GRU layer in TensorFlow - Cheat Sheet & Quick Revision

Recall & Review

beginner

What does GRU stand for in machine learning?

GRU stands for Gated Recurrent Unit, a type of recurrent neural network layer used to process sequences.

beginner

How does a GRU layer help in handling sequence data?

A GRU layer controls information flow using gates to remember or forget parts of the sequence, helping the model learn long-term dependencies.

intermediate

Name the two main gates inside a GRU cell.

The two main gates are the update gate and the reset gate. They decide what information to keep or discard.

intermediate

Compare GRU and LSTM layers in simple terms.

Both handle sequences well, but GRUs are simpler with fewer gates, making them faster and easier to train, while LSTMs have more gates and can capture more complex patterns.

intermediate

What is the typical output shape of a GRU layer in TensorFlow when return_sequences=True?

The output shape is (batch_size, timesteps, units), where units is the number of GRU cells and timesteps is the sequence length.

What is the main purpose of the update gate in a GRU layer?

A To decide how much past information to keep

B To reset the entire memory

C To increase the learning rate

D To add noise to the input

The update gate controls how much past information is kept for the current step.

Which of these is NOT a characteristic of a GRU layer?

A Has reset and update gates

B Uses convolutional filters

C Processes sequence data

D Requires fewer parameters than LSTM

GRU layers do not use convolutional filters; they are recurrent layers.

In TensorFlow, what argument makes a GRU layer return output for every time step?

A return_sequences=True

B return_state=True

C stateful=True

D activation='relu'

Setting return_sequences=True makes the GRU output a sequence for each time step.

Why might you choose a GRU over an LSTM?

A GRUs have more gates than LSTMs

B GRUs use convolutional layers

C GRUs always give better accuracy

D GRUs are simpler and faster to train

GRUs have fewer gates, making them simpler and faster to train than LSTMs.

What type of data is a GRU layer best suited for?

A Tabular data without order

B Static images

C Sequence or time series data

D Single numeric values

GRU layers are designed to handle sequence or time series data.

Explain how the reset and update gates work inside a GRU cell.

Think about how the GRU decides what to remember or forget at each step.

Describe a simple example where using a GRU layer would be helpful in a real-life application.

Consider tasks like predicting the next word in a sentence or forecasting weather.