ComeçarComece de graça

Better sentiment classification

In this exercise, you go back to the sentiment classification problem seen in Chapter 1.

You are going to add more complexity to the model and improve its accuracy. You will use an Embedding layer to train word vectors on the training set and two LSTM layers to keep track of longer texts. Also, you will add an extra Dense layer before the output.

This is no longer a simple model, and the training can take some time. For this reason, a pre-trained model is available by loading its weights with the method .load_weights() from the keras.models.Sequential class. The model was trained with 10 epochs and its weights are available on the file model_weights.h5.

The following modules are loaded on the environment: Sequential, Embedding, LSTM, Dropout, Dense.

Este exercício faz parte do curso

Recurrent Neural Networks (RNNs) for Language Modeling with Keras

Ver curso

Instruções do exercício

  • Add an Embedding layer as the first layer of the model.
  • Add a second LSTM layer with 64 units returning the sequences.
  • Add an extra Dense layer with 16 units.
  • Evaluate the model to print the accuracy on the training set.

Exercício interativo prático

Experimente este exercício completando este código de exemplo.

# Build and compile the model
model = Sequential()
model.add(____(vocabulary_size, wordvec_dim, trainable=True, input_length=max_text_len))
model.add(____(64, return_sequences=____, dropout=0.2, recurrent_dropout=0.15))
model.add(LSTM(64, return_sequences=False, dropout=0.2, recurrent_dropout=0.15))
model.add(____(16))
model.add(Dropout(rate=0.25))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

# Load pre-trained weights
model.load_weights('model_weights.h5')

# Print the obtained loss and accuracy
print("Loss: {0}\nAccuracy: {1}".format(*model.____(X_test, y_test, verbose=0)))
Editar e executar o código