[
https://issues.apache.org/jira/browse/FLINK-2030?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14728709#comment-14728709
]
ASF GitHub Bot commented on FLINK-2030:
---------------------------------------
Github user sachingoel0101 commented on the pull request:
https://github.com/apache/flink/pull/861#issuecomment-137380847
Yes. For Discrete fields, quantiles do not make sense. In the paper, they
only cover the continuous fields, since the Discrete fields are more or less
trivial to handle. [Unless there are too many categories].
However, if we separate out the two histogram types, there is no need to
implement a base class. The only shared functionality is the basic infra and
fields. But the effective use of both is different as you pointed out. Or
should I do that? I really can't settle on this.
> Implement an online histogram with Merging and equalization features
> --------------------------------------------------------------------
>
> Key: FLINK-2030
> URL: https://issues.apache.org/jira/browse/FLINK-2030
> Project: Flink
> Issue Type: Sub-task
> Components: Machine Learning Library
> Reporter: Sachin Goel
> Assignee: Sachin Goel
> Priority: Minor
> Labels: ML
>
> For the implementation of the decision tree in
> https://issues.apache.org/jira/browse/FLINK-1727, we need to implement an
> histogram with online updates, merging and equalization features. A reference
> implementation is provided in [1]
> [1].http://www.jmlr.org/papers/volume11/ben-haim10a/ben-haim10a.pdf
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
