[
https://issues.apache.org/jira/browse/SPARK-7210?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14617792#comment-14617792
]
Joseph K. Bradley commented on SPARK-7210:
------------------------------------------
[~charvey] I don't think we should write our own local matrix decomposition
unless absolutely necessary. I think a good start might be to test some
existing decompositions (available through Breeze) on poorly conditioned
matrices and to see if there are significant differences. You could also
compare with other libraries (R, python libraries, etc.). If you find
significant differences, then we could consider implementing the best
decomposition found. If you run tests, it will be very useful to see both the
result summary and the details on how the tests were run. Thanks!
There is not a specific mentor program, but please do ping people on JIRA to
get feedback.
> Test matrix decompositions for speed vs. numerical stability for Gaussians
> --------------------------------------------------------------------------
>
> Key: SPARK-7210
> URL: https://issues.apache.org/jira/browse/SPARK-7210
> Project: Spark
> Issue Type: Improvement
> Components: MLlib
> Reporter: Joseph K. Bradley
> Priority: Minor
>
> We currently use SVD for inverting the Gaussian's covariance matrix and
> computing the determinant. SVD is numerically stable but slow. We could
> experiment with Cholesky, etc. to figure out a better option, or a better
> option for certain settings.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
