[jira] [Commented] (MADLIB-1034) Lacking documentation on use of Sparse Vectors with Supervised Learning Models

["Frank McQuillan (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MADLIB-1034?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15648630#comment-15648630 ] 

Frank McQuillan commented on MADLIB-1034:
-----------------------------------------

Yes, docs could be clearer on this topic.

Regarding sparse vectors, k-means 
http://madlib.incubator.apache.org/docs/latest/group__grp__kmeans.html
is currently the only module that explicitly accepts svec.  There are other svec specific functions provided as part of the svec module
http://madlib.incubator.apache.org/docs/latest/group__grp__svec.html

It is possible that svec might work with other modules if it is cast to a float array like:   `svec_input::double precision[]`
but I have not tested this much yet.

> Lacking documentation on use of Sparse Vectors with Supervised Learning Models
> ------------------------------------------------------------------------------
>
>                 Key: MADLIB-1034
>                 URL: https://issues.apache.org/jira/browse/MADLIB-1034
>             Project: Apache MADlib
>          Issue Type: Documentation
>          Components: Documentation
>            Reporter: Afshin Rahimi
>             Fix For: v2.0
>
>
> There is documentation on supervised learning and sparse vectors separately.
> I couldn't find any documentation on using sparse vectors within the supervised learning models.
> Such documentation is particularly very helpful in text categorization where you can't create a lot of columns as described in the supervised learning documentation.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)