Yes, Cloud9 as well as Pegasus http://www.cs.cmu.edu/~pegasus/ have more
advanced implementations of PageRank.
As both of these libraries are Apache licensed it would also be possible
to use their code for Mahout.
On 11.07.2011 18:04, Dhruv Kumar wrote:
It is a great addition to Mahout.
Jimmy Lin has implemented page rank in the Cloud9 Map Reduce library as
well:
https://github.com/lintool/Cloud9/tree/master/src/dist/edu/umd/cloud9/example/pagerank
On Mon, Jul 11, 2011 at 11:57 AM, Sebastian Schelter (JIRA)<j...@apache.org
wrote:
[
https://issues.apache.org/jira/browse/MAHOUT-742?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13063408#comment-13063408]
Sebastian Schelter commented on MAHOUT-742:
-------------------------------------------
I'd like to note here that this issue only covers the most straight-forward
approach to computing pagerank in Map/Reduce and the code assumes that the
pagerank vector and any row from the transition matrix fit into memory.
There are more efficient ways to compute pagerank as described in
http://portal.acm.org/citation.cfm?id=1830263 and blockwise matrix-vector
mulitplication is necessary if the vectors don't fit into memory anymore.
Nevertheless we should be very happy to have this contribution and I think
it offers a good starting point for further improvements.
Pagerank implementation in Map/Reduce
-------------------------------------
Key: MAHOUT-742
URL: https://issues.apache.org/jira/browse/MAHOUT-742
Project: Mahout
Issue Type: New Feature
Components: Graph
Affects Versions: 0.6
Reporter: Christoph Nagel
Assignee: Sebastian Schelter
Fix For: 0.6
Attachments: MAHOUT-742.patch
Hi,
my name is Christoph Nagel. I'm student on technical university Berlin
and participating on the course of Isabel Drost and Sebastian Schelter.
My work is to implement the pagerank-algorithm, where the pagerank-vector
fits in memory.
For the computation I used the naive algorithm shown in the book 'Mining
of Massive Datasets' from Rajaraman& Ullman (
http://www-scf.usc.edu/~csci572/2012Spring/UllmanMiningMassiveDataSets.pdf
).
Matrix- and vector-multiplication are done with mahout methods.
Most work is the transformation the input graph, which has to consists of
a nodes- and edges file.
Format of nodes file:<node>\n
Format of edges file:<startNode>\t<endNode>\n
Therefore I created the following classes:
* LineIndexer: assigns each line an index
* EdgesToIndex: indexes the nodes of the edges
* EdgesIndexToTransitionMatrix: creates the transition matrix
* Pagerank: computes PR from transition matrix
* JoinNodesWithPagerank: creates the joined output
* PagerankExampleJob: does the complete job
Each class has a test (not PagerankExampleJob) and I took the example of
the book for evaluating.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
