MATLAB equivalent functions in Keras

layers = [ ...

sequenceInputLayer(inputSize)

lstmLayer(numHiddenUnits1)

lstmLayer(numHiddenUnits2)

fullyConnectedLayer(numResponses)

regressionLayer

];

What would be these layers be in Keras?

[Ruhi Thomas]

i know lstmLayer is tf.keras.layers.LSTM

What about the others?

Answer 1

[Aneela]

Hi Ruhi Thomas,

If “tf.keras” is the way you imported Keras from TensorFlow, the above layers are equivalent to the following layers in Keras:

sequenceInputLayer(inputSize) –

 inputLayer= tf.keras.layers.Input(shape=(None, inputSize)) 

lstmLayer(numHiddenUnits1) –

 lstm_layer1=tf.keras.layers.LSTM(numHiddenUnits1, return_sequences=True)(inputLayer) 

lstmLayer(numHiddenUnits2) –

 lstm_layer2=tf.keras.layers.LSTM(numHiddenUnits2, return_sequences=True)(inputLayer) 

fullyConnectedLayer(numResponses) –

 dense_layer = tf.keras.Layers.Dense(numResponses)(lstm_layer2) 

regressionLayer –

• In keras, there is no separate need for regression layer, instead we specify the loss function as part of the model compilation.
• For a regression task, loss functions like “mean_squared_error”, “mean_absolute_error” are typically used.

 model = Model(inputs=input_layer, outputs=dense_layer) 
 model.compile(optimizer='adam', loss='mean_squared_error') 

Hope this helps!!