USING LSTM TO CLASSIFY DATA

Question

Ernest Modise - Kgamane 2024-6-7

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2126386-using-lstm-to-classify-data

移动：Matt J 2024-6-10

采纳的回答： Ayush Aniket

LSTMdataIn.xlsx

在 MATLAB Online 中打开

Please see my code below

% Step 1: Load the data from the Excel file
data = readmatrix('LSTMdataIn.xlsx');
% Step 2: Create labels
labels = [ones(200, 1); 2*ones(200, 1); 3*ones(200, 1); 4*ones(200, 1); 5*ones(200, 1)];
% Step 3: Reshape the data
numTimeSteps = 100;
numFeatures = 1;
reshapedData = reshape(data', numFeatures, numTimeSteps, []);
% Step 4: Split the data into training and testing sets
cv = cvpartition(labels, 'HoldOut', 0.2);
trainIdx = training(cv);
testIdx = test(cv);
XTrain = reshapedData(:, :, trainIdx);
YTrain = labels(trainIdx);
XTest = reshapedData(:, :, testIdx);
YTest = labels(testIdx);
% Step 5: Create and train the LSTM network
numHiddenUnits = 100;
layers = [ ...
    sequenceInputLayer(100)
    lstmLayer(numHiddenUnits)
    fullyConnectedLayer(5)
    softmaxLayer
    classificationLayer];
options = trainingOptions('adam', 'MaxEpochs', 10, 'MiniBatchSize', 32);
net = trainNetwork(XTrain, categorical(YTrain), layers, options);
% Step 6: Evaluate the trained network
YTestPred = classify(net, XTest);
accuracy = sum(YTestPred == categorical(YTest)) / numel(YTest);

I get the following error

Error using trainNetwork (line 191)

The training sequences are of feature dimension 1 100 but the input layer expects sequences of feature dimension 100.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Matt J 2024-6-10

This question looks the same as,

https://www.mathworks.com/matlabcentral/answers/2126426-function-to-trainnetwork-returns-an-unexpected-error?s_tid=srchtitle

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Ayush Aniket 2024-6-10

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2126386-using-lstm-to-classify-data#answer_1469511

在 MATLAB Online 中打开

Hi ernest,

Based on your data, I am assuming that you want to perform a 'sequence-to-label' classification task i.e. given a sequence the network should return the label.

There are several errors in your code that need modification:

The error you observe is due to the fact that the sequenceInputLayer expects as argument the number of features in the sequence dataset which is '1' for your data and not '100'. Refer to the following link: https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.sequenceinputlayer.html#mw_9b13d129-0114-47c8-9ea3-e2977dd4fed1
The trainNetwork function expects training data in the format of 'Nx1' cell array, where 'N' represents the number of observations and each cell should have the data in the format of 'c-by-s' matrix, where 'c' is the number of features of the sequence and 's' is the sequence length. For your data, that would be '1x100'. Here's how you can restructure your data:

    % Assuming reshapedData is of size [1, numTimeSteps, numSamples]
    % Convert reshapedData into a cell array where each cell contains a single sequence
    XTrainCell = num2cell(reshapedData(:, :, trainIdx), [1,2]);
    XTestCell = num2cell(reshapedData(:, :, testIdx), [1,2]);
    % Each cell now contains a 1xnumTimeSteps matrix, which is a single sequence for the LSTM
    % Adjust the training and testing data preparation
    XTrain = squeeze(XTrainCell);
    XTest = squeeze(XTestCell

For 'sequence-to-label' classification task, you have to provide an additional 'Name-Value' argument 'OutputMode' in the lstmLayer of your network as shown below. Refer to the following documentation link to read the default value for the argument: https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.lstmlayer.html#mw_9f7c5f93-4bf2-4ddb-b922-b1c122668b9a_sep_mw_4fd4123f-c49f-4a38-95d1-85f9e542d641

    layers = [ ...
        sequenceInputLayer(numFeatures)
        lstmLayer(numHiddenUnits,OutputMode="last")
        fullyConnectedLayer(5)
        softmaxLayer
        classificationLayer];
    

Note: Starting in MATLAB R2024a, the classificationLayer and the trainNetwork functions are not recommended to be used. Instead you can use trainnet function to train your LSTM network:

https://www.mathworks.com/help/deeplearning/ref/trainnet.html