ASCATEGORICAL must have one element for each variable

Marketing Campaign Value Count Percent 0 2 66.67% 1 1 33.33% Training Set Value Count Percent 0 2 100.00% Test Set Value Count Percent 1 1 100.00%

Error using classreg.regr.FitObject/checkAsCat (line 912)

ASCATEGORICAL must have one element for each variable.

Error in classreg.regr.FitObject/assignData (line 314) viIsCategorical = classreg.regr.FitObject.checkAsCat(viIsCategorical,asCat,nvars,false,viName);

Error in classreg.regr.TermsRegression/assignData (line 349) model = [e-mail](model,X,y,w,asCat,varNames,excl);

Error in GeneralizedLinearModel/assignData (line 800) model = [e-mail](model,X,y,w,asCat,dummyCoding,varNames,excl);

Error in GeneralizedLinearModel.fit (line 1297) model = assignData(model,X,y,weights,offset,binomN,asCatVar,dummyCoding,model.Formula.VariableNames,exclude);

Error in crater (line 235) glm = GeneralizedLinearModel.fit(Xtrain,double(Ytrain)-1,'linear','Distribution','binomial','link','logit','CategoricalVars',catPred);

if true
  % code

clc clear all

% Load Datasets

Dataset = 'F:\Mtech\Mtech-3rd\Research\Matlab\CDA\Machine_Learning\data'; Testset = 'F:\Mtech\Mtech-3rd\Research\Matlab\CDA\Machine_Learning\data';

% we need to process the images first. % Convert your images into grayscale % Resize the images

width=100; height=100; DataSet = cell([], 1);

 for i=1:length(dir(fullfile(Dataset,'*.jpg')))

     % Training set process
     k = dir(fullfile(Dataset,'*.jpg'));
     k = {k(~[k.isdir]).name};
     for j=1:length(k)
        tempImage       = imread(horzcat(Dataset,filesep,k{j})); 
        imgInfo         = imfinfo(horzcat(Dataset,filesep,k{j}));

         % Image transformation
         if strcmp(imgInfo.ColorType,'grayscale')
%            tempfeaturepoints   = detectSURFFeatures(ReadImage);  
%            [f,vpts]            = extractFeatures(ReadImage, %tempfeaturepoints);
%            DataSet{j}          = f;
             DataSet{j}   = double(imresize(tempImage,[width height])); % array of images
         else
%              tempImage           = rgb2gray(ReadImage);
%              tempfeaturepoints   = detectSURFFeatures(tempImage);
%              [f,vpts]            = extractFeatures(tempImage, tempfeaturepoints);
%              DataSet{j}          = f;
            DataSet{j}   = double(imresize(rgb2gray(tempImage),[width height])); % array of images
         end
     end
 end
TestSet =  cell([], 1);
  for i=1:length(dir(fullfile(Testset,'*.jpg')))

     % Training set process
     k = dir(fullfile(Testset,'*.jpg'));
     k = {k(~[k.isdir]).name};
     for j=1:length(k)
        tempImage       = imread(horzcat(Testset,filesep,k{j}));
        imgInfo         = imfinfo(horzcat(Testset,filesep,k{j}));

         % Image transformation
         if strcmp(imgInfo.ColorType,'grayscale')
            TestSet{j}   = double(imresize(tempImage,[width height])); % array of images
         else
            TestSet{j}   = double(imresize(rgb2gray(tempImage),[width height])); % array of images
         end
     end
  end

% Prepare class label for first run of svm % I have arranged labels 1 & 2 as per my convenience. % It is always better to label your images numerically % Please note that for every image in our Dataset we need to provide one label. % we have 30 images and we divided it into two label groups here. train_label = zeros(size(30,1),1); train_label(1:15,1) = 1; % 1 = Airplanes train_label(16:30,1) = 2; % 2 = Faces

% Prepare numeric matrix for svmtrain Training_Set=[]; for i=1:length(DataSet) Training_Set_tmp = reshape(DataSet{i},1, 100*100); Training_Set=[Training_Set;Training_Set_tmp]; end

Test_Set=[]; for j=1:length(TestSet) Test_set_tmp = reshape(TestSet{j},1, 100*100); Test_Set=[Test_Set;Test_set_tmp]; end bank=Training_Set; Response=zeros(3,1); for i=1:length(Response) if(i<2) Response(i,1)=1; %1=crater% else Response(i,1)=0; %0=non-crater% end end

%% Prepare the Data: Response and Predictors % We can segregate the data into response and predictors. This will make it % easier to call subsequent functions which expect the data in this format.

