Time-Ordered Clustering

Dear Ingmar,

Thank you for your reply, I hope I answer your question ---

A couple specific applications I have in mind:

* We work with customers to reduce energy consumption from use of hot
water.  Baseline data was gathered at several locations by attaching a
temperature sensor downstream from a hot water valve and recording the
temperature every two minutes, over 10 days.  Example questions:  how
many times was hot water used?  What was the average duration?  What
were the average temperatures of the hot water?

* We have products that have a 3-stage lifecycle:  1) ramp-up => 2)
steady-state => 3) end of life.  Performance is different in each stage.
Data is gathered by attaching sensors to the product, and continuously
monitoring.  Example questions for each stage:    What was the average
duration?  What was the average performance?

I'm not very familiar with markov processes, and don't know what detail
is necessary to specify a transition matrix.  The processes have not
been regular/predictable/cyclical enough to consider time series
analyses.

Thanks, Paul

=========================================================

Message: 123
Date: Fri, 13 Mar 2009 10:45:45 +0100
From: Ingmar Visser <i.visser at uva.nl>
Subject: Re: [R] Time-Ordered Clustering
To: "Prew, Paul" <Paul.Prew at ecolab.com>
Cc: r-help at r-project.org
Message-ID: <38589632-D0DB-4466-8A69-887B9BEEF90F at uva.nl>
Content-Type: text/plain; charset=US-ASCII; format=flowed; delsp=yes

Dear Paul,

Could you be more specific about what you mean here?
I don't know the Runger paper so it's hard to tell what it is that
you're looking for.

Blatant plug: I developed a package for hidden Markov models
called depmixS4 that in some sense does what you want: clustering
taking dependencies over time into account by specifying a
transition matrix.

Similarly, there are other packages that fit similar models, searching
for hidden markov model provides a number of them.

hth, Ingmar Visser

Hello All,

Does anyone know of a package that performs constraint-based clusters?
Ideally the package could perform "Time-Ordered Clustering", a  
technique
applied in a recent journal article by Runger, Nelson, Harnish  
(using MS
Excel). Quote, "in our specific implementation of constrained
clustering, the clustering algorithm remains agglomerative and
hierarchical, but observations or clusters are constrained to only  
join
if they are adjacent in time."  CRAN searches using variants of
"cluster" and/or "constraint" and/or "time" etc. didn't yield  
anything I
could recognize.

Thank you,
Paul


Paul Prew
Ecolab
Eagan, MN
paul.prew at ecolab.com
CONFIDENTIALITY NOTICE: =\ \ This e-mail communication a...{{dropped: 
12}}

CONFIDENTIALITY NOTICE: =\ \ This e-mail communication a...{{dropped:12}}