High-Dimensional Feature Selection of Medical Data

Optimized LASSO feature selection

High-Dimensional Feature Selection of Medical Data/Code/FeatureSelect1.m

@@ -0,0 +1,32 @@
+%% Read dataset and preprocessing
+x=csvread('train_data1.csv',0,0,[0 0 699 128]);
+y=csvread('train_data1.csv',0,129,[0 129 699 140]);
+y(y(:,:)==-1) = 0;
+
+%% T test
+pSum = zeros(1,129);
+for i=1:12
+    nowY = y(:,i);
+    X0 = x(nowY==0,:);
+    X1 = x(nowY==1,:);
+    [~,p,~,~] = ttest2(X0,X1,'Vartype','unequal');
+    pSum = pSum+p;
+end
+
+[~,featureIdxSortbyP]= sort(pSum);
+x=x(:,featureIdxSortbyP(1:50));
+
+%% LASSO regression feature selection
+opts = statset('UseParallel',true);
+featureSum = zeros(12,1);
+featureWeight = zeros(50,1);
+for i=1:12
+    [B,S] = lassoglm(x,y(:,i),'binomial','DFmax',30,'CV',10,'Alpha',0.5,'Options',opts);
+    featureSum(i,1) = sum(B(:,S.IndexMinDeviance)~=0);
+    featureWeight = featureWeight+B(:,S.IndexMinDeviance);
+end
+
+%% Filter
+sum=floor(sum(featureSum)/12);
+[~,featureIdxSortbyLasso]= sort(featureWeight);
+x=x(:,featureIdxSortbyLasso(1:sum));

High-Dimensional Feature Selection of Medical Data/Code/Question1.m

@@ -0,0 +1,59 @@
+%% Read dataset
+x=csvread('train_data1.csv',0,0,[0 0 699 128]);
+y=csvread('train_data1.csv',0,129,[0 129 699 140]);
+testX = csvread('test_feature_data1.csv',0,1);
+number = csvread('test_feature_data1.csv',0,0,[0 0 203 0]);
+
+%% Preprocessing
+y(y(:,:)==-1) = 0;
+net = patternnet(15);
+
+net.trainFcn = 'trainrp';
+net.trainParam.max_fail = 20;
+net.trainParam.epochs = 1000;
+
+net.divideParam.trainRatio = 70/100;
+net.divideParam.valRatio = 15/100;
+net.divideParam.testRatio = 15/100;
+
+results = zeros(204,13);
+results(:,1)=number;
+errors = zeros(1,700);
+
+%% Divide into 12 sub problems
+for i=1:12
+
+    % T Test feature selection
+    nowY = y(:,i);
+    X0 = x(nowY==0,:);
+    X1 = x(nowY==1,:);
+    [~,p,~,~] = ttest2(X0,X1,'Vartype','unequal');
+    [~,featureIdxSortbyP]= sort(p);
+    nowX = x(:,featureIdxSortbyP(1:70));
+
+    % LASSO regression feature selection
+    [B,S] = lassoglm(nowX,nowY,'binomial','DFmax',30,'CV',10,'Alpha',0.5);
+    model = B(:,S.IndexMinDeviance)~=0;
+    nowX = nowX(:,model);
+
+    nowTestX = testX(:,featureIdxSortbyP(1:70));
+    nowTestX = nowTestX(:,model);
+
+    % Train and use NN array
+    outputs = zeros(1,204);
+    for j=1:5
+        rand = randperm(700);
+        nowX = nowX(rand,:);
+        nowY = nowY(rand,:);
+        nowNet = train(net,nowX',nowY');
+        errors = errors+gsubtract(nowY',nowNet(nowX'));
+        outputs = outputs+nowNet(nowTestX');
+    end
+
+    % Generate result
+    results(:,i+1) = outputs'/5;
+end
+
+%% Output result
+disp(sum(abs(errors))/(700*60));
+csvwrite('result1.csv',results);

