How many LSTM blocks are there in bidirectional LSTM layers?

Hi,

How can i relate hidden layers with number of lstm blocks?

inputSize = 17

numHiddenUnits = 50;

numClasses = 2;

maxEpochs = 15;

miniBatchSize = 1;

layers = [ ...

sequenceInputLayer(inputSize)

bilstmLayer(numHiddenUnits,'OutputMode','last')

fullyConnectedLayer(numClasses)

softmaxLayer

classificationLayer]

options = trainingOptions('adam', ...

'ExecutionEnvironment','auto', ...

'GradientThreshold',1, ...

'MaxEpochs',maxEpochs, ...

'MiniBatchSize',miniBatchSize, ...

'SequenceLength','longest', ...

'Shuffle','never', ...

'Verbose',0, ...

'Plots','training-progress');

Answer 1

[Shantanu Dixit]

Hi Shweta,

Assuming that by hidden layers you mean numHiddenUnits, the numHiddenUnits refer to the number of LSTM blocks per direction (same for both LSTM and BiLSTM). So here the number of LSTM blocks are 50 (numHiddenUnits = 50).

Refer the documentation for lstm and bilstm layer:

Long short-term memory (LSTM) layer for recurrent neural network (RNN) - MATLAB - MathWorks India

Bidirectional long short-term memory (BiLSTM) layer for recurrent neural network (RNN) - MATLAB - MathWorks India