基于tensorflow的栈式自编码器实现

这周完全没有想法要看栈式编码器的，谁知误入桃花源，就暂且把栈式自编码器看了吧。由于手上有很多数值型的数据，仅仅是数据，没有标签，所以，迫切需要通过聚类抽出特征。无意间看到别人家公司的推荐系统里面用到sdae，于是，找了个ae程序，建了个sdae，跑一下自己的数据。希望sdae在后面的推荐系统中能有用。

啰嗦了那么多，先看看原理吧。http://ufldl.stanford.edu/wiki/index.php/%E6%A0%88%E5%BC%8F%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95

斯坦福的这篇文章原理讲的很到位了。

 

一．基本原理

AE的原理是先通过一个encode层对输入进行编码，这个编码就是特征，然后利用encode乘第2层参数（也可以是encode层的参数的转置乘特征并加偏执），重构（解码）输入，然后用重构的输入和实际输入的损失训练参数。

对于我的应用来说，我要做的首先是抽取特征。AE的encode的过程很简单，是这样的：

SAE是这样的：

 

训练过程中，两个SAE分别训练，第一个SAE训练完之后，其encode的输出作为第二个SAE的输入，接着训练。训练完后，用第二层的特征，通过softmax分类器（由于分类器 还是得要带标签的数据来训练，所以，实际应用总，提取特征后不一定是拿来分类），将所有特征分为n类，得到n类标签。训练网络图

          step 1

              step 2

                      step 3    

 按照UFLDL的说明，在各自训练到快要收敛的时候，要对整个网络通过反向传播做微调，但是不能从一开始就整个网络调节。

两个SAE训练完后，每个encode的输出作为两个隐层的特征，然后重新构建网络，新网络不做训练，只做预测。网络如下：

 

二．程序

1.      首先建立自编码器的网络

采用添加高斯噪声的dae，这是tensorflow 的models-master里面提供的一种自编码器。

#coding: utf-8
import tensorflow as tf
import numpy as npimport Utilsclass AdditiveGaussianNoiseAutoencoder(object): 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 # model self.x = tf.placeholder(tf.float32, [None, self.n_input])
        #编码
        self.hidden = self.transfer(tf.add(tf.matmul(self.x + scale * tf.random_normal((n_input,)),
                self.weights['w1']),
                self.weights['b1']))
        #解码
        self.reconstruction = tf.add(tf.matmul(self.hidden, self.weights['w2']), self.weights['b2'])

        # cost
        self.cost = 0.5 * tf.reduce_sum(tf.pow(tf.sub(self.reconstruction, self.x), 2.0))
        self.optimizer = optimizer.minimize(self.cost)

        init = tf.initialize_all_variables()
        self.sess = tf.Session()
        self.sess.run(init)

    def _initialize_weights(self):
        all_weights = dict()
        all_weights['w1'] = tf.Variable(Utils.xavier_init(self.n_input, self.n_hidden))
        all_weights['b1'] = tf.Variable(tf.zeros([self.n_hidden], dtype = tf.float32))
        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
    #优化参数
    def partial_fit(self, X):
        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):
        return self.sess.run(self.cost, feed_dict = {self.x: X,
                                                     self.scale: self.training_scale
                                                     })

    def transform(self, X):
        return self.sess.run(self.hidden, feed_dict = {self.x: X,
                                                       self.scale: self.training_scale
                                                       })

    def generate(self, hidden = None):
        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 self.sess.run(self.reconstruction, feed_dict = {self.x: X,
                                                               self.scale: self.training_scale
                                                               })

    def getWeights(self):
        return self.sess.run(self.weights['w1'])

    def getBiases(self):
        return self.sess.run(self.weights['b1'])

2. sdae建立

#coding: utf-8
from __future__ import division, print_function, absolute_import

import numpy as np
import tensorflow as tf
from autoencoder_models.DenoisingAutoencoder import AdditiveGaussianNoiseAutoencoder
import prepare_data

#prepare_data.py是自己写的数据整理文件，通过这个文件清理数据，并将其分成M-1份测试数据，1份测试数据
pd = prepare_data.TidyData(file1='data/dim_tv_mall_lng_lat.txt',
                           file2='data/ali-hdfs_2017-03-22_18-24-37.log',
                           M=8,
                           seed=12,
                           k=5)
pd.read_mall_location_np()
pd.calc_min_distance()
print ('data read finished!')

#定义训练参数
training_epochs = 5
batch_size = 1000
display_step = 1
stack_size = 3  #栈中包含3个ae
hidden_size = [20, 20, 20]
input_n_size = [3, 200, 200]

def get_random_block_from_data(data, batch_size):
    start_index = np.random.randint(0, len(data) - batch_size)
    return data[start_index:(start_index + batch_size)]

#建立sdae图
sdae = []
for i in xrange(stack_size):
    if i == 0:
        ae = AdditiveGaussianNoiseAutoencoder(n_input = 2,
                                               n_hidden = hidden_size[i],
                                               transfer_function = tf.nn.softplus,
                                               optimizer = tf.train.AdamOptimizer(learning_rate = 0.001),
                                               scale = 0.01)
        ae._initialize_weights()
        sdae.append(ae)
    else:
        ae = AdditiveGaussianNoiseAutoencoder(n_input=hidden_size[i-1],
                                              n_hidden=hidden_size[i],
                                              transfer_function=tf.nn.softplus,
                                              optimizer=tf.train.AdamOptimizer(learning_rate=0.001),
                                              scale=0.01)
        ae._initialize_weights()
        sdae.append(ae)

		
W = []
b = []
Hidden_feature = [] #保存每个ae的特征
X_train = np.array([0])
for j in xrange(stack_size):
    #输入
    if j == 0:
        X_train = np.array(pd.train_set)
        X_test = np.array(pd.test_set)
    else:
        X_train_pre = X_train
        X_train = sdae[j-1].transform(X_train_pre)
        print (X_train.shape)
        Hidden_feature.append(X_train)
	
	#训练
    for epoch in range(training_epochs):
        avg_cost = 0.
        total_batch = int(X_train.shape[1] / batch_size)
        # Loop over all batches
        for k in range(total_batch):
            batch_xs = get_random_block_from_data(X_train, batch_size)

            # Fit training using batch data
            cost = sdae[j].partial_fit(batch_xs)
            # Compute average loss
            avg_cost += cost / X_train.shape[1] * batch_size

        # Display logs per epoch step
        #if epoch % display_step == 0:
        print ("Epoch:", '%04d' % (epoch + 1), "cost=", "{:.9f}".format(avg_cost))
	
	#保存每个ae的参数
    weight = sdae[j].getWeights()
    #print (weight)
    W.append(weight)
    b.append(sdae[j].getBiases())