High-Dimensional Feature Selection of Medical Data/Code/Question2.m

@@ -0,0 +1,86 @@
+%% Read dataset and preprocessing
+x=csvread('train_data2.csv',0,0,[0 0 3999 409]);
+y=csvread('train_data2.csv',0,410,[0 410 3999 410]);
+X1 = x(y==0,:);
+X2 = x(y==1,:);
+X3 = x(y==2,:);
+
+%% T Test
+[~,p12,~,~] = ttest2(X1,X2,'Vartype','unequal');
+[~,p13,~,~] = ttest2(X2,X3,'Vartype','unequal');
+[~,p23,~,~] = ttest2(X1,X3,'Vartype','unequal');
+p=p12+p13+p23;
+[~,featureIdxSortbyP]= sort(p);
+
+%% LASSO regression feature selection
+
+x=x(:,featureIdxSortbyP(1:100));
+opts = statset('UseParallel',true);
+% Linear Regression
+[B,S] = lasso(x,y,'DFmax',50,'CV',10,'Alpha',0.5,'Options',opts);
+% Poisson Regression
+% [B,S] = lassoglm(x,y,'poisson','DFmax',50,'CV',10,'Alpha',0.5,'Options',opts);
+model = B(:,S.Index1SE)~=0;
+x=x(:,model);
+
+clear X1 X2 X3 p p12 p13 p23 B S
+
+%% Process class label
+newY=zeros(4000,3);
+newY(y==0,1)=1;
+newY(y==1,2)=1;
+newY(y==2,3)=1;
+y=newY;
+clear newY;
+
+%% Build NN parameters
+neurons = sum(model~=0);
+net = patternnet(floor(neurons/2));
+net.divideParam.trainRatio = 70/100;
+net.divideParam.valRatio = 15/100;
+net.divideParam.testRatio = 15/100;
+
+net.trainFcn = 'trainrp';
+net.trainParam.max_fail = 20;
+net.trainParam.epochs = 1000;
+
+%% Train NN array
+errorsArray=zeros(1,20);
+nets = cell(1,20);
+
+for i=1:20
+    % Randomly disorganize sample order
+    rand = randperm(4000);
+    x = x(rand,:);
+    y = y(rand,:);
+
+    nets{i} = train(net,x',y');
+    neti=nets{i};
+    outputs = neti(x');
+    errors = gsubtract(y',outputs);
+    errorsArray(i) = sum(sum(abs(errors)))/12000;
+end
+
+%% Select 10 best NN by error information
+
+[~,IdxSortbyerrors]= sort(errorsArray,'ascend');
+nets = nets(IdxSortbyerrors(1:10));
+disp(mean(errorsArray(IdxSortbyerrors(1:10))));
+clear errors errorsArray i net neti neurons opts outputs rand IdxSortbyerrors
+
+%% Read test data
+testX = csvread('test_feature_data2.csv',0,1);
+number = csvread('test_feature_data2.csv',0,0,[0 0 1030 0]);
+testX = testX(:,featureIdxSortbyP(1:100));
+testX = testX(:,model);
+
+%% Classify by NN
+testY = zeros(3,1031);
+for i=1:10
+    neti = nets{i};
+    testY = testY+neti(testX');
+end
+testY = testY/10;
+testY = (vec2ind(testY)-1)';
+testY = [number,testY];
+csvwrite('result2.csv',testY);

High-Dimensional Feature Selection of Medical Data/Code/lasso.m

@@ -0,0 +1,12 @@
+%% Read dataset
+x=csvread('train_data2.csv',0,0,[0 0 3999 409]);
+y=csvread('train_data2.csv',0,410,[0 410 3999 410]);
+
+%% LASSO regression feature selection 
+
+opts = statset('UseParallel',true);
+% Linear Regression
+[B,S] = lasso(x,y,'DFmax',100,'CV',10,'Alpha',0.5,'Options',opts);
+% Poisson Regression
+[B,S] = lassoglm(x,y,'poisson','DFmax',100,'CV',10,'Alpha',0.5,'Options',opts);
+model = B(:,S.Index1SE)~=0;

