tensorflow2.0 循环神经网络--情感分类实战

tensorflow2.0 循环神经网络–情感分类实战代码

本文主要是情感分类单层实战RNN Cell代码

import os
import numpy as np
import tensorflow as tf
config = tf.compat.v1.ConfigProto(gpu_options=tf.compat.v1.GPUOptions(allow_growth=True))
sess = tf.compat.v1.Session(config=config)
from tensorflow import keras
from tensorflow.keras import layers
tf.random.set_seed(220)
np.random.seed(22)
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
assert tf.__version__.startswith('2.')
#x为用户的评语，y为标记  数据集为imdb
total_words=10000
max_review_len=80
embedding_len=100
batchsz=128
(x_train,y_train),(x_test,y_test)=keras.datasets.imdb.load_data(num_words=total_words)
#x_train:[b,80],number b个句子，每个句子有80个单词
#x_test:[b,80]
x_train=keras.preprocessing.sequence.pad_sequences(x_train,maxlen=max_review_len)#数据预处理  maxlen：None或整数，为序列句子的最大长度。大于此长度的序列将被截短，小于此长度的序列将在后部填0.
x_test=keras.preprocessing.sequence.pad_sequences(x_test,maxlen=max_review_len)
db_train=tf.data.Dataset.from_tensor_slices((x_train,y_train))
db_train=db_train.shuffle(1000).batch(batchsz,drop_remainder=True)#drop_remainder:最后一个batch小于batchsz就会丢弃掉
db_test=tf.data.Dataset.from_tensor_slices((x_test,y_test))
db_test=db_test.shuffle(1000).batch(batchsz,drop_remainder=True)
print('x_train shape:',x_train.shape,tf.reduce_max(y_train),tf.reduce_min(y_train))
print('x_test shape:',x_test.shape)

#构建网络结构
class MyRNN(keras.Model):
    def __init__(self,units):
        super(MyRNN,self).__init__()#父类的初始化函数

        self.state0=[tf.zeros([batchsz,units])]
        self.state1=[tf.zeros([batchsz,units])]

        #将[b,80]转换为[b,80,100]
        self.embedding=layers.Embedding(total_words,embedding_len,
                                        input_length=max_review_len)#一个单词有100维，80个单词
        #[b,80,100],h_dim:64
        #RNN:cell1,cell2,cell3
        #SimpleRNN类
        self.rnn_cell0=layers.SimpleRNNCell(units,dropout=0.2)# 将单词的100维的向量进行rnn信息提取 dropout 防止过拟合
        self.rnn_cell1=layers.SimpleRNNCell(units,dropout=0.2)
        #分类 fc,[b,80,100] -->[b,64]-->[b,1]
        self.outlayer=layers.Dense(1)#1个输入节点 [b,64]-->[b,1]

    #前向运算的过程
    def call(self,inputs,training=None):
        '''
        net(x) net(x,training=True):train mode
        net(x,training=False)
        #前项运算
        :param inputs:  [b,80]
        :param training:
        :return:
        '''
        #[b,80]
        x=inputs
        #embedding:[b,80] -->[b,80,100]
        x=self.embedding(x)
        #rnn cell computer
        #[b,80,100]-->[b,64]
        state0=self.state0
        state1= self.state1
        for word in tf.unstack(x,axis=1): #word:[b,100]#在一维上进行展开
            #h1=x*wxh+h0*whh
            out,state0=self.rnn_cell0(word,state0,training)
        #out:[b,64]-->[b,1]
        x=self.outlayer(out)
        #p(y is pos|x)
        prob=tf.sigmoid(x)
        return prob

def main():
    units=64
    epochs=4
    model=MyRNN(units)
    model.compile(optimizer=keras.optimizers.Adam(0.001),
                  loss=tf.losses.BinaryCrossentropy(),
                  metrics=['accuracy'],
                  experimental_run_tf_function=False
                 )
    model.fit(db_train,epochs=epochs,validation_data=db_test)

    #测试
    model.evaluate(db_test)
if __name__ == '__main__':
    main()

文最后附上霍金的一句话：
人工智能的强力崛起，可能是人类历史上最好的事情，也可能是最糟糕的事情。
------史蒂芬·霍金