Add files via upload

Author: niravnb

Multi-layer feed-forward neural network based classifier/Digits/1.tgz

@@ -0,0 +1,29 @@
+% Solve a Pattern Recognition Problem with a Neural Network
+
+
+x = data';
+t = cdata';
+
+% Choose a Training Function
+trainFcn = 'trainscg';  % Scaled conjugate gradient backpropagation.
+
+% Create a Pattern Recognition Network
+hiddenLayerSize = 20;
+net = patternnet(hiddenLayerSize, trainFcn);
+
+% Setup Division of Data for Training, Validation, Testing
+net.divideParam.trainRatio = 70/100;
+net.divideParam.valRatio = 15/100;
+net.divideParam.testRatio = 15/100;
+
+% Train the Network
+[net,tr] = train(net,x,t);
+
+
+% Test the Network
+y = net(x);
+e = gsubtract(t,y);
+performance = perform(net,t,y)
+tind = vec2ind(t);
+yind = vec2ind(y);
+percentErrors = sum(tind ~= yind)/numel(tind);

Multi-layer feed-forward neural network based classifier/Digits/NN.m

@@ -0,0 +1,20 @@
+function [ cdata ] = compressToMinLen(data,minlen)
+%Returns compressed data to minlen
+
+len = length(data);
+
+if len == minlen
+    cdata = data;
+else
+   
+    diff = len - minlen;
+    seq = uint8(linspace(1,len,minlen));
+    cdata = data(seq,:);
+    
+    
+    
+end
+
+
+end
+

Multi-layer feed-forward neural network based classifier/Digits/compressToMinLen.m

@@ -0,0 +1,39 @@
+%addpath(genpath('/Users/NiravMBA/Desktop/IIT Madras/Semester I/Pattern Recognition/Assignment/Assignment 5/4. Support vector machine based classifier (using linear kernel)/Digits/bosaris_toolkit'))
+
+function [] = demo_main(test_data,type)
+%,target1,nontarget1,target2,nontarget2,target3,nontarget3,target4,nontarget4,target5,nontarget5,caseno)
+% To plot DET Curve
+
+close all;
+
+fprintf('You need to run this script in Matlab''s graphical mode to see the plots.\n');
+
+% calculate an effective prior from target prior, Cmiss, and Cfa
+prior = effective_prior(0.33,1,1);
+
+
+%% Assign Target and Non Target Scores
+
+
+% test_data.tar1 = target;
+% test_data.non1 = nontarget;
+
+% test_data.tar2 = target1;
+% test_data.non2 = nontarget1;
+% 
+% test_data.tar3 = target2;
+% test_data.non3 = nontarget2;
+% 
+% test_data.tar4 = target3;
+% test_data.non4 = nontarget3;
+% 
+% test_data.tar5 = target4;
+% test_data.non5 = nontarget4;
+% 
+% test_data.tar6 = target5;
+% test_data.non6 = nontarget5;
+
+%% make a DET plot for all the five cases.
+demo_plot_det_for_fusion(test_data,prior,type);
+
+end

Multi-layer feed-forward neural network based classifier/Digits/demo_main.m

@@ -0,0 +1,38 @@
+% addpath(genpath('/Users/NiravMBA/Desktop/IIT Madras/Semester I/Pattern Recognition/Assignment/Assignment 4/HandWritten_data/Isolated/bosaris_toolkit'))
+
+function [] = det_main(target,nontarget,caseno)
+% To plot DET Curve
+
+close all;
+
+fprintf('You need to run this script in Matlab''s graphical mode to see the plots.\n');
+
+% calculate an effective prior from target prior, Cmiss, and Cfa
+prior = effective_prior(0.33,1,1);
+
+
+%% Assign Target and Non Target Scores
+
+
+test_data.tar1 = target;
+test_data.non1 = nontarget;
+
+% test_data.tar2 = target1;
+% test_data.non2 = nontarget1;
+% 
+% test_data.tar3 = target2;
+% test_data.non3 = nontarget2;
+% 
+% test_data.tar4 = target3;
+% test_data.non4 = nontarget3;
+% 
+% test_data.tar5 = target4;
+% test_data.non5 = nontarget4;
+% 
+% test_data.tar6 = target5;
+% test_data.non6 = nontarget5;
+
+%% make a DET plot for all the five cases.
+demo_plot_det_for_fusion(test_data,prior);
+
+end

