다음은 양방향 RNN만 적용한 모델의 결과입니다.
Model: "sequential_4"
________________________________________________________________
Layer (type) Output Shape Param #
================================================================
embedding_2 (Embedding) (None, None, 64) 523840
________________________________________________________________
bidirectional_2 (Bidirection (None, 128) 66048
________________________________________________________________
dense_16 (Dense) (None, 64) 8256
________________________________________________________________
dense_17 (Dense) (None, 1) 65
================================================================
Total params: 598,209
Trainable params: 598,209
Non-trainable params: 0
________________________________________________________________
모델을 컴파일하고 훈련시킵니다.
코드 8-24 모델 훈련
model.compile(loss='binary_crossentropy',
optimizer=tf.keras.optimizers.Adam(1e-4), metrics=['accuracy'])
history = model.fit(train_batches, epochs=5, validation_data=test_batches,
validation_steps=30)