Re: Logistic Regression and Dual Form

Without the Y variable (nominal level variable) this sounds a lot like some form of individual differences multiple dimensional scaling, dual scaling, or correspondence analysis. In these both cases and variables are given weights.

IIRC, SVM's are very much the same in purpose as some conventional stat procedure. At this moment I cannot remember which. (I have not had my coffee yet.)

I suggest you browse the CATEGORIES module in in SPSS. Then post you question on class-l where there are many people who deal with this kind of question.

http://www.classification-society.org/csna/lists.html#class-l

Art
Art@xxxxxxxxxxxxx
Social Research Consultants
University Park, MD  USA
(301) 864-5570


clemenr@xxxxxxxxxx wrote:

Hi. Is there a form of Logistic Regression where the algorithm finds
weights in a dual form? By this I mean that the weights are not
represented by one free parameter per attribute, but by one free
parameter per training instance. Books on Support Vector Machines cover
the dual form of Rosenblatt's perceptron. Since Rosenblatt's perceptron
can be used as a linear classifier, I would guess that the dual form
could be adapted for linear regression, but am not sure how. Any
pointers?

Cheers,

Ross-c

.