Error using classreg.l​earning.in​ternal.cla​ssCount (line 31) You passed an unknown class '0.7641' of type double.

I am trying to run the HAR simulink to android example using the "bag" method instead of "adaboostM2", the lines of codes are as:

load humanactivity

rng('default') %for reproducibility
Partition = cvpartition(actid, 'Holdout', 0.10);
trainingInds = training(Partition); % Indices for training set
XTrain = feat(trainingInds,:);
YTrain = actid(trainingInds);
testInds = test(Partition); % Indices for test set
XTest = feat(testInds,:);
YTest = feat(testInds);

tTrain = array2table([XTrain YTrain]);

tTrain.Properties.VariableNames = [featlabels' 'Activities'];

template = templateTree('MaxNumSplits', 20, 'Reproducible', true);
classificationEnsemble = fitcensemble(XTrain, YTrain,...
    'Method', 'bag',...
    'NumLearningCycles', 30,...
    'Learners', template, ...
    'ClassNames',[1;2;3;4;5]);

partitionedModel = crossval(classificationEnsemble, 'KFold', 5);
validationAccuracy = 1-kfoldLoss(partitionedModel)

testAccuracy = 1-loss(classificationEnsemble,XTest,YTest)

I am getting the following errors shile testing:

Error using classreg.learning.internal.classCount (line 31)
You passed an unknown class '0.7641' of type double.

Error in classreg.learning.classif.ClassificationModel/prepareDataForLoss (line 316)
            C = classreg.learning.internal.classCount(this.ClassSummary.ClassNames,Y);

Error in classreg.learning.classif.CompactClassificationEnsemble/loss (line 339)
            [X,C,W] = prepareDataForLoss(this,X,Y,W,[],true,true);

Error in HAR (line 33)
testAccuracy = 1-loss(classificationEnsemble,XTest,YTest)

Please help me debug the code.

Thank you for your time!

Answer 1

[Walter Roberson]

load humanactivity

rng('default') %for reproducibility
Partition = cvpartition(actid, 'Holdout', 0.10);
trainingInds = training(Partition); % Indices for training set
XTrain = feat(trainingInds,:);
YTrain = actid(trainingInds);
testInds = test(Partition); % Indices for test set
XTest = feat(testInds,:);
YTest = actid(testInds);    %<---- this was your error

tTrain = array2table([XTrain YTrain]);

tTrain.Properties.VariableNames = [featlabels' 'Activities'];

template = templateTree('MaxNumSplits', 20, 'Reproducible', true);
classificationEnsemble = fitcensemble(XTrain, YTrain,...
    'Method', 'bag',...
    'NumLearningCycles', 30,...
    'Learners', template, ...
    'ClassNames',[1;2;3;4;5]);

partitionedModel = crossval(classificationEnsemble, 'KFold', 5);
validationAccuracy = 1-kfoldLoss(partitionedModel)

validationAccuracy = 0.9673

testAccuracy = 1-loss(classificationEnsemble,XTest,YTest)

testAccuracy = 0.9613

[Doli Hazarika]

Thanks, I understood the error.

[Doli Hazarika]

After building the model, I am trying to run it with the "slexHARAndroidExample" in simulink but getting the following error: Error(s) encountered while building "slexHARAndroidExample": ### Failed to generate all binary outputs.

I have also copied the calibration matrix file slexHARAndroidCalibrationMatrix.mat to the cwd but still the error persists. Am I missing something here?

[Walter Roberson]

fitcensemble() cannot be deployed to Android (or other hardware).

The workflow for deploying classifiers and approximators, is that you have to do the training with a real MATLAB session first, and you save the classification results to a file. Then you create a new project that load()'s the file and arranges input data as appropriate for the hardware, and uses predict() or whichever function is appropriate for the object class. The training portion cannot take place on the deployed hardware, but the results of training can be used on the hardware to classify or approximate.

[Doli Hazarika]

I understood that part. My problem is when I am trying to run the simulink model given in "slexHARAndroidExample", then it is showing error in calibration block. The error is given in my above answer. So do I need to remove the calibration block from the simlink diagram so as to run the model to an android app?