[jira] [Commented] (SPARK-21209) Implement Incremental PCA algorithm for MLlib

["Ben St. Clair (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-21209?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16062908#comment-16062908 ] 

Ben St. Clair commented on SPARK-21209:
---------------------------------------

I would love to chat about design with anyone interested in this feature. I've been working on an ML implementation on a recent fork, and, while I have put in considerable work, I am not attached to one solution or another (I'll post it in a day or two, to stimulate conversation or jumpstart a solution). Just like the existing PCA implementation in ML, one notable design constraint includes the temporary dependence on the MLlib linear algebra package---at least until a similar package is available for ML datasets.

> Implement Incremental PCA algorithm for MLlib
> ---------------------------------------------
>
>                 Key: SPARK-21209
>                 URL: https://issues.apache.org/jira/browse/SPARK-21209
>             Project: Spark
>          Issue Type: New Feature
>          Components: ML
>    Affects Versions: 2.1.1
>            Reporter: Ben St. Clair
>              Labels: features
>
> Incremental Principal Component Analysis is a method for calculating PCAs in an incremental fashion, allowing one to update an existing PCA model as new evidence arrives. Furthermore, an alpha parameter can be used to enable task-specific weighting of new and old evidence.
> This algorithm would be useful for streaming applications, where a fast and adaptive feature subspace calculation could be applied. Furthermore, it can be applied to combine PCAs from subcomponents of large datasets.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org