Hands-On Machine Learning for Algorithmic Trading

How to define a two-layer RNN using a single LSTM layer

Having created input/output pairs from our stock index time series and split the pairs into training/testing sets, we can now begin setting up our RNN. Keras makes it very straightforward to build a two-hidden-layer RNN of the following specifications:

• Layer 1 uses an LSTM module with 20 hidden units (note here the bit that reads input_shape = (window_size,1)).
• Layer 2 uses a fully connected module with one unit.
• We use the mean_squared_error loss because we are performing regression.

This can be constructed using just a few lines, as follows:

rnn = Sequential([    LSTM(units=20,          input_shape=(window_size, n_features), name='LSTM'),    Dense(1, name='Output')])

The summary shows that the ...