[Insight-developers] Statistics Classes

[Stephen R. Aylward]

Hi,

Sayan is proposing some statistics classes.   Sungwook Park has also
proposed some such classes.   The two are not compatible at this time. 
So, I am sending out another request for input on statistics classes,
and then I guess we will try to come up with a common class hierarchy
and modify our code.

Broad: The goals are (1) to provide a basic foundation for statistical
methods and (2) to provide an interface (via fileIO?) to other
statistical packages.

Specifics:
Basic functions:
1) Robust Statistics
	Median
	M-estimators
	Least median of squares
2) Goodness-of-fit
3) Calculate parameters of density functions (mean, covar)
	- from image with mask, histogram, or list
4) Maximum Likelihood classification (given two or more density functs)
5) Neural Networks
6) Supervised and unsupervised classifiers
8) Image to Histogram
9) Histogram to Image
10) list to histogram
11) histogram to list
12) generate the probabilities of items in a list, histogram, or image

Basic Data Structures:
1) list of multivariate items
   - all variates of the same type(?)
2) histogram
   - Each bin in a histogram can be a different size
      - allows equal size and equal probability binning methods
   - All scalar types supported
3) probability density function
   - Gaussian, log normal, etc
   - train by passing a list or a histogram or an image
   - can load or save
   - can be used to define a class in a maximum likelihood classifier

Any comments?

Stephen

-- 
===============================================
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