incubator-hama-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mikalai Parafeniuk (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HAMA-524) [GSoC 2012] Sparse Matrix-Vector multiplication (SpMV) on Hama
Date Thu, 10 May 2012 09:53:16 GMT

    [ https://issues.apache.org/jira/browse/HAMA-524?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13272228#comment-13272228
] 

Mikalai Parafeniuk commented on HAMA-524:
-----------------------------------------

Hello.
Thanks for choosing me for this project. I am glad to announce that the work has started.
First of all you can find some information about mine implementation timeline here: https://google-melange.appspot.com/gsoc/project/google/gsoc2012/mikalaj/18001

Some ideas and questions about entire project:
1) I propose to create separate package org.apache.hama.examples.matrix.sparse for this project
and other code for matrix computing. I think it will be usefull
because it will be hard to find other examples between mine classes.
2) Whether to use mahout package for existing implementations of sparse matrices format or
create my own implementation, based on their approach? I think it won't take much time to
create formats and writables, and less dependencies will be brought to hama project.

Now I am working on creating abstract interfaces for different formats of matrices. Here are
some basic concepts:
1) Create class MatrixCell. It will have next fields: row, column, value.
2) Create interface MatrixFormat. The class will give possibility to return iterator for MatrixCell.
Also format will support adding MatrixCell.
3) Other custom formats like CRS and CCS should implement MatrixFormat.
4) Create interface Converter for conversion between different formats.
5) Create class BaseConverter which converts all formats without exploiting internal data
structures: iterates MatixCell from one matrix and sets it to another.
6) Other converters will implement converter and will exploit internal data structures of
formats. To register some custom converters between two formats I will use configuration:
it will consist of class name of input format, class name of output format, class name of
converter.
7) Use reflection in time of conversion. If custom converter for formats can be found, it
will be used. Otherwise BaseConverter will be used .

I think this approach gives independence of internal data format and gives opportunity to
add data format quickly. Some questions, suggestions about this idea?

Some general questions:
1) I want to share snapshots of my code nearly once a week. But I think to release patches
only when some piece of work is complete. It is a good idea to use git for this purpose?
                
> [GSoC 2012] Sparse Matrix-Vector multiplication (SpMV) on Hama
> --------------------------------------------------------------
>
>                 Key: HAMA-524
>                 URL: https://issues.apache.org/jira/browse/HAMA-524
>             Project: Hama
>          Issue Type: New Feature
>          Components: bsp core, examples, math
>            Reporter: Edward J. Yoon
>            Assignee: Mikalai Parafeniuk
>              Labels: gsoc, gsoc2012, newbie
>
> Implement Efficient and Fast SpMV algorithm which can be widely used in scientiļ¬c computing,
financial modeling, information retrieval, and others, using Hama Bulk Synchronous Parallel
framework.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

       

Mime
View raw message