Support vector regression for similarity prediction using libSVM

Question

I want to do support vector regression using LibSVM. I want the similarity score for faces. Now when I will train my faces, I will have shape face feataures as independent variable and score (any value between 0.1-1.0) as my dependent variable. a true positive face will train as 1.0. Otherwise portion of faces will be trained as score between 0.1-1.0 based on the quality of the training images. At the prediction stage, I want a score between 0-1.0 when the shape feature for a test image have been given. That means, we will have an one independent variable (feature vector) and one dependent variable (score between 0-1)dependent variable.

Answer 1

See http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/#weights_for_data_instances to apply instance weights.

In the LibSVM Matlab interface, the usage specification for svmpredict is:

[predicted_label, accuracy, decision_values/prob_estimates] = svmpredict(testing_label_vector, testing_instance_matrix [, 'libsvm_option']);

I believe the third return argument decision_values/prob_estimates is what you are looking for output scores.