Tensorflow2.0笔记 - LSTM做IMDB数据集分析

        本笔记记录使用LSTMCell做IMDB情感分析，代码和前一个笔记使用SimpleRNN的代码基本一致，除了创建state的时候tensor的list包含两个状态参数外，基本没有任何区别。

import os
import time
import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics, Input

os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
#tf.random.set_seed(12345)
#np.random.seed(22)
tf.__version__

#取常见的10000个单词
total_words = 10000
#句子最长的单词数量设置为80
max_review_len = 80
#embedding设置为100,表示每个单词用100维向量表示
embedding_len = 100
#加载IMDB数据集
(x_train,y_train), (x_test, y_test) = datasets.imdb.load_data(num_words = total_words)
#对训练数据和测试数据的句子进行填充或截断
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
batchsize = 128
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)

#x_train包含25000个句子，每个句子包含80个单词，y_train标签为1表示好评，0表示差评
print('x_train: shape - ', x_train.shape, ' y_train: max/min -', tf.reduce_max(y_train).numpy(), '/', tf.reduce_min(y_train).numpy())
print('x_test: shape - ', x_test.shape)

class MyLSTM(keras.Model):
    #units：state的维度
    def __init__(self, total_words, embedding_len, max_review_len, units):
        super(MyLSTM, self).__init__()
        #初始的序列状态初始化为0（第0时刻的状态）h, c
        self.state0 = [tf.zeros([batchsize, units]), tf.zeros([batchsize, units])]
        self.state1 = [tf.zeros([batchsize, units]), tf.zeros([batchsize, units])]
        #embedding层，将文本转换为embedding表示
        #[b, 80] => [b, 80, 100]
        self.embedding = layers.Embedding(total_words, embedding_len, input_length=max_review_len)
        #[b, 80, 100] , units： 64 - 转换为64维的state [b, 64]
        self.lstm_cell0 = layers.LSTMCell(units, dropout=0.5)
        self.lstm_cell1 = layers.LSTMCell(units, dropout=0.5)
        #全连接层 [[b, 64] => [b, 1]
        self.outlayer = layers.Dense(1)
    #inputs: [b, 80] 
    def call(self, inputs, training=None):
        x = inputs
        #做embedding，[b,80] => [b, 80, 100]
        x = self.embedding(x)
        #做RNN cell计算
        #[b, 80, 100] => [b,  64]
        #遍历句子中的每个单词
        # word: [b, 100]
        state0 = self.state0
        state1 = self.state1
        for word in tf.unstack(x, axis=1):
            #h1 = x*w_xh + h0*w_hh
            out0, state0 = self.lstm_cell0(word, state0, training)
            out1, state1 = self.lstm_cell1(out0, state1, training)
        #循环完毕后，得到的out为[b, 64]，表示每个句子最终得到的状态
        x = self.outlayer(out1)
        #计算最终评价结果
        prob = tf.sigmoid(x)
        return prob

def main():
    units = 64
    epochs = 5
    lr = 0.001

    model = MyLSTM(total_words, embedding_len, max_review_len, units)
    model.compile(optimizer = optimizers.Adam(lr), loss = tf.losses.BinaryCrossentropy(),
                 metrics=['accuracy'])
    model.fit(db_train, epochs=epochs, validation_data=db_test)

    model.evaluate(db_test)

if __name__ == '__main__':
    main()

运行结果：