High-Dimensional Feature Selection of Medical Data/Code/lassoSelect.m

@@ -0,0 +1,26 @@
+%% Read dataset and preprocessing
+x=csvread('train_data2.csv',0,0,[0 0 3999 409]);
+y=csvread('train_data2.csv',0,410,[0 410 3999 410]);
+X1 = x(y==0,:);
+X2 = x(y==1,:);
+X3 = x(y==2,:);
+
+%% T test
+[~,p12,~,~] = ttest2(X1,X2,'Vartype','unequal');
+[~,p13,~,~] = ttest2(X2,X3,'Vartype','unequal');
+[~,p23,~,~] = ttest2(X1,X3,'Vartype','unequal');
+p=p12+p13+p23;
+[~,featureIdxSortbyP]= sort(p);
+
+%% LASSO regression feature selection
+
+x=x(:,featureIdxSortbyP(1:100));
+opts = statset('UseParallel',true);
+% Linear Regression
+[B,S] = lasso(x,y,'DFmax',100,'CV',10,'Alpha',0.5,'Options',opts);
+% Poisson Regression
+% [B,S] = lassoglm(x,y,'poisson','DFmax',100,'CV',10,'Alpha',0.5,'Options',opts);
+model = B(:,S.Index1SE)~=0;
+x=x(:,model);
+
+%% SVM

High-Dimensional Feature Selection of Medical Data/Code/useNN.m

@@ -0,0 +1,14 @@
+%% Read dataset
+
+testX = csvread('test_feature_data2.csv',0,1);
+number = csvread('test_feature_data2.csv',0,0,[0 0 1030 0]);
+testX = testX(:,featureIdxSortbyP(1:100));
+testX = testX(:,model);
+
+%% Neural Network
+testY = net(testX');
+testY = (vec2ind(testY)-1)';
+testY = [number,testY];
+
+%% Outpur data
+csvwrite('result2.csv',testY);

High-Dimensional Feature Selection of Medical Data/Data/example_task/example_task1.csv

