Error in training Neural Network: Invalid training data. For a recurrent layer with output mode 'last', responses must be a categorical column vector.

Question

John Schwarcz 2020-5-8

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/524116-error-in-training-neural-network-invalid-training-data-for-a-recurrent-layer-with-output-mode-las

评论： shan liu 2022-5-23

I am trying to train an LSTM on a time-series dataset.

The training data is of dimensions (time x samples) and in preprocessing I put it into a cell (samples x 1) with each array in the cell having dimension (1 x time)

The target data is of dimensions (1 x samples) and specifies the category the sample belongs to. In preprocessing I create a cell (samples x 1) of categorical labels with dimension (1 x categories). The labels are arrays with one-hot encoding to indicate the category.

I'm getting the error: "Invalid training data. For a recurrent layer with output mode 'last', responses must be a categorical column vector."

I'm not sure what is causing this error, any help would be greatly appreciated! Thank you!

Code:

%training
train_on = full(getfield(train_data,'spikes'));
targets = mod(getfield(train_data,'target_direction'), 360);
tm = getfield(train_data,'target_motion');
directions = unique(targets);
directions_index = 1:length(directions);
%% preprocess data
true = zeros(length(directions), length(targets));
for i = 1:length(targets)
    ind = find(directions == targets(i));
    true(ind, i) = 1;
end
train_on = train_on(tm(1): end-1, :);
train_on = permute(train_on, [2, 1]);
true = permute(true, [2, 1]);
input = cell(size(train_on, 1), 1);
labels = cell(size(train_on, 1), 1);
for i = 1:size(train_on,1)
    input{i, 1} = [train_on(i, :)];
    labels{i, 1} = categorical(true(i,:));
end
%% make LSTM
numFeatures = 1; numHiddenUnits = 100; numClasses = length(directions);
layers = [ ...
    sequenceInputLayer(numFeatures)
    lstmLayer(numHiddenUnits,'OutputMode','last')
    fullyConnectedLayer(numClasses)
    softmaxLayer
    classificationLayer];
maxEpochs = 10; miniBatchSize = 20;
options = trainingOptions('adam', ...
    'ExecutionEnvironment','cpu', ...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'GradientThreshold',1, ...
    'Verbose',false, ...
    'Plots','training-progress');
net = trainNetwork(input,labels,layers,options);

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

shan liu 2022-5-23

I got a same error as you and resolved it.

at your code there:

labels = cell(size(train_on, 1), 1);

for i = 1:size(train_on,1)

input{i, 1} = [train_on(i, :)];

labels{i, 1} = categorical(true(i,:));

end

Check your labels. It is a cell, not a categorical. Make it to size(train_on,1)*1 categorical.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Harsha Priya Daggubati 2020-5-11

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/524116-error-in-training-neural-network-invalid-training-data-for-a-recurrent-layer-with-output-mode-las#answer_431684

Hi,

As it is mentioned in the error message, responses (i.e. training data labels) should be of categorical type. In your case can you check the type of labels, just to ensure it is of categorical type with size samplesx1.

Also, are you sure that each sample has information about only one feature?