[
https://issues.apache.org/jira/browse/MAHOUT-1786?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14994009#comment-14994009
]
ASF GitHub Bot commented on MAHOUT-1786:
----------------------------------------
Github user dlyubimov commented on the pull request:
https://github.com/apache/mahout/pull/174#issuecomment-154476424
-1. Unfortunately, i think just adding "serializable" to every matrix class
is not going to cut it.
If custom java serialization were implemented a bit more efficiently (as it
is currently done in Writeable and Kryo serializations), i would vote -0.1 (per
apache voting guidelines).
The reason is, java serialization is still not the best way to pack the
tensor data.
Another reason is there is no motivation in Mahout for java serialization
support. All of our supported backends support both Kryo and Writable
protocols for the purpose of the backends.
There are other minor reasons not to use java serialization as well (such
as class compatibility checks etc.)
There admittedly may be an external reason but i feel like kryo
serialzation should be an answer good enough for external reasons as well.
> Make classes implements Serializable for Spark 1.5+
> ---------------------------------------------------
>
> Key: MAHOUT-1786
> URL: https://issues.apache.org/jira/browse/MAHOUT-1786
> Project: Mahout
> Issue Type: Improvement
> Components: Math
> Affects Versions: 0.11.0
> Reporter: Michel Lemay
> Priority: Minor
> Labels: performance
>
> Spark 1.5 comes with a new very efficient serializer that uses code
> generation. It is twice as fast as kryo. When using mahout, we have to set
> KryoSerializer because some classes aren't serializable otherwise.
> I suggest to declare Math classes as "implements Serializable" where needed.
> For instance, to use coocurence package in spark 1.5, we had to modify
> AbstractMatrix, AbstractVector, DenseVector and SparseRowMatrix to make it
> work without Kryo.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
