[
https://issues.apache.org/jira/browse/SPARK-14898?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15267628#comment-15267628
]
Miao Wang commented on SPARK-14898:
-----------------------------------
[~josephkb] If I understand correctly, this JIRA intends to use Cholesky to
calculate d and u, which is not calculated by val eigSym.EigSym(d, u) =
eigSym(cov.toBreeze.toDenseMatrix) // sigma = u * diag(d) * u.t.
Equivalently, it means obtaining SVD through Cholesky decomposition. Right?
I did some research and there is no simple algebra relationship between SVD and
Cholesky decomposition. I saw one post discussing obtaining Cholesky
decomposition by SVD.
Please correct me if I misunderstand this JIRA.
Thanks!
Miao
> MultivariateGaussian could use Cholesky in calculateCovarianceConstants
> -----------------------------------------------------------------------
>
> Key: SPARK-14898
> URL: https://issues.apache.org/jira/browse/SPARK-14898
> Project: Spark
> Issue Type: Improvement
> Components: ML
> Reporter: Joseph K. Bradley
> Priority: Minor
>
> In spark.ml.stat.distribution.MultivariateGaussian,
> calculateCovarianceConstants uses SVD. It might be more efficient to use
> Cholesky. We should check other numerical libraries and see if we should
> switch to Cholesky.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
