I keep getting a warning and poor classification using fitcecoc.

Hi, I'm wondering if anyone may be able to help with a problem I'm having in classifiying some data?

I'm trying to classify a predictor which is a 4 x 154 double array against a response which is a 80 x 1 double array.

I'm very new to using the machine learning tools and have been trying to follow examples, but get the warning (see below), I'm struggling to see the problem. I wondered if anyone may be able to help or if this is even the best way of classifying data in this format?

Kind regards,

Andy

I'm using the following code:

%Create a classifier that takes as input the condition indicators and returns the combined fault flag. 
%Train a support vector machine that uses a 2nd order polynomial kernel. 
%Use the cvpartition command to partition the ensemble members into a set for training and a set for validation.
 
rng('default') % for reproducibility
predictors = scTrain(:,:); 
response = data.CombinedFlag;
cvp = cvpartition(size(predictors,1),'KFold',5);

% Create and train the classifier
template = templateSVM(...
    'KernelFunction', 'polynomial', ...
    'PolynomialOrder', 2, ...
    'KernelScale', 'auto', ...
    'BoxConstraint', 1, ...
    'Standardize', true);
combinedClassifier = fitcecoc(...
    predictors(cvp.training(1),:), ...
    response(cvp.training(1),:), ...
    'Learners', template, ...
    'Coding', 'onevsone', ...
    'ClassNames', [0; 1; 2; 3; 4; 5; 6; 7]);

% Check performance by computing and plotting the confusion matrix
   actualValue = response(cvp.test(1),:);
   predictedValue = predict(combinedClassifier, predictors(cvp.test(1),:));
   confdata = confusionmat(actualValue,predictedValue);
   
figure,
labels = {'None','Capacitor','Thermal','Bondwire','Bondwire & Capacitor', ...
    'Thermal & Bondwire','Thermal & Capacitor','All'};
  h = heatmap(confdata,...
    'YLabel', 'Actual IGBT fault', ...
    'YDisplayLabels', labels, ...
    'XLabel', 'Predicted IGBT fault', ...
    'XDisplayLabels', labels, ...
    'ColorbarVisible','off');

However, I keep getting the following warning for each learner which in turn leads to no classes being classified.

Warning: The number of folds K is greater than the number of observations N. K will be set to the
value of N. 
> In internal.stats.cvpartitionInMemoryImpl (line 158)
  In cvpartition (line 175)
  In Boost_DT (line 206) 
Warning: Unable to fit learner 1 (SVM) because: No class names are found in input labels. 
> In ClassificationECOC>localFitECOC/loopBody (line 647)
  In internal.stats.parallel.smartForSliceout (line 174)
  In ClassificationECOC>localFitECOC (line 571)
  In ClassificationECOC (line 171)
  In classreg.learning/FitTemplate/fit (line 263)
  In ClassificationECOC.fit (line 116)
  In fitcecoc (line 356)
  In Boost_DT (line 215)

Answer 1

[Prince Kumar]

Hi,

Please have a look at your data first, it seems you data is not in correct format.

You are having 4 x 154 double array, it means you have 4 data point of 154 dimension each. But you are trying to do 5-fold validation.

Please go through k-fold documentation for more information.

Hope this helps!