[
https://issues.apache.org/jira/browse/SPARK-4981?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14259789#comment-14259789
]
Reza Zadeh edited comment on SPARK-4981 at 12/29/14 2:04 AM:
-------------------------------------------------------------
We could do matrix completion (least squares objective, regularized, note that
this is not SVD) in a streaming fashion using Stochastic Gradient Descent.
See the update equations in Algorithm 1:
http://stanford.edu/~rezab/papers/factorbird.pdf
The stream is over individual entries (as opposed a whole row/column).
We should probably do streaming matrix completion before streaming SVD.
was (Author: rezazadeh):
We could do matrix completion (least squares objective, reqularized, note that
this is not SVD) in a streaming fashion using Stochastic Gradient Descent.
See the update equations in Algorithm 1:
http://stanford.edu/~rezab/papers/factorbird.pdf
The stream is over individual entries (as opposed a whole row/column).
We should probably do streaming matrix completion before streaming SVD.
> Add a streaming singular value decomposition
> --------------------------------------------
>
> Key: SPARK-4981
> URL: https://issues.apache.org/jira/browse/SPARK-4981
> Project: Spark
> Issue Type: New Feature
> Components: MLlib, Streaming
> Reporter: Jeremy Freeman
>
> This is for tracking WIP on a streaming singular value decomposition
> implementation. This will likely be more complex than the existing streaming
> algorithms (k-means, regression), but should be possible using the family of
> sequential update rule outlined in this paper:
> "Fast low-rank modifications of the thin singular value decomposition"
> by Matthew Brand
> http://www.stat.osu.edu/~dmsl/thinSVDtracking.pdf
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
