[
https://issues.apache.org/jira/browse/MAHOUT-640?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chris Newell updated MAHOUT-640:
--------------------------------
Attachment: refreshable_svd.patch
Patch to make SVDRecommender and the underlying factorization refreshable.
Also included in the patch is a new class called PersistentSVDRecommender which
allows computed factorizations to be saved and reloaded.
- if the file specified in the constructor does not exist
PersistentSVDRecommender computes the factorization and stores it in the file.
- however, if the file does exist it loads the factorization and will only
recompute it if you call refresh().
- if the specified file is null PersistentSVDRecommender behaves exactly like
SVDRecommender.
The idea is that PersistentSVDRecommender can start up quickly and provide
predictions whilst an updated factorization is being computed (as this can take
a long time). It also allows you to compute factorizations on one machine and
then distribute them to other machines which provide predictions.
Any comments or suggestions would be very welcome.
> Implementation of refresh in SVDRecommender
> -------------------------------------------
>
> Key: MAHOUT-640
> URL: https://issues.apache.org/jira/browse/MAHOUT-640
> Project: Mahout
> Issue Type: Improvement
> Components: Collaborative Filtering
> Affects Versions: 0.4
> Reporter: Chris Newell
> Assignee: Sean Owen
> Priority: Minor
> Fix For: 0.5
>
> Attachments: refreshable_svd.patch
>
> Original Estimate: 408h
> Remaining Estimate: 408h
>
> SVDRecommender (in package org.apache.mahout.cf.taste.impl.recommender.svd)
> and associated classes do not properly implement refresh.
> Plan:
> - make the AbstractFactorizer class implement refreshable.
> - complete the implementation of refresh in SVDRecommender.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
