TensorFlow2.0之RNN情感分类问题实战

TensorFlow2.0之RNN情感分类问题实战

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import Sequential, layers, datasets, optimizers, losses
import numpy as np
import os

tf.random.set_seed(22)
np.random.seed(22)
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
assert tf.__version__.startswith('2.')

# 加载数据
# 批大小
batchsize = 128
# 词汇表大小
total_words = 10000
# 句子最大长度s,大于的句子部分将截断，小于的将填充
max_review_len = 80
# 词向量特征长度n
embedding_len = 100
# 加载IMDB数据集，此处的数据采用数字编码，一个数字代表一个单词
(x_train, y_train), (x_test, y_test) = keras.datasets.imdb.load_data(num_words=total_words)
# 打印输入的形状，标签的形状
# print(x_train.shape, len(x_train[0]), y_train.shape)  # (25000,) 218 (25000,)
# print(x_test.shape, len(x_test[0]), y_test.shape)  # (25000,) 68 (25000,)
# 数字编码表
word_index = keras.datasets.imdb.get_word_index()
# 打印出编码表的单词和对应的数字
# for k, v in word_index.items():
#     print(k, v)
# 前面四个ID是特殊位
word_index = {k: (v + 3) for k, v in word_index.items()}
# 填充标志
word_index["<PAD>"] = 0
# 起始标志
word_index["<START>"] = 1
# 未知单词的标志
word_index["<UNK>"] = 2
# 没有用到单词的标志
word_index["<UNUSED>"] = 3
# 翻转编码表
reverse_word_index = dict([(value, key) for (key, value) in word_index.items()])


# 将数字编码的句子转换为字符串数据
def decode_review(text):
    return ''.join([reverse_word_index.get(i, '?') for i in text])


decode_review(x_train[0])
# print(x)
# 截断和填充句子，使得等长，此处长句子保留句子后面的部分，短句子在前面填充
x_train = keras.preprocessing.sequence.pad_sequences(x_train, maxlen=max_review_len)
x_test = keras.preprocessing.sequence.pad_sequences(x_test, maxlen=max_review_len)
# 构建数据集，打散，批量，并丢掉最后一个不够batchsize的batch
db_train = tf.data.Dataset.from_tensor_slices((x_train, y_train))
db_train = db_train.shuffle(1000).batch(batchsize, drop_remainder=True)
db_test = tf.data.Dataset.from_tensor_slices((x_test, y_test))
db_test = db_test.batch(batchsize, drop_remainder=True)


# 统计数据集属性
# print('x_train shape:', x_train.shape, tf.reduce_max(y_train), tf.reduce_min(y_train))
# print('x_test shape:', x_test.shape)


# 网络模型
class MyRNN(keras.Model):
    # Cell方式构建多层网络
    def __init__(self, units):
        super(MyRNN, self).__init__()
        # [b, 64], 构建Cell初始化状态向量，重复使用
        self.state0 = [tf.zeros([batchsize, units])]
        self.state1 = [tf.zeros([batchsize, units])]
        # 词向量编码[b, 80] => [b, 80, 100]
        self.embedding = layers.Embedding(total_words, embedding_len, input_length=max_review_len)
        # 构建两个Cell，使用dropout技术防止过拟合
        self.rnn_cell0 = layers.SimpleRNNCell(units, dropout=0.5)
        self.rnn_cell1 = layers.SimpleRNNCell(units, dropout=0.5)
        # 构建分类网络，用于将CELL的输出特征进行分类，2分类
        # [b, 80, 100] => [b, 64] => [b, 1]
        self.outlayer = Sequential([
            layers.Dense(units),
            layers.Dropout(rate=0.5),
            layers.ReLU(),
            layers.Dense(1)])

    def call(self, inputs, training=None):
        # [b, 80]
        x = inputs
        # 获取词向量:[b, 80] => [b, 80, 100]
        x = self.embedding(x)
        # 通过2个RNN CELL，[b, 80, 100] => [b, 64]
        state0 = self.state0
        state1 = self.state1
        # word:[b, 100]
        for word in tf.unstack(x, axis=1):
            out0, state0 = self.rnn_cell0(word, state0, training)
            out1, state1 = self.rnn_cell1(out0, state1, training)
        # 末层最后一个输出作为分类网络的输入：[b, 64] => [b, 1]
        x = self.outlayer(out1, training)
        # 通过激活函数,p(y is pos|x)
        prob = tf.sigmoid(x)

        return prob


def main():
    # RNN 状态向量长度n
    units = 64
    # 训练世代
    epochs = 50
    # 创建模型
    model = MyRNN(units)
    # 装配
    model.compile(optimizer=optimizers.Adam(1e-3), loss=losses.BinaryCrossentropy(), metrics=['accuracy'],
                  experimental_run_tf_function=False)
    # 训练和验证
    model.fit(db_train, epochs=epochs, validation_data=db_test)
    # 测试
    model.evaluate(db_test)


if __name__ == '__main__':
    main()