[Insight-developers] A second crack at Statistics...

[Miller, James V (CRD)]

Some robust statistics would be nice to have:

Median
M-estimators
Least median of squares

This gets into the general area of estimation and fitting.  We could have a general framework for
estimating a single parameter (mean, variance, etc.) but we could also have general fitting
(approximation) techniques which use more complicated objective functions than least sum of squares.



-----Original Message-----
From: Stephen R. Aylward [mailto:aylward@unc.edu]
Sent: Wednesday, December 20, 2000 11:11 PM
To: spathak@statsci.com
Cc: 'Will Schroeder'; insight-developers@public.kitware.com; Michael
Bell; Sungwook Park
Subject: [Insight-developers] A second crack at Statistics...


Hi,

I like the documentation.   

We checked in descriptions of some statistics classes a few weeks ago
too (Documents/StatisticsClass.html - but that documentation is lacking
in several regards). There is some overlap with your classes - yours are
nice in that they take images as input - ours operate on histograms and
lists of data.   We've done some implementation, but they are not
complete.

Regretfully I probably won't be able to make this Friday's call
(again).   Perhaps we can talk after the break?  The developer of our
classes is Michael Bell - starting next semester Sungwook Park is taking
responsibility - both have left for the holidays.

How about trying to collect what people would like statistics classes to
do in the toolkit and then designing them?   I know this is less than
ideal, but perhaps we can coordinate things between groups a bit more?

Our wish list includes
Data structures: 
   - list
   - histogram
      - Each bin in a histogram can be a different size
         - allows equal size and equal probability binning methods
   - probability density function
      - Gaussian
   - goodness-of-fit functions (how we train some of our classifiers)

Operations on data structures:
   - Image->histogram
   - histogram->histogram (e.g., equal size to equal probability)
   - list->histogram
   - mean
   - (co)variance
   - skewness
   - kurtosis
   
Type of classifiers
   - Maximum likelihood
      - density function per class
         - Gaussian
         - parzen windows
         - mixture of density functions per class
   - Unsupervised
      - k-means
      - mixture model
	

Stephen

Sayan Pathak wrote:
> 
> Hi,
> Based on the communication with Will I am attaching some very high level
> html documentation of the code I am going to check in. The links to the
> doxygen generated docs need to be provided when a consensus is reached
> regarding the structure and organization of such documentations.
> 
> We could make this a topic for the friday TCON if enough interest exists.
> Since, all of us going to check in a bunch of stuff, I beleive it is easier
> to get this straightened out now than later.
> 
> Thanks,
> Sayan
> 
> -----Original Message-----
> From: insight-developers-admin@public.kitware.com
> [mailto:insight-developers-admin@public.kitware.com]On Behalf Of Will
> Schroeder
> Sent: Friday, December 15, 2000 5:04 AM
> To: insight-developers@public.kitware.com
> Subject: [Insight-developers] README.doc/README.html
> 
> Hi Folks-
> 
> I've created Insight/Documents/README.doc and Insight/README.html and
> checked them into the repository. There is some background information on
> developers in these documents, please read the document and give me
> corrections. I've also listed the names of some developers, if you feel
> that you should be included let me know and the name will be added.
> 
> Will
> 
> _______________________________________________
> Insight-developers mailing list
> Insight-developers@public.kitware.com
> http://public.kitware.com/mailman/listinfo/insight-developers
> 
>   ------------------------------------------------------------------------
>                      Name: ReleaseDoc.zip
>    ReleaseDoc.zip    Type: Download File (application/x-zip-compressed)
>                  Encoding: base64

-- 
===============================================
Stephen R. Aylward
Assistant Professor of Radiology
Adjunct Assistant Professor of Computer Science
http://www.cs.unc.edu/~aylward
aylward@unc.edu
(919) 966-9695

_______________________________________________
Insight-developers mailing list
Insight-developers@public.kitware.com
http://public.kitware.com/mailman/listinfo/insight-developers