Recurrent Layers

Introduction

Recurrent layers process sequential data by maintaining internal state and passing information across timesteps.

LSTM

from tensorflow.keras import layers

model = keras.Sequential([
    layers.LSTM(64, return_sequences=True, input_shape=(timesteps, features)),
    layers.LSTM(32),
    layers.Dense(10, activation='softmax')
])

# Bidirectional LSTM
model.add(layers.Bidirectional(layers.LSTM(64)))

GRU

# GRU - fewer parameters than LSTM
model = keras.Sequential([
    layers.GRU(64, return_sequences=True, input_shape=(timesteps, features)),
    layers.GRU(32),
    layers.Dense(10)
])

Stacking RNN Layers

model = keras.Sequential([
    layers.LSTM(64, return_sequences=True),
    layers.LSTM(32, return_sequences=True),
    layers.LSTM(16),
    layers.Dense(10)
])

RNN with State

# Stateful RNN maintains state across batches
model = keras.Sequential([
    layers.LSTM(64, batch_input_shape=(8, timesteps, features), stateful=True)
])

# Manual state reset
model.reset_states()

Practice Problems

1. Build LSTM for sequence prediction
2. Use Bidirectional LSTM
3. Stack multiple LSTM layers
4. Implement GRU
5. Handle variable length sequences