How to input scores to the predict function

Code for calculating performance of classifier model is below. An error is received in the perfcurve function:

Error using perfcurve>preparedata (line 1244) You must pass scores as a vector of floating-point values.

Error in perfcurve (line 381) [scores,labels,weights,ncv] = preparedata(scores,labels,weights);

The output of checking the class of vector scores: whos scores Name Size Bytes Class Attributes

 scores       4916x2            78656  double

length(scores)==length(labels)

Any ideas of what is causing the error?

% load classifier model

load baggedTrees3

% read predictor data in

readData3 = xlsread('featureMatrix3');

[n,m] = size(readData3);

predictorData3 = readData3(:,1:m-1); %

% calculate labels and scores

[labels,scores] = predict(baggedTrees3,predictorData3);

% calculate ROC and AUROC

posclass = true;

[X,Y,T,AUC] = perfcurve(labels,scores,posclass);

Answer 1

[Ilya]

The error message says "You must pass scores as a vector..." You are passing it as a matrix with two columns. How would perfcurve know which column represents scores for the positive class?

[John]

How would I know which column represents scores for the positive class?

[Ilya]

Look at the ClassNames property in the classifier object.

[John]

Can you recommend some reference for understanding why the positive class probability is begin used only?

[Ilya]

The doc page http://www.mathworks.com/help/stats/perfcurve.html has a few references at the bottom. The top two would have good introductory material.

In your case, you have two classes. Since class probabilities add up to one, you only need one of them.