% Response Y = Response; disp('Marketing Campaign') tabulate(Y) % Predictor matrix X = double(bank(:,1:end-1));

%% Category Predictor

S = ['rays ';'peak ';'walls']; C = cellstr(S); [bank_row,bank_col]= size(bank); p=1;

for i=1:bank_row if(p>3) p=1; end crater_attrib{i}= C{p}; p=p+1; end attribute=[crater_attrib]';

%% category predictor %bank=cell2dataset(bank); catPred=false(bank_row,1); for i=1:bank_row if(strcmp(bank(i,1),C(1)) strcmp(bank(i,1),C(3))) catPred(i)=true; end end bank=(bank)'; catPred=(catPred)'; %% Adding category to data

bank=vertcat(crater_attrib, num2cell(bank));

%% Cross Validation

% Cross validation is almost an inherent part of machine learning. Cross % validation may be used to compare the performance of different predictive % modeling techniques. In this example, we use holdout validation. Other % techniques including k-fold and leave-one-out cross validation are also % available. % % In this example, we partition the data into training set and test set. % The training set will be used to calibrate/train the model parameters. % The trained model is then used to make a prediction on the test set. % Predicted values will be compared with actual data to compute the % confusion matrix. Confusion matrix is one way to visualize the % performance of a machine learning technique.

% In this example, we will hold 40% of the data, selected randomly, for % test phase. cv = cvpartition(length(bank(1, 1:end)),'holdout',0.40); Vtest=training(cv); % Training set Xtrain = X(training(cv),:); Ytrain = Y(training(cv),:); % Test set Xtest = X(test(cv),:); Ytest = Y(test(cv),:);

disp('Training Set') tabulate(Ytrain) disp('Test Set') tabulate(Ytest) % % %% Prepare Predictors/Response for Neural Networks % % When using neural networks the appropriate way to include categorical % % predictors is as dummy indicator variables. An indicator variable has % % values 0 and 1. % % [XtrainNN, YtrainNN, XtestNN, YtestNN] = preparedataNN(bank, catPred, cv); % [XtrainNN, YtrainNN, XtestNN, YtestNN] = preparedataNN(bank, cv);

% %% Speed up Computations using Parallel Computing % % If Parallel Computing Toolbox is available, the computation will be % % distributed to 2 workers for speeding up the evaluation. % % if matlabpool('size') == 0 % matlabpool open 2 % end % % %% Neural Networks % % Neural Network Toolbox supports supervised learning with feedforward, % % radial basis, and dynamic networks. It supports both classification and % % regression algorithms. It also supports unsupervised learning with % % self-organizing maps and competitive layers. % % % % One can make use of the interactive tools to setup, train and validate a % % neural network. It is then possible to auto-generate the code for the % % purpose of automation. In this example, the auto-generated code has been % % updated to utilize a pool of workers, if available. This is achieved by % % simply setting the useParallel flag while making a call to train. % % % % [net,~] = train(net,inputs,targets,'useParallel','yes'); % % % % If a GPU is available, it may be utilized by setting the useGPU flag. % % % % The trained network is used to make a prediction on the test data and % % confusion matrix is generated for comparison with other techniques. % % % Use modified autogenerated code to train the network % [~, net] = NNfun(XtrainNN,YtrainNN); % % % Make a prediction for the test set % Y_nn = net(XtestNN'); % Y_nn = round(Y_nn'); % % % Compute the confusion matrix % C_nn = confusionmat(YtestNN,Y_nn); % % Examine the confusion matrix for each class as a percentage of the true class % C_nn = bsxfun(@rdivide,C_nn,sum(C_nn,2)) * 100 %#ok<*NOPTS>

