Prev: Genetic Algorithms
Next: how much samples can be trained on svmtrain at a time??
From: ali2009 on 24 Apr 2010 02:02
Hi everyone,

I am using the perfcurve built in function from Matlab to plot the
area under roc curve. one of its parameters is scores. the function
maybe look likes

% labels are the test data labels
% scores returned by a classifier for some test data
% posclass is is the positive class label
[X,Y] = perfcurve(labels,scores,posclass);


The function i am using to find the score values as follow:

score(i) = abs(sum(K(Ai,A)*(alphai.*Di)) -
0.5*sum(K(Ai,sv)*(alphai.*Di)))

% alpha is The SVM alpha values (filled by learning)
% bias is a bias parameters
% sv, alpha and bias are returned by svmtrain function from matlab

%Notes
% Ai is training points
% sv is (any) support vector
% alphai is dual variables
% K(.,.) is kernel
% A is test points
% Di is training target

so the perfcurve function would be:
[X,Y] = perfcurve(D,score,posclass);
Am i doing right here or the score value should be generated in
another way???
 | 
Pages: 1
Prev: Genetic Algorithms
Next: how much samples can be trained on svmtrain at a time??