Input shape for the LSTM model

Question

Yongho Son 2021-7-28

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/887604-input-shape-for-the-lstm-model

回答： Prince Kumar 2021-9-7

Train dataset X has a shape of 1x50000 and each of 50000 elements has 5x1 data. Train dataset Y also has a shape of 1x50000 and each elements has 1x1 data. I wonder if the shape of the two datasets are valid to train lstm model like below because it keeps giving me an error msg " Invalid training data. Responses must be a matrix of numeric responses, or a N-by-1 cell array of sequences, where N is the number of sequences. The feature dimension of all sequences must be the same.

numFeatures = 5;
numHiddenUnits = 200;
numClasses = 1;
layers = [ ...
    sequenceInputLayer(numFeatures)
    lstmLayer(numHiddenUnits)
    dropoutLayer(0.5)
    fullyConnectedLayer(numClasses)
    regressionLayer];
options = trainingOptions('adam', ...
    'MaxEpochs',50, ...
    'ValidationData',{test_x, test_y}, ...
    'GradientThreshold',1, ...
    'Verbose',1, ...
    'Shuffle','never', ...
    'ExecutionEnvironment','gpu', ...
    'Plots','training-progress');
net = trainNetwork(train_x, train_y, layers, options);