Stacking RNN layers
Deep RNN models can have tens to hundreds of layers in order to achieve state-of-the-art results.
In this exercise, you will get a glimpse of how to create deep RNN models by stacking layers of LSTM cells one after the other.
To do this, you will set the return_sequences
argument to True
on the firsts two LSTM
layers and to False
on the last LSTM
layer.
To create models with even more layers, you can keep adding them one after the other or create a function that uses the .add()
method inside a loop to add many layers with few lines of code.
Este ejercicio forma parte del curso
Recurrent Neural Networks (RNNs) for Language Modeling with Keras
Instrucciones del ejercicio
- Import the
LSTM
layer. - Return the sequences in the first two layers and don't return the sequences in the last
LSTM
layer. - Load the pre-trained weights.
- Print the loss and accuracy obtained.
Ejercicio interactivo práctico
Prueba este ejercicio y completa el código de muestra.
# Import the LSTM layer
from tensorflow.keras.layers import ____
# Build model
model = Sequential()
model.add(LSTM(units=128, input_shape=(None, 1), return_sequences=____))
model.add(LSTM(units=128, return_sequences=____))
model.add(LSTM(units=128, return_sequences=____))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
# Load pre-trained weights
model.____('lstm_stack_model_weights.h5')
____("Loss: %0.04f\nAccuracy: %0.04f" % tuple(model.evaluate(X_test, y_test, verbose=0)))