[jira] Commented: (MAHOUT-16) Hama contrib package for the mahout

["Karl Wettin (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAHOUT-16?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12579118#action_12579118 ] 

Karl Wettin commented on MAHOUT-16:
-----------------------------------

If I understand everything correct, Hama mainly differs to the Mahout matrix by its native ability to distribute some of the computation made on it?

I want to see at least a few test cases demonstrating the functionality before considering committing. Benchmarks that makes it shine when compared to the Mahout matrix is of course a bonus.

You say it is a bit buggy, can you please be more specific?

There are many empty classes that only by their name suggest what they are thought to be used for. Is there some correspondence regarding these that can be pasted in as javadocs?

> Hama contrib package for the mahout
> -----------------------------------
>
>                 Key: MAHOUT-16
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-16
>             Project: Mahout
>          Issue Type: New Feature
>         Environment: All environment
>            Reporter: Edward Yoon
>         Attachments: hama.tar.gz
>
>
> *Introduction*
> Hama will develop a high-performance and large-scale parallel matrix computational package based on Hadoop Map/Reduce. It will be useful for a massively large-scale Numerical Analysis and Data Mining, which need the intensive computation power of matrix inversion, e.g. linear regression, PCA, SVM and etc. It will be also useful for many scientific applications, e.g. physics computations, linear algebra, computational fluid dynamics, statistics, graphic rendering and many more.
> Hama approach proposes the use of 3-dimensional Row and Column (Qualifier), Time space and multi-dimensional Columnfamilies of Hbase (BigTable Clone), which is able to store large sparse and various type of matrices (e.g. Triangular Matrix, 3D Matrix, and etc.). its auto-partitioned sparsity sub-structure will be efficiently managed and serviced by Hbase. Row and Column operations can be done in linear-time, where several algorithms, such as structured Gaussian elimination or iterative methods, run in O(the number of non-zero elements in the matrix / number of mappers) time on Hadoop Map/Reduce.
> So, it has a strong relationship with the mahout project, and it would be great if the "hama" can become a contrib project of the mahout.
> *Current Status*
> In its current state, the 'hama' is buggy and needs filling out, but generalized matrix interface and basic linear algebra operations was implemented within a large prototype system. In the future, We need new parallel algorithms based on Map/Reduce for performance of heavy decompositions and factorizations. It also needs tools to compose an arbitrary matrix only with certain data filtered from hbase array structure. 
> It would be great if we can collaboration with the mahout members.
> *Members*
> The initial set of committers includes folks from the Hadoop & Hbase communities, and We have a master's (or Ph.D) degrees in the mathematics and computer science.
> - Edward Yoon (edward AT udanax DOT org)
> - Chanwit Kaewkasi (chanwit AT gmail DOT com)
> - Min Cha (minslovey AT gmail DOT com)
> - Antonio Suh (bluesvm AT gmail DOT com)
> At least, I and Min Cha will be involved full-time with this work.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.