% %% Other Machine Learning Techniques % % Statistics Toolbox features a number of supervised and unsupervised % % machine learning techniques. It supports both classification and % % regression algorithms. The supervised learning techniques range from % % non-linear regression, generalized linear regression, discriminant % % analysis, SVMs to decision trees and ensemble methods. % % % % In this example, we make use of some of these techniques to perform % % predictive modeling. Observe that once the data has been prepared, the % % syntax to utilize the different modeling techniques is very similar. Most % % of these techniques can handle categorical predictors. The user can % % conveniently supply information about different parameters associated % % with the different algorithms. % % %% Generalized Linear Model - Logistic Regression % % In this example, a logistic regression model is leveraged. Response may % % follow normal, binomial, Poisson, gamma, or inverse Gaussian % % distribution. % % % % Since the response in this data set is binary, binomial distribution is % % suitable. % % % Train the classifier glm = GeneralizedLinearModel.fit(Xtrain,double(Ytrain)-1,'linear','Distribution','binomial','link','logit','CategoricalVars',catPred);

% Make a prediction for the test set Y_glm = glm.predict(Xtest); Y_glm = round(Y_glm) + 1;

% Compute the confusion matrix C_glm = confusionmat(double(Ytest),Y_glm); % Examine the confusion matrix for each class as a percentage of the true class C_glm = bsxfun(@rdivide,C_glm,sum(C_glm,2)) * 100

%% Discriminant Analysis % Discriminant analysis is a classification method. It assumes that % different classes generate data based on different Gaussian % distributions. Linear discriminant analysis is also known as the Fisher % discriminant. % % Here, a quadratic discriminant classifier is used.

% Train the classifier da = ClassificationDiscriminant.fit(Xtrain,Ytrain,'discrimType','linear');

% Make a prediction for the test set Y_da = da.predict(Xtest);

% Compute the confusion matrix C_da = confusionmat(Ytest,Y_da); % Examine the confusion matrix for each class as a percentage of the true class C_da = bsxfun(@rdivide,C_da,sum(C_da,2)) * 100

%% Classification Using Nearest Neighbors % Categorizing query points based on their distance to points in a training % dataset can be a simple yet effective way of classifying new points. % Various distance metrics such as euclidean, correlation, hamming, % mahalonobis or your own distance metric may be used.

% Train the classifier knn = ClassificationKNN.fit(Xtrain,Ytrain,'Distance','seuclidean');

% Make a prediction for the test set Y_knn = knn.predict(Xtest);

% Compute the confusion matrix C_knn = confusionmat(Ytest,Y_knn); % Examine the confusion matrix for each class as a percentage of the true class C_knn = bsxfun(@rdivide,C_knn,sum(C_knn,2)) * 100 % % %% Naive Bayes Classification % % Naive Bayes classification is based on estimating P(X|Y), the probability % % or probability density of features X given class Y. The Naive Bayes % % classification object provides support for normal (Gaussian), kernel, % % multinomial, and multivariate multinomial distributions % % % The multivariate multinomial distribution (mvmn) is appropriate for % % categorical features dist = repmat({'normal'},1,bank_col-1); dist(catPred) = {'mvmn'};

% Train the classifier Nb = NaiveBayes.fit(Xtrain,Ytrain,'Distribution',dist);

% Make a prediction for the test set Y_Nb = Nb.predict(Xtest);

% Compute the confusion matrix C_nb = confusionmat(Ytest,Y_Nb); % Examine the confusion matrix for each class as a percentage of the true class C_nb = bsxfun(@rdivide,C_nb,sum(C_nb,2)) * 100

%% Support Vector Machines % Support vector machine (SVM) is supported for binary response variables. % An SVM classifies data by finding the best hyperplane that separates all % data points of one class from those of the other class.

