版权声明：本文为博主原创文章，未经博主允许不得转载。 https://blog.csdn.net/u013818406/article/details/66479518
自编码器（autoencoder）顾名思义，即可以使用自身的高阶特征编码自己。自编码器借助稀疏编码的思想，目的是使用稀疏的高阶特征重新组合来重构自己。因此，它的特点非常明显：第一，期望输入与输出一致；第二，希望使用高阶特征来重构自己，而不只是复制像素点。
首先使用tensorflow实现一个基本的自编码器类
import tensorflow as tf
class Autoencoder(object):
def __init__(self, n_input, n_hidden, transfer_function=tf.nn.softplus, optimizer = tf.train.AdamOptimizer()):
self.n_input = n_input
self.n_hidden = n_hidden
self.transfer = transfer_function
network_weights = self._initialize_weights()
self.weights = network_weights
# model
self.x = tf.placeholder(tf.float32, [None, self.n_input])
self.hidden = self.transfer(tf.add(tf.matmul(self.x, self.weights['w1']), self.weights['b1']))
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.subtract(self.reconstruction, self.x), 2.0))
self.optimizer = optimizer.minimize(self.cost)
init = tf.global_variables_initializer()
self.sess = tf.Session()
self.sess.run(init)
def _initialize_weights(self):
all_weights = dict()
all_weights['w1'] = tf.get_variable("w1", shape=[self.n_input, self.n_hidden],
initializer=tf.contrib.layers.xavier_initializer())
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
def partial_fit(self, X):
cost, opt = self.sess.run((self.cost, self.optimizer), feed_dict={self.x: X})
return cost
def calc_total_cost(self, X):
return self.sess.run(self.cost, feed_dict = {self.x: X})
def transform(self, X):
return self.sess.run(self.hidden, feed_dict={self.x: X})
def generate(self, hidden = None):
if hidden is None:
hidden = self.sess.run(tf.random_normal([1, self.n_hidden]))
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})
def getWeights(self):
return self.sess.run(self.weights['w1'])
def getBiases(self):
return self.sess.run(self.weights['b1'])
接下来谈谈自编码器的衍生降噪自编码器（Denoising Autoencoder），人类具有这样一种能力：当遮住一只狗的一小部分让你判断时，你很大可能还是能判断出这是一只狗。
这样的能力对应在自编码器上就是当我给予神经网络一定的噪声时，神经网络依然能给出正确的输出，这就是降噪自编码的思想。使用tensorflow实现时有两种不同的实现方式，一种是在输入层直接人为给输入加上一个高斯分布的噪声，另一种是使用dropout层让部分输入直接损失掉，这里使用第二种写法，第一种直接在hidden那里加上scale*tf.normal((n_input,))就可以实现了。
class DenoisingAutoencoder(object):
def __init__(self, n_input, n_hidden, transfer_function = tf.nn.softplus, optimizer = tf.train.AdamOptimizer(),
dropout_probability = 0.95):
self.n_input = n_input
self.n_hidden = n_hidden
self.transfer = transfer_function
self.dropout_probability = dropout_probability
self.keep_prob = tf.placeholder(tf.float32)
network_weights = self._initialize_weights()
self.weights = network_weights
# model
self.x = tf.placeholder(tf.float32, [None, self.n_input])
self.hidden = self.transfer(tf.add(tf.matmul(tf.nn.dropout(self.x, self.keep_prob), self.weights['w1']),
self.weights['b1']))
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.subtract(self.reconstruction, self.x), 2.0))
self.optimizer = optimizer.minimize(self.cost)
init = tf.global_variables_initializer()
self.sess = tf.Session()
self.sess.run(init)
def _initialize_weights(self):
all_weights = dict()
all_weights['w1'] = tf.get_variable("w1", shape=[self.n_input, self.n_hidden],
initializer=tf.contrib.layers.xavier_initializer())
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
def partial_fit(self, X):
cost, opt = self.sess.run((self.cost, self.optimizer),
feed_dict = {self.x: X, self.keep_prob: self.dropout_probability})
return cost
def calc_total_cost(self, X):
return self.sess.run(self.cost, feed_dict = {self.x: X, self.keep_prob: 1.0})
def transform(self, X):
return self.sess.run(self.hidden, feed_dict = {self.x: X, self.keep_prob: 1.0})
def generate(self, hidden=None):
if hidden is None:
hidden = self.sess.run(tf.random_normal([1, self.n_hidden]))
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.keep_prob: 1.0})
def getWeights(self):
return self.sess.run(self.weights['w1'])
def getBiases(self):
return self.sess.run(self.weights['b1'])
变分自编码器（variational autoencoders）不同点在于其隐藏代码来自于训练期间学习到的概率分布。具体流程如图
想要看详细的推理过程的可以去下面的链接
直接上代码
import tensorflow as tf
import numpy as np
class VariationalAutoencoder(object):
def __init__(self, n_input, n_hidden, optimizer = tf.train.AdamOptimizer()):
self.n_input = n_input
self.n_hidden = n_hidden
network_weights = self._initialize_weights()
self.weights = network_weights
# model
self.x = tf.placeholder(tf.float32, [None, self.n_input])
self.z_mean = tf.add(tf.matmul(self.x, self.weights['w1']), self.weights['b1'])
self.z_log_sigma_sq = tf.add(tf.matmul(self.x, self.weights['log_sigma_w1']), self.weights['log_sigma_b1'])
# sample from gaussian distribution
eps = tf.random_normal(tf.stack([tf.shape(self.x)[0], self.n_hidden]), 0, 1, dtype = tf.float32)
self.z = tf.add(self.z_mean, tf.multiply(tf.sqrt(tf.exp(self.z_log_sigma_sq)), eps))
self.reconstruction = tf.add(tf.matmul(self.z, self.weights['w2']), self.weights['b2'])
# cost
reconstr_loss = 0.5 * tf.reduce_sum(tf.pow(tf.subtract(self.reconstruction, self.x), 2.0))
latent_loss = -0.5 * tf.reduce_sum(1 + self.z_log_sigma_sq
- tf.square(self.z_mean)
- tf.exp(self.z_log_sigma_sq), 1)
self.cost = tf.reduce_mean(reconstr_loss + latent_loss)
self.optimizer = optimizer.minimize(self.cost)
init = tf.global_variables_initializer()
self.sess = tf.Session()
self.sess.run(init)
def _initialize_weights(self):
all_weights = dict()
all_weights['w1'] = tf.get_variable("w1", shape=[self.n_input, self.n_hidden],
initializer=tf.contrib.layers.xavier_initializer())
all_weights['log_sigma_w1'] = tf.get_variable("log_sigma_w1", shape=[self.n_input, self.n_hidden],
initializer=tf.contrib.layers.xavier_initializer())
all_weights['b1'] = tf.Variable(tf.zeros([self.n_hidden], dtype=tf.float32))
all_weights['log_sigma_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
def partial_fit(self, X):
cost, opt = self.sess.run((self.cost, self.optimizer), feed_dict={self.x: X})
return cost
def calc_total_cost(self, X):
return self.sess.run(self.cost, feed_dict = {self.x: X})
def transform(self, X):
return self.sess.run(self.z_mean, feed_dict={self.x: X})
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.z_mean: hidden})
def reconstruct(self, X):
return self.sess.run(self.reconstruction, feed_dict={self.x: X})
def getWeights(self):
return self.sess.run(self.weights['w1'])
def getBiases(self):
return self.sess.run(self.weights['b1'])