Class names in SVM Leave-one-out cross validation

Hi everyone,

I'm currently working on my Bachelor thesis and trying to do LOO cross validation on a dataset I clustered with kmeans. My data Matrix is a 57x707 double called dyn_post_cluster_final and the Matrix which assigns each patient to a cluster is called Index_cluster{i,j} with i being k from kmeans and j being the number of runs. Index_cluster is a 57x1 double.

I want to do cross validation for each pair of clusters, so in this case k = 2: 1 vs. 2 and later k=3 1 vs. 2, 1 vs. 3 and 2 vs. 3

Here is the code:

k = 1;
num_runs = 1;
classification_rate = zeros(57, 1);

Index_cluster_k = Index_cluster{k, num_runs};

for p = find(Index_cluster_k == 1 | Index_cluster_k == 2)
    testPatient = dyn_post_cluster_final(p, :);
    testCluster = Index_cluster_k(p);

    trainPatients = dyn_post_cluster_final(Index_cluster_k == 1 | Index_cluster_k == 2, :);
    trainPatients(p, :) = [];
    trainClusters = Index_cluster_k(Index_cluster_k == 1 | Index_cluster_k == 2);
    trainClusters(p) = [];

    trainClusters = categorical(trainClusters);
    svmModel = fitcsvm(trainPatients, trainClusters, 'KernelFunction', 'linear', 'ClassNames', [1, 2]);
    predictedCluster = predict(svmModel, testPatient);

    classification_rate(p) = (predictedCluster == testCluster);
end

accuracy = mean(classification_rate(Index_cluster_k == 1 | Index_cluster_k == 2));

I always seem to incounter the same error which is:

Error using ClassificationSVM.prepareData
No class names are found in input labels.

Error in classreg.learning.FitTemplate/fit (line 246)
                    this.PrepareData(X,Y,this.BaseFitObjectArgs{:});

Error in ClassificationSVM.fit (line 243)
            this = fit(temp,X,Y);

Error in fitcsvm (line 342)
    obj = ClassificationSVM.fit(X,Y,RemainingArgs{:});

Error in svm (line 37)
    svmModel = fitcsvm(trainPatients, trainClusters, 'KernelFunction', 'linear', 'ClassNames', [1, 2]);

I hope you can help me.

Thanks!

Answer 1

[Pratyush]

Hi Philip. It is my understanding that you are trying to train an SVM model using fitcsvm method but you get errors related to column names. Since I do not have access to your data, I tried to train an SVM using the fisheriris dataset available on MATLAB in your way. Here is what I think the problem is.

load fisheriris
inds = ~strcmp(species,'setosa');
X = meas(inds,3:4);
y = species(inds);

SVMModel = fitcsvm(X,y,'ClassNames',['versicolor', 'virginica']);

Error using classreg.learning.classif.FullClassificationModel.processClassNames
Requested class names are not found among passed class labels.

Error in ClassificationSVM.prepareData (line 715)
classreg.learning.classif.FullClassificationModel.processClassNames(...

Error in classreg.learning.FitTemplate/fit (line 246)
this.PrepareData(X,Y,this.BaseFitObjectArgs{:});

Error in ClassificationSVM.fit (line 243)
this = fit(temp,X,Y);

Error in fitcsvm (line 342)
obj = ClassificationSVM.fit(X,Y,RemainingArgs{:});

If we try to execute the above cell, we would get similar errors. This happens because the 'ClassNames' attributes expect a cell array containing class names in char array. Below is the correct way to write the above line.

SVMModel = fitcsvm(X,y,'ClassNames',{'versicolor', 'virginica'});

If we change the svmModel line in your code to the following, it may work

svmModel = fitcsvm(trainPatients, trainClusters, 'KernelFunction', 'linear', 'ClassNames', {'class1','class2'});

You may replace class1 and class2 with actual names of your dataset classes. Hope this helps.

Here are some docs to help you understand more on the fitcsvm method.

Train support vector machine classifier