opts = statset('MaxIter',30000); % Train the classifier svmStruct = svmtrain(Xtrain,Ytrain,'kernel_function','rbf','kktviolationlevel',0.1,'options',opts);

% Make a prediction for the test set Y_svm = svmclassify(svmStruct,Xtest); C_svm = confusionmat(Ytest,Y_svm); % Examine the confusion matrix for each class as a percentage of the true class C_svm = bsxfun(@rdivide,C_svm,sum(C_svm,2)) * 100

% %% Decision Trees % % Classification trees and regression trees are two kinds of decision % % trees. A decision tree is a flow-chart like structure in which internal % % node represents test on an attribute, each branch represents outcome of % % test and each leaf node represents a response (decision taken after % % computing all attributes). Classification trees give responses that are % % nominal, such as 'true' or 'false'. Regression trees give numeric % % responses. % % tic % % Train the classifier % t = ClassificationTree.fit(Xtrain,Ytrain,'CategoricalPredictors',catPred); % toc % % % Make a prediction for the test set % Y_t = t.predict(Xtest); % % % Compute the confusion matrix % C_t = confusionmat(Ytest,Y_t); % % Examine the confusion matrix for each class as a percentage of the true class % C_t = bsxfun(@rdivide,C_t,sum(C_t,2)) * 100

% %% Ensemble Learning: TreeBagger % % Bagging stands for bootstrap aggregation. Every tree in the ensemble is % % grown on an independently drawn sample of input data. To compute % % prediction for the ensemble of trees, TreeBagger takes an average of % % predictions from individual trees (for regression) or takes votes from % % individual trees (for classification). Ensemble techniques such as % % bagging combine many weak learners to produce a strong learner. % % % % From a marketing perspective, as we are creating this predictive model, % % it may be more important for us to classify yes correctly than a % % no. If that is the case, we can include our opinion using the cost % % matrix. Here, cost matrix specifies that it is 5 times more costly to % % classify a yes as a no.

% Cost of misclassification % cost = [0 1 % 5 0]; % opts = statset('UseParallel',true); % % Train the classifier % tb = TreeBagger(150,Xtrain,Ytrain,'method','classification','categorical',catPred,'Options',opts,'OOBVarImp','on','cost',cost); % % % Make a prediction for the test set % [Y_tb, classifScore] = tb.predict(Xtest); % Y_tb = nominal(Y_tb); % % % Compute the confusion matrix % C_tb = confusionmat(Ytest,Y_tb); % % Examine the confusion matrix for each class as a percentage of the true class % C_tb = bsxfun(@rdivide,C_tb,sum(C_tb,2)) * 100 % % %% Compare Results % % This visualization function is making use of a couple files downloaded % % from MATLAB Central, the user % % community website. We are leveraging social computing along the way to % % help us in our effort. % % Cmat = [C_nn C_glm C_da C_knn C_nb C_svm C_t C_tb]; % labels = {'Neural Net ', 'Logistic Regression ', 'Discriminant Analysis ',... % 'k-nearest Neighbors ', 'Naive Bayes ', 'Support VM ', 'Decision Trees ', 'TreeBagger '}; Cmat = [ C_da C_knn C_nb C_svm ]; labels = { 'Discriminant Analysis ',... 'k-nearest Neighbors ', 'Naive Bayes ', 'Support VM '};

comparisonPlot( Cmat, labels )

%% ROC Curve for Classification by TreeBagger % Another way of exploring the performance of a classification ensemble is % to plot its Receiver Operating Characteristic (ROC) curve.

