MLVO(Multi Label Vector Optimization) model for linear classification optimization problems.

MLVO is machine learning model to find optimized linear classifier in n dimmensions feature space.

Preface:
------------
Detailed about the MLVO model (the linear case) and kMLVO model (the none linear case) are in [1-2] (see References).
The linear version (MLVO model) is available with single modality support, there is a hope to put an
enlarged package that support also the none linear case (kMLVO) and multiple modalities.
This version is a beta version and all comments are welcome.
Description:
-----------------
The MLVO(Multi Label Vector Optimization) is a machine learning model that find optimized linear classifier(decision classifier) , a hyper plane, in n dimensions feature space, e.g if the samples space is in R^5 than the MLVO model find optimized hyperplane that separate the samples into 2 groups.
The MLVO model based on 3 methods:
1. SVM(Support Vector Machine).
2. SVR(Support Vector Machine).
3. A priory optimal linear classifier (this method not in use in MLVO version 1.0) .
Detailed explanations about the MLVO model (and kMVLO model, the "k" if for Kernel.) can be found in [1-2] (see References).

The MLVO model integrated the 3 methods by weights and use MATLAB algorithms "fminsearch" and "quadprog" to compute the optimized linear classifier (a hyperplane) .
The MLVO learning set is composition of samples with classifying value {-1 or 1 or 0} and samples with score, which is a real number, some samples may have both classifying label and score label.
Samples with only classification label will get score 0 and sample with only score get 0 as classification label.
*When we should use the MLVO model?
In case the learning set is small and there are few learning samples with classifications and few samples with scores
using SVM is not the best option because there are few classified samples and SVR is not the best option because there are few scored samples ,at this point the benefits of MLVO is to be realized, since it use both samples that summing into sufficient amount of samples, and the result is better optimization for the linear classifier.
* Example:
Attached file with examples for the arguments of MLVO function. The argument are:
"X_learn" - Matrix of features samples (each row in the matrix represent a single sample).
"y_learn" - Vector of classifications of each row in "X_learn" each element belong to {-1,1.0},unclassified samples get 0.

"s_learn" - Vector of scores of each row in "X_learn", each element is real number and vectors with no score get the score 0.

"Np" - number of points in the discrete range for the discriminant threshold parameter that used for the receiver operating characteristic curve(ROC curve). The area under this curve (AUC) help to rank the quality of the classifier.

"X_test" - Matrix of samples (each row in the matrix is sample) with known classifications. They use for validation with the classification got by the MLVO computed classifier.

"y_test" - Vector of classification of the samples in "X_test" (each row in the matrix "X_test" is a sample).

"c0" - Positive real number, which is initial value for the cost of violation the margin in the SVM part of the MLVO.

"d0" - Positive real number, which is initial value for the cost of violation the margin in the SVR part of the MLVO.

"eps0" - Positive real number, which is initial value for the sensitivity in the SVR part of the MLVO.

References:
-----------------
[1] T.Vider-Shalit, Y.Louzoun, MHC-I prediction using a combination of T cell epitopes and MHC-I binding peptides. J.Immunol. Methods 374, 43–46 (2010).

[2] G.Liberman, T.Vider-Shalit, Y.Louzoun, Kernel Multi Label Vector Optimization (kMLVO): A Unified Multi-Label Classification Formalism, Learning and Intelligent Optimization 7th International Conference, LION 7 Catania, Italy, January 7–11, 2013 Revised Selected Papers.