自定义数据自编码器--tensorflow

#导入必要的库
import numpy as np
import sklearn.preprocessing as prep
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

# 自己定义一个参数初始化函数 --Xavier initialization  --标准的均匀分布 Xaiver 初始化器
def xavier_initialization(fan_in,fan_out,constant = 1):
    low = -constant * np.sqrt(6.0 / (fan_in + fan_out))
    high = constant * np.sqrt(6.0 / (fan_in + fan_out))
    return tf.random_uniform((fan_in,fan_out),
                             minval = low,maxval = high,
                             dtype = tf.float32)

# 定义一个 去噪自编码 class
class AdditiveGaussianNoiseAutoEncoder(object):
    '''
    n_input:输入的变量个数
    n_hidden: 隐藏层的节点数
    transfer_function:隐层的激活函数，默认softplus
    optimizer:优化器，默认Adam
    scale:高斯噪声系数，默认0.1 --training_scale
    weights:使用_initialize_weights()进行weights初始化
    function:_initialize_weights()、
    ps:这里只用了一个隐含层
    '''
    def __init__(self,n_input,n_hidden,transfer_function = tf.nn.softplus,
                optimizer = tf.train.AdamOptimizer(),scale = 0.1):
        self.n_input = n_input
        self.n_hidden = n_hidden
        self.transfer =transfer_function
        self.scale = tf.placeholder(tf.float32)
        self.training_scale = scale
        network_weights = self._initialize_weights()
        self.weights = network_weights
        # 输入x
        self.x = tf.placeholder(tf.float32,[None,n_input])
        #计算隐层的输出（默认一层）
        self.hidden = self.transfer(tf.add(tf.matmul(
            self.x + scale * tf.random_normal((n_input,)),
            self.weights['w1']),self.weights['b1']))
        #数据的复原重建
        self.reconstruction = tf.add(tf.matmul(self.hidden,
                                               self.weights['w2']),
                                     self.weights['b2'])
        
        #自定义编码loss function --平方误差函数
        self.cost = 0.5 * tf.reduce_sum(tf.pow(tf.subtract(
            self.reconstruction,self.x),2.0))
        
        #定义优化器
        self.optimizer = optimizer.minimize(self.cost)
        #创建Session
        init = tf.global_variables_initializer()
        self.sess = tf.Session()
        self.sess.run(init)
    def  _initialize_weights(self):
        '''
        初始化权重参数函数
        '''
        all_weights = dict()
        all_weights['w1'] = tf.Variable(xavier_initialization(self.n_input,self.n_hidden))
        all_weights['b1'] = tf.Variable(tf.zeros([self.n_hidden]),dtype = tf.float32)
        #输出层没有使用激活函数，w2和b2初始值设为零即可
        all_weights['w2'] = tf.Variable(tf.zeros([self.n_hidden,self.n_input]),dtype = tf.float32)
        all_weights['b2'] = tf.Variable(tf.zeros([self.n_input]),dtype = tf.float32)
        return all_weights
    def partial_fit(self,X):
        '''
        训练数据X，返回损失cost
        feed_dict: 训练时需要feed的数据字典（输入数据x和高斯系数scale）
        '''
        cost,opt = self.sess.run((self.cost,self.optimizer),
                                 feed_dict = {self.x:X,self.scale:self.training_scale})
        return cost
    def calc_total_cost(self,X):
        '''
        用于计算模型测试时用的函数，只求损失，不触发训练
        '''
        return self.sess.run(self.cost,
                             feed_dict = {self.x:X,self.scale:self.training_scale})
    def tansform(self,X):
        '''
        函数学习数据高级特征，返回自编码器隐层的输出结果
        '''
        return self.sess.run(self.hidden,
                             feed_dict = {self.x:X,self.scale:self.training_scale})
    def generate(self,hidden=None):
        '''
        将隐层的输出结果作为输入，复原提取到的高阶特征数据并返回
        '''
        if hidden is None:
            hidden = np.random.normal(size = self.weights['b1'])
        return self.sess.run(self.reconstruction,feed_dict = {self.hidden:hidden})
    def reconstruct(self,X):
        '''
        return重建或复原隐层输出的自编码数据--等同于先后执行transform和generate函数
        '''
        return self.sess.run(self.reconstruction,
                             feed_dict = {self.x:X,self.scale:self.training_scale})
    def getWeights(self):
        '''
        获取隐层权重w1
        '''
        return self.sess.run(self.weights['w1'])
    def getBiases(self):
        '''
        获取隐层偏置b1
        '''
        return self.sess.run(self.weights['b1'])
    

#读取数据mnist
mnist = input_data.read_data_sets('MNIST_data',one_hot = True)

Extracting MNIST_data\train-images-idx3-ubyte.gz
Extracting MNIST_data\train-labels-idx1-ubyte.gz
Extracting MNIST_data\t10k-images-idx3-ubyte.gz
Extracting MNIST_data\t10k-labels-idx1-ubyte.gz

def standard_scale(X_train,X_test):
    '''
    数据标准化处理
    '''
    preprocess = prep.StandardScaler().fit(X_train)
    X_train = preprocess.transform(X_train)
    X_test = preprocess.transform(X_test)
    return X_train,X_test
def get_random_block_from_data(data,batch_size):
    '''
    从数据中随机获取batch_size大小的训练集，不放回的抽取
    '''
    start_index = np.random.randint(0,len(data) - batch_size)
    return data[start_index:(start_index + batch_size)]

#提取训练数据
X_train,X_test = standard_scale(mnist.train.images,mnist.test.images)

#定义常用的训练参数
n_samples = int(mnist.train.num_examples)
training_epochs = 20
batch_size = 128
display_step = 1#每一轮epoch显示一次cost

#定义一个AGN自编码器实例
AGN = AdditiveGaussianNoiseAutoEncoder(n_input=784,
                                       n_hidden=200,
                                      transfer_function=tf.nn.softplus,
                                      optimizer=tf.train.AdamOptimizer(learning_rate=0.001),
                                      scale = 0.01)

#开始训练
for epoch in range(training_epochs):
    avg_cost = 0.
    total_batch = int(n_samples / batch_size)
    for i in range(total_batch):
        batch_xs = get_random_block_from_data(X_train,batch_size)
        
        cost = AGN.partial_fit(batch_xs)
        avg_cost += cost / n_samples * batch_size
        
    if epoch % display_step ==0:
        print("Epoch:",'%04d' % (epoch + 1),"cost=","{:.9f}".format(avg_cost))

Epoch: 0001 cost= 18546.710154545
Epoch: 0002 cost= 13127.617153409
Epoch: 0003 cost= 10881.075117045
Epoch: 0004 cost= 9427.339469318
Epoch: 0005 cost= 9333.850439773
Epoch: 0006 cost= 9067.035839773
Epoch: 0007 cost= 9929.138722727
Epoch: 0008 cost= 9507.725019886
Epoch: 0009 cost= 8539.850939205
Epoch: 0010 cost= 9277.037648295
Epoch: 0011 cost= 8203.704468182
Epoch: 0012 cost= 7974.056809091
Epoch: 0013 cost= 7523.191403409
Epoch: 0014 cost= 8728.067648864
Epoch: 0015 cost= 8446.473536364
Epoch: 0016 cost= 8368.747946591
Epoch: 0017 cost= 8495.513869886
Epoch: 0018 cost= 8680.677721591
Epoch: 0019 cost= 7734.869307955
Epoch: 0020 cost= 8421.617193750

print("Total cost:" + str(AGN.calc_total_cost(X_test)))

Total cost:618007.6