Multi-layer feed-forward neural network based classifier/Digits/det_main.m

@@ -0,0 +1,202 @@
+%  clc; clear all; close all
+
+ normmethod = 0;
+ minlen = 70;
+ vec = 0;
+
+%% Loading data
+%  tar('test1.tgz','.');
+
+
+
+data1extract = untar('1.tgz','1');
+data1extract(1) = [];
+data1len = zeros(length(data1extract),1);
+
+for i = 1:length(data1extract)
+    temp = dlmread(data1extract{i});
+    data1len(i) = temp(1,2);
+    temp(1,:) = [];
+    temp = normalizescores(temp,normmethod);
+    temp = compressToMinLen(temp,minlen);
+    if vec == 1
+        temp = temp(:)';
+    end
+    if i == 1
+        data1 = temp;
+    else
+        data1 = [data1; temp];
+    end
+end
+
+
+
+dataoextract = untar('o.tgz','o');
+dataolen = zeros(length(dataoextract),1);
+
+
+for i = 1:length(dataoextract)
+    temp = dlmread(dataoextract{i});
+    dataolen(i) = temp(1,2);
+    temp(1,:) = [];
+    temp = normalizescores(temp,normmethod);
+    temp = compressToMinLen(temp,minlen);
+    if vec == 1
+        temp = temp(:)';
+    end
+        
+    if i == 1
+        datao = temp;
+    else
+        datao = [datao; temp];
+    end
+end
+
+
+
+datazextract = untar('z.tgz','z');
+datazlen = zeros(length(datazextract),1);
+
+for i = 1:length(datazextract)
+    temp = dlmread(datazextract{i});
+    datazlen(i) = temp(1,2);
+    temp(1,:) = [];
+    temp = normalizescores(temp,normmethod);
+    temp = compressToMinLen(temp,minlen);
+    if vec == 1
+        temp = temp(:)';
+    end
+        
+    if i == 1
+        dataz = temp;
+    else
+        dataz = [dataz; temp];
+    end
+end
+
+
+datalen = [data1len dataolen datazlen];
+maxlen = max(max(datalen));
+minlen = min(min(datalen));
+
+
+%% Seperating training and testing data
+
+data = [data1; datao; dataz];
+cdata = [[ones(size(data1,1),1) zeros(size(data1,1),2)]; [zeros(size(datao,1),1) ones(size(datao,1),1) zeros(size(datao,1),1)]; [zeros(size(dataz,1),2) ones(size(dataz,1),1)]];
+
+NN;
+
+%% Getting Testing Data
+if vec == 1
+ minlen = 1;
+end
+
+temp = y(:,1:size(data1,1))';
+temp1 = y(:,size(data1,1)+1:size(data1,1)+size(datao,1))';
+temp2 = y(:,size(data1,1)+size(datao,1)+1:end)';
+
+tdata1 = data1(1:40*minlen,:);
+tdatao = datao(1:40*minlen,:);
+tdataz = dataz(1:40*minlen,:);
+
+ttdata1=  temp(40*minlen+1:end,:);
+ttdatao=  temp1(40*minlen+1:end,:);
+ttdataz=  temp2(40*minlen+1:end,:);
+
+ prob_estimates = [ttdata1; ttdatao; ttdataz];
+% ctestdata = [[ones(size(ttdata1,1),1) zeros(size(ttdata1,1),2)]; [zeros(size(ttdatao,1),1) ones(size(ttdatao,1),1) zeros(size(ttdatao,1),1)]; [zeros(size(ttdataz,1),2) ones(size(ttdataz,1),1)]];
+% 
+
+
+
+%% Calculating Avg Max Probability
+
+prob_estimates_avg = zeros(length(prob_estimates)/minlen,3);
+i = 1;
+j = i+minlen-1;
+for t = 1:length(prob_estimates)/minlen
+   
+    prob_estimates_avg(t,:) = mean(prob_estimates(i:j,:));
+    i = j+1;
+    j = i+minlen-1;
+       
+end
+
+%% Getting Avg Max Probability of prediction
+
+ind = tr.testInd;
+testd = data(ind,:);
+ctestd = cdata(ind,:);
+
+[cte,actual] = max(ctestd,[],2);
+
+
+%% Predicting Class
+
+prob_estimates_avg = y(:,ind)';
+[predict,predictclass] = max(prob_estimates_avg,[],2);
+ actual1 = [ones(1,17) ones(1,17)*2 ones(1,17)*3];
+% predictclass = predict_label;
+ myConfusionPlot(predictclass',actual1,1,1);
+
+
+%% Plotting Confusion
+
+%  [ind,fpredicted] = max(scores_linear_avg,[],2);
+%  myConfusionPlot(fpredicted',actual,1,1);
+%   [ind,fpredicted] = max(scores_gaussian_avg,[],2);
+%  myConfusionPlot(fpredicted',actual,1,2);
+%   [ind,fpredicted] = max(scores_linear_vector,[],2);
+%  myConfusionPlot(fpredicted',actual,2,1);
+%   [ind,fpredicted] = max(scores_gaussian_vector,[],2);
+%  myConfusionPlot(fpredicted',actual,2,2);
+
+ 
+ %% Plotting ROC
+%  
+%  scores_avg = prob_estimates_avg;
+% scores_vector = prob_estimates_avg;
+
+
+ myPlotRoc(scores_avg,actual1);
+ myPlotRoc(scores_vector,actual);
+ 
+
+ 
+ legend('Max Avg Prob','Speech Vector','Location','southeast');
+ 
+t = 0:.5:1;
+plot(t,t,'--k');
+title(['ROC Curve of Speech Dataset for NN']);
+set(gca,'FontSize',12,'FontWeight','bold')
+hold off
+print('-djpeg', ['ROC_Speech_NN.jpg'], '-r300');
+close all;
+ 
+ 
+  %% Plotting DET
+
+ 
+   scores = scores_avg;
+  target_linear_avg = [scores(1:17,1); scores(18:34,2); scores(35:51,3)];
+  nontarget_linear_avg = [scores(18:51,1); scores(1:17,2); scores(35:51,2); scores(1:34,3)]; 
+ 
+  scores = scores_vector;
+   target_gaussian_avg = [scores(1:13,1); scores(14:34,2); scores(35:51,3)];
+   nontarget_gaussian_avg = [scores(14:51,1); scores(1:13,2); scores(35:51,2); scores(1:34,3)]; 
+
+
+test_data.tar1 = target_linear_avg;
+test_data.non1 = nontarget_linear_avg;
+
+test_data.tar2 = target_gaussian_avg;
+test_data.non2 = nontarget_gaussian_avg;
+
+
+demo_main(test_data,'big');
+title(['DET Curve of Speech Dataset for NN']);
+set(gca,'FontSize',12,'FontWeight','bold')
+print('-djpeg', ['DET_Speech_NN.jpg'], '-r300');
+close all;
+

