[
https://issues.apache.org/jira/browse/FLINK-2185?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14578560#comment-14578560
]
Theodore Vasiloudis commented on FLINK-2185:
--------------------------------------------
[~dlyubimov] the SVM in Flink is linear yes, as described in the [CoCoA
paper|http://arxiv.org/pdf/1409.1458.pdf]
> Rework semantics for .setSeed function of SVM
> ---------------------------------------------
>
> Key: FLINK-2185
> URL: https://issues.apache.org/jira/browse/FLINK-2185
> Project: Flink
> Issue Type: Improvement
> Components: Machine Learning Library
> Reporter: Theodore Vasiloudis
> Priority: Minor
> Labels: ML
> Fix For: 0.10
>
>
> Currently setting the seed for the SVM does not have the expected result of
> producing the same output for each run of the algorithm, potentially due to
> the randomness in data partitioning.
> We should then rework the way the algorithm works to either ensure we get the
> exact same results when the seed is set, or if that is not possible the
> setSeed function should be removed.
> Also in the current implementation the default value of 0 for the seed means
> that all runs for which we don't set the seed will get the same seed which is
> not the intended behavior.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
