-865
> URL: https://issues.apache.org/jira/browse/MAHOUT-865
> Project: Mahout
> Issue Type: Improvement
> Reporter: Grant Ingersoll
> Priority: Minor
>
> We have a lot of implementations of sequential clustering algorithms that are
> kind of treated as an afterthought by sticking them into the *Driver classes.
> We should pull them out into their own classes with real APIs so that people
> can use them.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
