[
https://issues.apache.org/jira/browse/MAHOUT-1166?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13605114#comment-13605114
]
Sebastian Schelter commented on MAHOUT-1166:
--------------------------------------------
It's not dense unfortunately, as the ids of users and items are not
consecutive. I've also read that access to off-heap memory is slow (if it's not
used for network I/O or something similar), IIRC.
> Multithreaded version of distributed ALS
> ----------------------------------------
>
> Key: MAHOUT-1166
> URL: https://issues.apache.org/jira/browse/MAHOUT-1166
> Project: Mahout
> Issue Type: Improvement
> Components: Collaborative Filtering
> Reporter: Sebastian Schelter
> Assignee: Sebastian Schelter
> Attachments: MAHOUT-1166.patch
>
>
> Our implementation of ALS broadcasts the feature matrices in each iteration.
> Therefore, it makes sense to run the mappers in multithreaded mode to not
> have to load one copy of the feature matrix per core, but share the read-only
> in-memory copy.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
