How to design LSTM-CNN on deep network designer?

Hello,

My project is on classification of ECG/EEG signals using deep learning. I have design based on sequence on LSTM layer. Now i want to design hybrid LSTM-CNN on deep network designer which i have problem with connection between LSTM and Convolutional layer. I used Sequencefolding layer (suggested by deep network designer) after LSTM and connect to Convolutionallayer2d. The problem is Sequencefolding layer have two output (1. output, 2. minibatchsize) , which i don't now where to connect this minibatchsize connection. Can somebody expert give me advice on this? Really appreciate on any advice.

Thanks in advance sir.

Answer 1

[Divya Gaddipati]

You have to use a sequenceUnfoldingLayer that takes two inputs, feature map and the miniBatchSize from the corresponding sequenceLayer. You can refer to this example for more information.

[NurAlisa Ali]

Thank you very much for this sir. From the example given, it is for hybrid CNN-LSTM, what i'm try to design is LSTM-CNN....

Answer 2

[Dreaman]

i have the same problem too, have u solved this problem?

[NurAlisa Ali]

Yeah i have try CNN-LSTM, but the input length must be not too long, otherwise will get out of memory even 32GB ram.

[Manoj Devaraju]

Hello Ali,

Evn I would like to apply CNN-LSTM network for the image data set classification problem. But unfortunately i am struggling to apply, can you please give me some insight, how can it be done?

Answer 3

[H W]

% Load data

[XTrain,YTrain] = japaneseVowelsTrainData;

% Define layers

layers = [ sequenceInputLayer(12,'Normalization','none', 'MinLength', 9);

convolution1dLayer(3, 16)

batchNormalizationLayer()

reluLayer()

maxPooling1dLayer(2)

convolution1dLayer(5, 32)

batchNormalizationLayer()

reluLayer()

averagePooling1dLayer(2)

lstmLayer(100, 'OutputMode', 'last')

fullyConnectedLayer(9)

softmaxLayer()

classificationLayer()];

options = trainingOptions('adam', ...

'MaxEpochs',10, ...

'MiniBatchSize',27, ...

'SequenceLength','longest');

% Train network

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