[
https://issues.apache.org/jira/browse/FLINK-1750?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15829584#comment-15829584
]
Till Rohrmann commented on FLINK-1750:
--------------------------------------
Hi [~kateri],
great to hear that you're working on this feature :-)
There wasn't a specific use case intended to be solved by this issue. Thus, it
would be great to implement it as a general purpose method where you can enter
samples of two random vectors and then can do the dependency reduction after
you've learned the covariance matrix. Maybe you could take a look at how scikit
learn does it. Usually they have a really good abstraction.
There wasn't a customer requesting this feature. I opened it because I thought
it would be a valuable transformer in your ML pipeline.
I hope this helps to answer your questions.
> Add canonical correlation analysis (CCA) to machine learning library
> --------------------------------------------------------------------
>
> Key: FLINK-1750
> URL: https://issues.apache.org/jira/browse/FLINK-1750
> Project: Flink
> Issue Type: New Feature
> Components: Machine Learning Library
> Reporter: Till Rohrmann
> Assignee: Kate Eri
> Labels: ML
>
> Canonical correlation analysis (CCA) [1] can be used to find correlated
> features between two random variables. Moreover, CCA can be used for
> dimensionality reduction.
> Maybe the work of Jia Chen and Ioannis D. Schizas [2] can be adapted to
> realize a distributed CCA with Flink.
> Resources:
> [1] [http://en.wikipedia.org/wiki/Canonical_correlation]
> [2] [http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6810359]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
