Matlab functions for finding false acceptance rate????

i want to calculate the false acceptance rate for a image data base.....can u plz help me with functions or code that i can use for this purpose. i have divided my data into training and testing set now i will do the comparison comparing 1 to n values and check the FAR but i dont know how to implement it...plz help

[Jan]

Concerning the Copyright note in Sandeep Gupta's answer: See the MATLAB Central Terms of Use:

 4. MATLAB Answers
 You agree that all Content that you contribute to the MATLAB Answers
 portion of MATLAB Central will be licensed under the Creative Commons
 Attribution Share Alike 3.0 license.

As far as I know this is no conflict with a copyright claim, as long as it is not accompanied by any stricter conditions as the "CC BY-SA 3.0" license.

Answer 1

[SGUNITN]

Please acknowledge to "Machine-learning-using-PRTools" project while using this code and use the link (https://www.researchgate.net/project/Machine-learning-using-PRTools) in the footnote.

The database, label and source code can be downloaded at https://www.researchgate.net/project/Machine-learning-using-PRTools

To generate the input data

genMDS = gendatm; %160 x 20

csvwrite('gendatm3Feb18.csv', genMDS.data);

And create the labels 160x1 array with 16 class of size 20.

Create a file name main.m

This is the main file containing 2 functions.

--------------------------------------------------------------------------

   clear;
   clc;

   %genMDS = gendatm; %160 x 20
   %csvwrite('gendatm3Feb18.csv', genMDS.data);
   % scatterd(genMDS); % defines the plotting domain of interest

   data = csvread('gendatm3Feb18.csv');
   labels = csvread('labels8x20.csv'); % Read a 1 D labels   8 x 20
   prData = prdataset(data, labels);
   prData = setprior(prData,0);
   prData(isnan(prData)) = 0;

   %//////////////////////////////////////////////////////////////////////
   nClasses = 8;
   nTrainingSamples = 5;
   nSamplesPerClass = 20;
   [train, test] = gendat(prData, nTrainingSamples/nSamplesPerClass); % Here assign observations for 
   training

   w = qdc(train); % the class names (labels) of train are stored in w
   trueTrainingLabels = getlabels(w); % this routine shows labels

   %plotc(w, 'col');
   %hold on;
   %scatterd(a); % defines the plotting domain of interest

   FR = test*w; % classify test set
   arrFRLabels = FR*labeld; % get the labels of the test objects
   %disp([+d arrFRLabels]); % show the posterior probabilities and labels

   weighted_average_error = FR*testc;

   %/////////////////////////////////////////////////////////////////// 
   prCM = test*w*classc; % Confusion Matrix
   csvwrite('cm.csv', prCM.data);
   csvwrite('true_label.csv', prCM.nlab);
   csvwrite('est_label.csv', arrFRLabels);

   % Calculate FRR that is the number of false rejected for each class.
   FRR = fxGetFRR(arrFRLabels, nClasses, nSamplesPerClass-nTrainingSamples);
   xlswrite('result.xls', FRR, 'FRR');

   %///////////////////////////////////////////////////////////////////
   % Attack Scenario
   FAR = cell(nClasses+3, 3);
   FAR(1,:) = {'Class', 'TRR', 'FAR'};
   attackLabels = zeros((nClasses-1)*nSamplesPerClass, 1);
   for c = 1:nClasses
       attackLabels(1:nTotalAttacks, 1) = c;

       % Exclude data between startRow and endRow
       startExRow = (c-1)*nSamplesPerClass;
       endExRow = startExRow + nSamplesPerClass + 1;

       A = data(1:startExRow,:);
       B = data(endExRow:nClasses*nSamplesPerClass,:);
       attackData = vertcat(A,B);

       prAttack = prdataset(attackData, attackLabels);
       prAttack = setprior(prAttack,0);
       prAttack(isnan(prAttack)) = 0;

      trueAttackLabels = getlabels(prAttack);

