이론과 관련된 내용은 여기를 참고하시면 됩니다.

간단한 RNN 모델

from keras.models import Sequential
from keras.layers import SimpleRNN

model = Sequential()
model.add(SimpleRNN(3, input_shape=(2,10)))
# model.add(SimpleRNN(3, input_length=2, input_dim=10))와 동일함.
model.summary()
----------------------------------------------------------------
Layer (type)                 Output Shape              Param #   
=================================================================
simple_rnn_1 (SimpleRNN)     (None, 3)                 42        
=================================================================
Total params: 42
Trainable params: 42
Non-trainable params: 0

model = Sequential()
model.add(SimpleRNN(3, batch_input_shape=(8,2,10)))
model.summary()
----------------------------------------------------------
Layer (type)                 Output Shape              Param #   
=================================================================
simple_rnn_2 (SimpleRNN)     (8, 3)                    42        
=================================================================
Total params: 42
Trainable params: 42
Non-trainable params: 0

model = Sequential()
model.add(SimpleRNN(3, batch_input_shape=(8,2,10), return_sequences=True))
model.summary()
------------------------------------------------------------------------
Layer (type)                 Output Shape              Param #   
=================================================================
simple_rnn_3 (SimpleRNN)    (8, 2, 3)                 42        
=================================================================
Total params: 42
Trainable params: 42
Non-trainable params: 0

rnn = SimpleRNN(3, return_sequences=True, return_state=True)
hidden_states, last_state = rnn(train_X)

print('hidden states : {}, shape: {}'.format(hidden_states, hidden_states.shape))
print('last hidden state : {}, shape: {}'.format(last_state, last_state.shape))
-----------------------------------------------------------------------------------
hidden states : [[[ 0.29839835 -0.99608386  0.2994854 ]
  [ 0.9160876   0.01154806  0.86181474]
  [-0.20252597 -0.9270214   0.9696659 ]
  [-0.5144398  -0.5037417   0.96605766]]], shape: (1, 4, 3)
last hidden state : [[-0.5144398  -0.5037417   0.96605766]], shape: (1, 3)

깊은 RNN 모델

model = Sequential()
model.add(SimpleRNN(hidden_size, return_sequences = True))
model.add(SimpleRNN(hidden_size, return_sequences = True))

Bidirectional RNN 모델

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import SimpleRNN, Bidirectional

model = Sequential()
model.add(Bidirectional(SimpleRNN(hidden_size, return_sequences = True), input_shape=(timesteps, input_dim)))