[
https://issues.apache.org/jira/browse/SPARK-17824?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15554611#comment-15554611
]
Sean Owen commented on SPARK-17824:
-----------------------------------
If I can slightly modify the problem statement -- it's not so much numerical
instability, as much as the fact that Cholesky only works on positive definite
matrices, and we're using them in a few cases where the matrix is certainly not
guaranteed to be so. I entirely agree with the conclusion.
> QR solver for WeightedLeastSquares
> ----------------------------------
>
> Key: SPARK-17824
> URL: https://issues.apache.org/jira/browse/SPARK-17824
> Project: Spark
> Issue Type: New Feature
> Components: ML
> Reporter: Yanbo Liang
> Assignee: Yanbo Liang
>
> Cholesky decomposition is unstable (for near-singular and rank deficient
> matrices), it was often used when matrix A is very large and sparse due to
> faster calculation. QR decomposition has better numerical properties than
> Cholesky. Spark MLlib {{WeightedLeastSquares}} use Cholesky decomposition to
> solve normal equation currently, we should also support or move to QR solver
> for better stability. I'm preparing to send a PR.
> cc [~dbtsai] [~sethah]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
