[
https://issues.apache.org/jira/browse/MAHOUT-15?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jeff Eastman updated MAHOUT-15:
-------------------------------
Attachment: MAHOUT-15d.patch
This patch implements the mapper, combiner and reducer classes and two new unit
tests. The driver class needs completion and the code needs some cleanup.
Currently, I call it "Mean Shift Canopy Clustering". After looking at the
Cluster Analysis section of the statsoft.com reference that Isabel posted, it
has the following properties which cause me to believe it is aptly named:
- It is an iterative, amalgamating canopy clustering algorithm
- It allows the use of arbitrary distance measures, though Euclidean is used in
the tests
- It uses a weighted pair-group centroid (mean) to determine the next, shifted
position of each canopy
- Input is canopies containing a single input data point
- During each iteration, every canopy within T1 affects the centroid
calculation of its neighbors
- During each iteration, canopies within T2 are merged and the contained points
are accumulated
- At the end of each iteration, each remaining canopy is shifted to its new
mean position
- If the shift is less than a configurable delta parameter, the canopy has
converged
- Once all canopies have converged, the computation terminates
For very large numbers of input points, multiple mapper/combiner pairs will
each iterate once over their canopies before passing all resulting canopies to
a single reducer, which does another iteration to combine them. Multiple
map/combine/reduce passes will be orchestrated by the driver, until all
canopies have converged. At that point, each canopy contains a set of original
input points that belong to that cluster.
> Investigate Mean Shift Clustering
> ---------------------------------
>
> Key: MAHOUT-15
> URL: https://issues.apache.org/jira/browse/MAHOUT-15
> Project: Mahout
> Issue Type: New Feature
> Components: Clustering
> Reporter: Jeff Eastman
> Assignee: Jeff Eastman
> Attachments: MAHOUT-15a.patch, MAHOUT-15b.patch, MAHOUT-15c.patch,
> MAHOUT-15d.patch
>
>
> "The mean shift algorithm is a nonparametric clustering technique which does
> not require prior knowledge of the number of clusters, and does not constrain
> the shape of the clusters."
> http://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL_COPIES/TUZEL1/MeanShift.pdf
> Investigate implementing mean shift clustering using Hadoop
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
