Overview

本篇文章我们记录一下怎么用TensorFlow 2.0当中的Keras模块来进行RNN和LSTM文本分类。

1. 加载IMDB评论数据

import tensorflow_datasets as tfds
import tensorflow as tf
from tensorflow.keras import models, layers, losses, optimizers, Sequential
import matplotlib.pyplot as plt

dataset, info = tfds.load('imdb_reviews/subwords8k', with_info=True, as_supervised=True)
train_dataset, test_dataset = dataset['train'], dataset['test']
encoder = info.features['text'].encoder
print('Vocabulary size: {}'.format(encoder.vocab_size))


BUFFER_SIZE = 10000
BATCH_SIZE = 64
train_dataset = train_dataset.shuffle(BUFFER_SIZE)
train_dataset = train_dataset.padded_batch(BATCH_SIZE)

test_dataset = test_dataset.padded_batch(BATCH_SIZE)

2. 构建RNN模型

model = Sequential([
    layers.Embedding(encoder.vocab_size, 64),
    layers.Bidirectional(layers.LSTM(64)),
    layers.Dense(64, activation='relu'),
    layers.Dense(1)
])

其中，Embedding层是用来做word embedding的，把稀疏向量转换成稠密向量；Bidirectional层是一个RNN的双向封装器，用于对序列进行前向和后向计算。

3. 模型编译训练

model.compile(loss=losses.BinaryCrossentropy(from_logits=True),
              optimizer=optimizers.Adam(1e-4),
              metrics=['acc'])

history = model.fit(train_dataset, epochs=10,
                    validation_data=test_dataset, 
                    validation_steps=30)
test_loss, test_acc = model.evaluate(test_dataset)

print('Test Loss: {}'.format(test_loss))
print('Test Accuracy: {}'.format(test_acc))

测试集效果如下：

391/391 [==============================] - 109s 280ms/step - loss: 0.4388 - acc: 0.8553
Test Loss: 0.43875041604042053
Test Accuracy: 0.8552799820899963

画个图，看看效果：

plt.rcParams['font.size'] = 14
def plot_graphs(history, metric):
    plt.plot(history.history[metric])
    plt.plot(history.history['val_'+metric], '')
    plt.xlabel("Epochs")
    plt.ylabel(metric)
    plt.legend([metric, 'val_'+metric])
    plt.show()
plot_graphs(history, 'acc')
plot_graphs(history, 'loss')

acc.png

loss.png

本文参考了如下文章：
Text classification with an RNN
TensorFlow 2 中文文档 - RNN LSTM 文本分类