@@ -0,0 +1,50 @@
+1,0.82,0.53,0.81,0.77,0.12,0.53,0.81,0.47,0.12,0.53,0.83,0.11
+2,0.72,0.53,0.81,0.71,0.1,0.51,0.74,0.77,0.72,0.93,0.83,0.11
+3,0.68,0.53,0.81,0.77,0.12,0.52,0.81,0.37,0.12,0.53,0.83,0.11
+4,0.62,0.53,0.83,0.77,0.12,0.53,0.63,0.27,0.72,0.63,0.83,0.11
+5,0.21,0.53,0.81,0.77,0.12,0.56,0.81,0.77,0.12,0.53,0.83,0.11
+6,0.12,0.53,0.81,0.77,0.11,0.53,0.22,0.17,0.92,0.43,0.83,0.11
+7,0.22,0.53,0.82,0.76,0.12,0.53,0.29,0.67,0.12,0.53,0.83,0.11
+8,0.12,0.53,0.81,0.77,0.14,0.58,0.71,0.37,0.12,0.73,0.83,0.11
+9,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.83,0.11
+10,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.83,0.11
+11,0.82,0.53,0.81,0.77,0.12,0.53,0.81,0.47,0.12,0.53,0.83,0.11
+12,0.72,0.53,0.81,0.71,0.1,0.51,0.74,0.77,0.72,0.93,0.83,0.11
+13,0.68,0.53,0.81,0.77,0.12,0.52,0.81,0.37,0.12,0.53,0.83,0.11
+14,0.62,0.53,0.83,0.77,0.12,0.53,0.63,0.27,0.72,0.63,0.83,0.11
+15,0.21,0.53,0.81,0.77,0.12,0.56,0.81,0.77,0.12,0.53,0.83,0.11
+16,0.12,0.53,0.81,0.77,0.11,0.53,0.22,0.17,0.92,0.43,0.83,0.11
+17,0.22,0.53,0.82,0.76,0.12,0.53,0.29,0.67,0.12,0.53,0.83,0.11
+18,0.12,0.53,0.81,0.77,0.14,0.58,0.71,0.37,0.12,0.73,0.83,0.11
+19,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.83,0.11
+20,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.83,0.11
+21,0.82,0.53,0.81,0.77,0.12,0.53,0.81,0.47,0.12,0.53,0.83,0.11
+22,0.72,0.53,0.81,0.71,0.1,0.51,0.74,0.77,0.72,0.93,0.75,0.11
+23,0.68,0.53,0.81,0.77,0.12,0.52,0.81,0.37,0.12,0.53,0.75,0.11
+24,0.62,0.53,0.83,0.77,0.12,0.53,0.63,0.27,0.72,0.63,0.75,0.11
+25,0.21,0.53,0.81,0.77,0.12,0.56,0.81,0.77,0.12,0.53,0.75,0.11
+26,0.12,0.53,0.81,0.77,0.11,0.53,0.22,0.17,0.92,0.43,0.75,0.11
+27,0.22,0.53,0.82,0.76,0.12,0.53,0.29,0.67,0.12,0.53,0.75,0.11
+28,0.12,0.53,0.81,0.77,0.14,0.58,0.71,0.37,0.12,0.73,0.75,0.11
+29,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.75,0.11
+30,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.75,0.11
+31,0.82,0.53,0.81,0.77,0.12,0.53,0.81,0.47,0.12,0.53,0.75,0.11
+32,0.72,0.53,0.81,0.71,0.1,0.51,0.74,0.77,0.72,0.93,0.75,0.11
+33,0.68,0.53,0.81,0.77,0.12,0.52,0.81,0.37,0.12,0.53,0.75,0.11
+34,0.62,0.53,0.83,0.77,0.12,0.53,0.63,0.27,0.72,0.63,0.75,0.11
+35,0.21,0.53,0.81,0.77,0.12,0.56,0.81,0.77,0.12,0.53,0.75,0.11
+36,0.12,0.53,0.81,0.77,0.11,0.53,0.22,0.17,0.92,0.43,0.75,0.11
+37,0.22,0.53,0.82,0.76,0.12,0.53,0.29,0.67,0.12,0.53,0.75,0.11
+38,0.12,0.53,0.81,0.77,0.14,0.58,0.71,0.37,0.12,0.73,0.64,0.11
+39,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.64,0.11
+40,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.64,0.11
+41,0.82,0.53,0.81,0.77,0.12,0.53,0.81,0.47,0.12,0.53,0.64,0.11
+42,0.72,0.53,0.81,0.71,0.1,0.51,0.74,0.77,0.72,0.93,0.64,0.11
+43,0.68,0.53,0.81,0.77,0.12,0.52,0.81,0.37,0.12,0.53,0.64,0.11
+44,0.62,0.53,0.83,0.77,0.12,0.53,0.63,0.27,0.72,0.63,0.64,0.11
+45,0.21,0.53,0.81,0.77,0.12,0.56,0.81,0.77,0.12,0.53,0.64,0.11
+46,0.12,0.53,0.81,0.77,0.11,0.53,0.22,0.17,0.92,0.43,0.64,0.11
+47,0.22,0.53,0.82,0.76,0.12,0.53,0.29,0.67,0.12,0.53,0.64,0.11
+48,0.12,0.53,0.81,0.77,0.14,0.58,0.71,0.37,0.12,0.73,0.64,0.11
+49,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.64,0.11
+50,0.92,0.53,0.81,0.77,0.12,0.53,0.81,0.87,0.12,0.83,0.64,0.11

High-Dimensional Feature Selection of Medical Data/Data/example_task/example_task2.csv

@@ -0,0 +1,100 @@
+1,1
+2,1
+3,1
+4,0
+5,1
+6,0
+7,2
+8,1
+9,0
+10,2
+11,1
+12,0
+13,1
+14,0
+15,1
+16,1
+17,1
+18,1
+19,1
+20,1
+21,1
+22,0
+23,1
+24,0
+25,2
+26,1
+27,0
+28,0
+29,2
+30,1
+31,1
+32,2
+33,1
+34,1
+35,1
+36,1
+37,1
+38,1
+39,1
+40,1
+41,1
+42,1
+43,1
+44,1
+45,1
+46,2
+47,2
+48,0
+49,1
+50,1
+51,1
+52,1
+53,0
+54,1
+55,1
+56,1
+57,0
+58,1
+59,0
+60,0
+61,2
+62,1
+63,1
+64,1
+65,1
+66,0
+67,1
+68,1
+69,0
+70,0
+71,1
+72,1
+73,2
+74,1
+75,0
+76,1
+77,1
+78,1
+79,1
+80,0
+81,0
+82,1
+83,1
+84,0
+85,1
+86,1
+87,2
+88,2
+89,1
+90,1
+91,2
+92,0
+93,1
+94,1
+95,2
+96,0
+97,1
+98,1
+99,1
+100,1