Building LSTMs

Now that we have some baseline models in place, let's proceed by constructing what this chapter is all about: an LSTM. We will first start with a plain one-layer LSTM with no dropout strategy, equipping it with 32 neurons as follows:

def simple_lstm():
    model = Sequential()
    model.add(LSTM(32, input_shape=(1, 7)))
    
    model.add(Dense(1, activation='linear'))
    
    model.compile(loss='mae', optimizer='adam')
    return model

We connect the LSTM layer to our dense regressor layer, and continue to use the same loss and optimizer and loss functions.

Stacked LSTM

Next, we simply stack two LSTM layers on top of each other, just like we did with the GRUs in the previous...