A Practical Approach to Timeseries Forecasting Using Python - GRU Models

The video tutorial discusses Gated Recurrent Units (GRUs) and compares them with Long Short-Term Memory (LSTM) networks. GRUs are computationally cheaper and faster but less accurate for longer sequences compared to LSTMs. The tutorial explains the three gates of GRUs: update, reset, and current memory gate, and how they function. It also covers the implementation of GRU and simple RNN layers in Python, highlighting the code similarities with LSTMs. The focus then shifts to LSTM models, discussing issues of underfitting and overfitting, and how to address them.

What is one advantage of GRUs over LSTMs?

• They have more gates than LSTMs.

• They have an internal cell state.

• They use less memory and are faster.

• They are more accurate with longer sequences.

Which gate in a GRU is responsible for determining how much past information should be retained?

• Reset gate

• Input gate

• Update gate

• Output gate

How does the reset gate in a GRU function?

• It combines the functions of the input and forget gates in LSTMs.

• It acts as the output gate in LSTMs.

• It maintains the internal cell state.

• It determines how much past information to retain.

In Python, how do you add a GRU layer to a model?

• Use model.add(Dense) and specify the number of neurons

• Use model.add(SimpleRNN) and specify the number of neurons

• Use model.add(GRU) and specify the number of neurons

• Use model.add(LSTM) and change to GRU

What is the purpose of setting 'return_sequences' to true in a GRU layer?

• To initialize the GRU layer

• To output the entire sequence of states

• To add a dense layer

• To output only the final state

Which type of RNN is most computationally expensive?

• Bi-directional LSTM

• GRU

• Simple RNN

• LSTM

What is a common issue when using LSTMs on datasets?

• They use too little memory.

• They are too fast.

• They always underfit.

• They can overfit or underfit.