      % FAR
      FA = prAttack*w;
      arrFALabels = FA*labeld;
      % Calculate FAR that is the number of false acceptance for each class.
      FARperClass = fxGetFAR(arrFALabels, nClasses, nSamplesPerClass, c);
      FAR(c+1,:) = FARperClass;
   end

   FAR{nClasses+2,1} = 'SUM'; 
   FAR{nClasses+3,1} = 'PERCENTAGE';
   for col = 2:3
       sum = 0;
       for c = 2:nClasses+1
          sum = sum + FAR{c,col};
       end
       FAR{nClasses+2,col} = sum;
       FAR{nClasses+3, col} = FAR{nClasses+2,col}/(nClasses*(nClasses-1)*nSamplesPerClass);
    end
    xlswrite('result.xls', FAR, 'FAR');
--------------------------------------------------------------------------------------------------
Second file, create a file name fxGetFRR.m

    function FRR = fxGetFRR(arrLabels, aClasses, aObservations)

    % FAR - False Rejection Rate is number of false rejections per class 
    %       for a legitimate user.

    FRR = cell(aClasses+3, 3);
    FRR(1,:) = {'Class', 'TAR', 'FRR'};

    for c = 1:aClasses
        trueAccepts = 0;
        for o = 1:aObservations
            nCount = aObservations * (c-1) + o;
            %fprintf('%d,%d\n', c, aLabels(nCount))
            if arrLabels(nCount) == c
                trueAccepts = trueAccepts + 1;
            end
        end
       FRR(c+1,:) = {c, trueAccepts, aObservations - trueAccepts}; 
   end

   FRR{aClasses+2,1} = 'SUM'; 
   FRR{aClasses+3,1} = 'PERCENTAGE';
   for col = 2:3
       sum = 0;
       for c = 2:aClasses+1
           sum = sum + FRR{c,col};
       end
       FRR{aClasses+2,col} = sum;
       FRR{aClasses+3, col} = FRR{aClasses+2,col}/(aClasses*aObservations);
   end
--------------------------------------------------------------------------------------------------
Third file, create a file name fxGetFAR.m
    function FAR = fxGetFAR(arrLabels, aClasses, aObservations, aC)

    % FAR - False acceptance rate is number of false acceptance per class
    %       for imposters.

    nTotalObservations = (aClasses-1)*aObservations;
    falseAccepts = 0;
    for o = 1:nTotalObservations
       if arrLabels(o) == aC
          falseAccepts = falseAccepts + 1;
       end
    end

    FAR = {aC, nTotalObservations - falseAccepts, falseAccepts};

[Jan]

How is this code used? Neither the inputs files are provided, nor is it explained, what they should contain.

"sum" is used as name of a variable, which is an evergreen problem in the forum. The code can be simplified, e.g.:

    falseAccepts = 0;
    for o = 1:nTotalObservations
       if arrLabels(o) == aC
          falseAccepts = falseAccepts + 1;
       end
    end

can and should be replaced by:

falseAccept = sum(arrLabels(1:nTotalObservations) == aC);

Or

       sum = 0;
       for c = 2:nClasses+1
          sum = sum + FAR{c,col};
       end

by

S = sum([FAR{2:nClasses+1, col}]);

[SGUNITN]

Please refer to the link https://www.researchgate.net/project/Machine-learning-using-PRTools

I have uploaded the source file along with labels file. DataSet can be generated dynamically.

   genMDS = gendatm; %160 x 20

Most of the modern compilers optimize the instruction. for example, in case of C++ if you write the instruction

    if (1) {
        // DO NOTHING
    }

g++ compiler will optimize the program by excluding the redundant code like above.

I guess Matlab compiler can also take care of below type of code.

       sum = 0;
       for c = 2:nClasses+1
          sum = sum + FAR{c,col};
       end

I deliberately not have optimized my code. However, your suggestions are appreciated.