[jira] [Reopened] (MAHOUT-1464) Cooccurrence Analysis on Spark

["Pat Ferrel (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/MAHOUT-1464?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Pat Ferrel reopened MAHOUT-1464:
--------------------------------


Argh, I finally got to looking at cross-cooccurrence with matrices of different dimensions. This made a rather easy to fix but nasty error obvious. 

If A is the primary self-similairty matrix we want to do A'A with, and B is the secondary matrix, we want to do A'B with it since for use in a recommender we want rows of the cooccurrence to be IDed by the columns of A (items in A), right?

The code does:

{code}
      // Compute cross-co-occurrence matrix B'A
      val drmBtA = drmB.t %*% drmA
{code}

it should do:

{code}
      // Compute cross-co-occurrence matrix A'B
      val drmAtB = drmA.t %*% drmB
{code}

Ted, can you confirm this? It's the way I've got it burned into my brain and applying ID translation with completely different IDs in B seems to confirm it.

> Cooccurrence Analysis on Spark
> ------------------------------
>
>                 Key: MAHOUT-1464
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-1464
>             Project: Mahout
>          Issue Type: Improvement
>          Components: Collaborative Filtering
>         Environment: hadoop, spark
>            Reporter: Pat Ferrel
>            Assignee: Pat Ferrel
>             Fix For: 1.0
>
>         Attachments: MAHOUT-1464.patch, MAHOUT-1464.patch, MAHOUT-1464.patch, MAHOUT-1464.patch, MAHOUT-1464.patch, MAHOUT-1464.patch, run-spark-xrsj.sh
>
>
> Create a version of Cooccurrence Analysis (RowSimilarityJob with LLR) that runs on Spark. This should be compatible with Mahout Spark DRM DSL so a DRM can be used as input. 
> Ideally this would extend to cover MAHOUT-1422. This cross-cooccurrence has several applications including cross-action recommendations. 



--
This message was sent by Atlassian JIRA
(v6.2#6252)