I'm having trouble with convolution1dLayer

layers = [
    featureInputLayer(24)
    convolution1dLayer(5, 32, 'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling1dLayer(2, 'Stride', 2)
    convolution1dLayer(5, 64, 'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling1dLayer(2, 'Stride', 2)
    convolution1dLayer(5, 128, 'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling1dLayer(2, 'Stride', 2)
    dropoutLayer(0.5)
    fullyConnectedLayer(5)
    softmaxLayer
    classificationLayer];
options = trainingOptions('adam', ...
    'MaxEpochs', 20, ...
    'MiniBatchSize', 128, ...
    'ValidationData', {XVal, YVal}, ...
    'ValidationFrequency', 50, ...
    'Shuffle', 'every-epoch', ...
    'Verbose', false, ...
    'Plots', 'training-progress');
%Xtrain = 5000x24 Ytrain = 5000x1 Xtest = 5000x24 Ytest=50000x1
net = trainNetwork(XTrain, YTrain, layers, options);
% Doğruluk oranını hesapla
YPred = classify(net, XTest);
accuracy = sum(YPred == YTest) / numel(YTest);
fprintf('Doğruluk oranı: %0.2f%%\n', 100*accuracy);

I have a dataset of 15300 records with 24 features. (size 15300x24) My output dataset consists of 5 classes (15300x1). I am trying to classify with cnn. When I write the Layer, I encounter the following error:

Caused by:

Layer 2: Input data must have one spatial dimension only, one temporal dimension only, or one of each.

Instead, it has 0 spatial dimensions and 0 temporal dimensions.

I haven't been able to solve it.

[Walter Roberson]

Your XTrain is empty, somehow.

[nagihan yagmur]

clc;
clear all;
load veri_seti.mat
X = MyData.Inp;
Y = categorical(MyData.Out); 
numClasses = 5;
layers = [    featureInputLayer(24,'Name','inputs')                 
    convolution1dLayer(128,3,'Stride',2)    
    reluLayer()    maxPooling1dLayer(2,'Stride',2)    
    batchNormalizationLayer()    
    convolution1dLayer(64,3,'Stride',1)   
    reluLayer()   
    maxPooling1dLayer(2,'Stride',2)    
    batchNormalizationLayer()   
    dropoutLayer(0.2)    
    convolution1dLayer(32,3,'Stride',1)    
    reluLayer()    
    batchNormalizationLayer()    
    convolution1dLayer(16,3,'Stride',1)   
    reluLayer()    
    batchNormalizationLayer()    
    dropoutLayer(0.2)    
    convolution1dLayer(8,3,'Stride',1)    
    reluLayer()    
    maxPooling1dLayer(2,'Stride',2)    
    globalMaxPooling1dLayer()   
    dropoutLayer(0.2)    
    batchNormalizationLayer()    
    fullyConnectedLayer(1024)    
    fullyConnectedLayer(1024)    
    softmaxLayer()    
    classificationLayer()];
% Öğrenme oranı ve diğer hiperparametreler
miniBatchSize = 128;
maxEpochs = 30;
initialLearningRate = 0.001;
learnRateDropFactor = 0.1;
learnRateDropPeriod = 10;
% Options nesnesi
options = trainingOptions('adam', ...
    'MiniBatchSize', miniBatchSize, ...
    'MaxEpochs', maxEpochs, ...
    'InitialLearnRate', initialLearningRate, ...
    'LearnRateSchedule', 'piecewise', ...
    'LearnRateDropFactor', learnRateDropFactor, ...
    'LearnRateDropPeriod', learnRateDropPeriod, ...
    'Shuffle', 'every-epoch', ...
    'Verbose', false, ...
    'Plots', 'training-progress');

Error using vertcat

Dimensions of arrays being concatenated are not consistent.

Error in example (line 9)

layers = [ featureInputLayer(24,'Name','inputs')

I'm constantly between these two errors

Answer 1

[Matt J]

Tech Support has suggested 2 workarounds to me. The simplest IMO is to recast the training as a 2D image classification problem, where one of the dimensions of the image is a singleton. This requires the use of an imageInputLayer as well as converting convolutional and pooling layers to 2D form, also specifying one of the dimensions as a singleton.

load veri_seti

XTrain = reshape(MyData.Inp',24,1,1,[]);  %Dimensions: 24x1x1xBatch
YTrain = reshape( categorical(MyData.Out),[],1); %Dimensions: Batchx1

layers = [ imageInputLayer([24,1],'Name','inputs')        %<---Use imageInputLayer         
    convolution2dLayer([5,1], 32, 'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer([2,1], 'Stride', [2,1])
    convolution2dLayer([5,1], 64, 'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer([2,1], 'Stride', [2,1])
    convolution2dLayer([5,1], 128, 'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer([2,1], 'Stride', [2,1])
    dropoutLayer(0.5)
    flattenLayer
    fullyConnectedLayer(5)
    softmaxLayer
    classificationLayer];

%analyzeNetwork(layers);



options = trainingOptions('adam', ...
    'MaxEpochs', 3, ...
    'MiniBatchSize', 128, ...
    'Verbose', false, ...
    'Plots', 'training-progress','ExecutionEnvironment','cpu');

net = trainNetwork(XTrain, YTrain(:), layers,options);

[Matt J]

I have only one problem left. It says iteration per epoch = 1. How can I change this? I think Xtrain takes the sample count as 24 and returns 1 even if I give the minibatch size 100.

With the reformulation above, I am seeing the expected iterations per epoch:

[nagihan yagmur]

I am very very grateful

[Matt J]

I'm glad, but please Accept-click the answer to indicate that it worked.

Answer 2

[Walter Roberson]

layers = [    featureInputLayer(24,'Name','inputs')                 
    convolution1dLayer(128,3,'Stride',2)    
    reluLayer()    maxPooling1dLayer(2,'Stride',2)    

Notice you have two layers on the same line.

[nagihan yagmur]

Yes, I made a mistake there. But I still couldn't solve the other error.

Error using trainNetwork

Invalid network.

Error in example (line 58)

net = trainNetwork(X', Y, layers, options);

Caused by:

Layer 2: Input data must have one spatial dimension only, one temporal dimension only, or one of each.

Instead, it has 0 spatial dimensions and 0 temporal dimensions.

[Walter Roberson]

Please show

whos -file veri_seti.mat
whos X Y

[nagihan yagmur]

The dataset is attached.