Multi-layer feed-forward neural network based classifier/Digits/main.m

@@ -0,0 +1,49 @@
+function [  ] = myConfusionPlot( predicttest,op,k,s )
+% Calculates & Plots Confusion Matrix
+
+
+%% Calculate Confusion Matrix
+
+conf=sum(predicttest==op)/length(predicttest);
+disp(['Accuracy of system from Confusion Matrix is : ',num2str(conf*100),' %'])
+
+yu=unique(predicttest);
+confMat=zeros(length(yu));
+for i=1:length(yu)
+    for j=1:length(yu)
+        confMat(i,j)=sum(predicttest==yu(i) & op==yu(j));
+    end
+end
+
+disp(['Confusion :'])
+disp(confMat)
+
+
+%% Plots Confusion Matrix
+figure;
+
+% Convert this data to a [numClasses x length(op)] matrix
+targets = zeros(3,length(op));
+outputs = zeros(3,length(op));
+targetsIdx = sub2ind(size(targets), op, 1:length(op));
+outputsIdx = sub2ind(size(outputs), predicttest, 1:length(op));
+targets(targetsIdx) = 1;
+outputs(outputsIdx) = 1;
+
+% Plot the confusion matrix for a 3-class problem
+plotconfusion(targets,outputs)
+
+h = gca;
+h.XTickLabel = {'Class ai','Class bA','Class lA',''};
+h.YTickLabel = {'Class ai','Class bA','Class lA',''};
+h.FontSize = 12;
+h.FontWeight = 'bold';
+h.YTickLabelRotation = 90;
+title(['Confusion with acc. = ',num2str(conf*100),'% & K = ',num2str(k),' & S = ',num2str(s),],'FontSize',12,'FontWeight','bold')
+set(gca,'FontSize',14,'FontWeight','bold')
+print('-djpeg', ['Confusion_Matrix_K_',num2str(k),'_S_',num2str(s),'.jpg'], '-r300');
+close all;
+% 
+
+end
+