% [xx,yy,~,auc] = perfcurve(Ytest, classifScore(:,2),'yes'); % figure; % plot(xx,yy) % xlabel('False positive rate'); % ylabel('True positive rate') % title('ROC curve for ''yes'', predicted vs. actual response (Test Set)') % text(0.5,0.25,{'TreeBagger with full feature set',strcat('Area Under Curve = ',num2str(auc))},'EdgeColor','k');

% %% Simplify Model - Optional % % One may choose to examine the models further. One may even be able to % % improve the performance of the models. It is also possible to estimate % % importance of the different features, reduce the dimensionality of % % feature set etc. % % %% Estimating a Good Ensemble Size % % Examining the out-of-bag error may give an insight into determining a % % good ensemble size. % % figure; % plot(oobError(tb)); % xlabel('Number of Grown Trees'); % ylabel('Out-of-Bag Classification Error/Misclassification Probability'); % % %% Estimating Feature Importance % % Feature importance measures the increase in prediction error if the % % values of that variable are permuted across the out-of-bag observations. % % This measure is computed for every tree, then averaged over the entire % % ensemble and divided by the standard deviation over the entire ensemble. % % figure; % bar(tb.OOBPermutedVarDeltaError); % ylabel('Out-Of-Bag Feature Importance'); % set(gca,'XTick',1:16) % names2 = names; % names2{5} = ' default'; % set(gca,'XTickLabel',names2) % % Use file submitted from a user at MATLAB Central to rotate labels % rotateXLabels( gca, 60 ) % [~,idxvarimp] = sort(tb.OOBPermutedVarDeltaError, 'descend'); % % %% Sequential Feature Selection % % Feature selection reduces the dimensionality of data by selecting only a % % subset of measured features (predictor variables) to create a model. % % Selection criteria involves the minimization of a specific measure of % % predictive error for models fit to different subsets. % % % % Sequential feature selection can be computationally intensive. It can % % benefit significantly from parallel computing. % % opts = statset('UseParallel',true); % critfun = @(Xtr,Ytr,Xte,Yte)featureImp(Xtr,Ytr,Xte,Yte,'TreeBagger'); % % The top 5 features determined in the previous step have been included, % % to reduce the number of combinations to be tried by sequentialfs % [fs,history] = sequentialfs(critfun,Xtrain,Ytrain,'options',opts,'keepin',idxvarimp(1:5)); % disp('Included features:'); % disp(names(fs)'); % % %% TreeBagger with Reduced Feature Set % % opts = statset('UseParallel',true); % tb_r = TreeBagger(120, Xtrain(:,fs),Ytrain,'method','classification','categorical',catPred(:,fs),'Options',opts,'cost',cost); % [Y_tb_r, classifScore] = tb_r.predict(Xtest(:,fs)); % Y_tb_r = nominal(Y_tb_r); % C_tb_r = confusionmat(Ytest,Y_tb_r); % C_tb_r = bsxfun(@rdivide,C_tb_r,sum(C_tb_r,2)) * 100 % % %% Compare Results % % Cmat = [C_nn C_glm C_da C_knn C_nb C_svm C_t C_tb C_tb_r]; % labels = {'Neural Net ', 'Logistic Regression ', 'Discriminant Analysis ',... % 'k-nearest Neighbors ', 'Naive Bayes ', 'Support VM ', 'Decision Trees ', 'TreeBagger ', 'Reduced TB '}; % % comparisonPlot( Cmat, labels ) % % %% ROC Curve for Classification by Reduced TreeBagger % % [xx,yy,~,auc] = perfcurve(Ytest, classifScore(:,2),'yes'); % figure; % plot(xx,yy) % xlabel('False positive rate'); % ylabel('True positive rate') % title('ROC curve for ''yes'', predicted vs. actual response (Test Set)') % text(0.5,0.25,{'TreeBagger with reduced feature set',strcat('Area Under Curve = ',num2str(auc))},'EdgeColor','k'); % % %% Shut Down Workers % % Release the workers if there is no more work for them % % if matlabpool('size') > 0 % matlabpool close % end % %% References % % # [Moro et al., 2011] S. Moro, R. Laureano and P. Cortez. Using Data Mining for Bank Direct Marketing: An Application of the CRISP-DM Methodology. % % In P. Novais et al. (Eds.), Proceedings of the European Simulation and Modelling Conference - ESM'2011, pp. 117-121, Guimarães, Portugal, October, 2011. EUROSIS.

    %code
  end

[Greg Heath]

Please

1. Explain your problem.

2. Please clarify what your inputs and outputs are.

3. Please format