Error using classreg.l​earning.Fu​llClassifi​cationRegr​essionMode​l.prepareD​ataCR (line 192) X must be a numeric matrix.

I am working on a deep learning Matlab simulation, this is my code:

********************************************

clear all

clc

% Load Alexnet for CNN convnet = alexnet;

% Setup training data

rootFolder = 'photo';

categories ={'BigSmile','Smile','Neutral'};

imds = imageDatastore(fullfile(rootFolder, categories), 'LabelSource', 'foldernames');

[imdsTrain, imdsTest] = splitEachLabel(imds, 0.8, 'randomize');

imageSize = [277 277 3]; augimdsTrain = augmentedImageDatastore(imageSize(1:2), imdsTrain, 'ColorPreprocessing', 'gray2rgb');

augimdsTest = augmentedImageDatastore(imageSize(1:2),imdsTest, 'ColorPreprocessing', 'gray2rgb');

% Extract features from the training images layer = 'fc7';

featuresTrain = activations(convnet,augimdsTrain,layer, 'OutputAs', 'channels');

featuresTest = activations(convnet,augimdsTest,layer, 'OutputAs', 'channels');

YTrain = imdsTrain.Labels;

YTest = imdsTest.Labels;

% Train the SVM classifier classifier = fitcecoc(featuresTrain,YTrain);

YPred = predict(classifier,featuresTest);

accuracy = mean(YPred == YTest);

**********************************************

I got an error: Error using classreg.learning.FullClassificationRegressionModel.prepareDataCR (line 192) X must be a numeric matrix. Error in example2 (line 44) classifier = fitcecoc(featuresTrain,YTrain);

why do I get this error and how should I fix it?

Answer 1

[Khadije El Zein]

change 'OutputAs', 'channels' to 'OutputAs', 'rows'. worked for me.

[Johanna Aguirre]

Khadije's answer works great!