变分自动编码器

变分自动编码器

对图片编码
对图片进行压缩，保留关键信息
对图片解码
学习训练图片之后，自动产生相似的图片

#tensorflow 概率分布计算
import tensorflow as tf
a=[[1.1],[2.1],[3.2],[4.5],[5.0]]
inputs=tf.Variable(a)
hidden = tf.layers.dense(inputs, 100, tf.nn.relu)
#均值
mean = tf.layers.dense(hidden, 10, None)
#标准差
stddev = tf.layers.dense(hidden, 10, tf.nn.softplus)
#分布
tfd = tf.contrib.distributions
#取样
dist = tfd.MultivariateNormalDiag(mean, stddev)
samples = dist.sample()
dist.log_prob(samples)
dist.prob(samples)

#概率分布只能接受浮点运算类型
dist = tfd.MultivariateNormalDiag([0.0,10.0], [1.0,5.0])
sess=tf.Session()
for i in range(10):
  samples = dist.sample()
  print(sess.run(samples))
  print(sess.run(dist.prob(samples)))
  
#回归
loss = -dist.log_prob(label)  
optimize = tf.train.AdamOptimizer().minimize(loss)
#分类
dist = tfd.Categorical(logit)
loss = -dist.log_prob(label)  
optimize = tf.train.AdamOptimizer().minimize(loss)

#tensorflow实现变分自动编码器
import numpy as np
import IPython.display as display
import tensorflow as tf

import matplotlib.pyplot as plt
import matplotlib as mpl

def make_prior(code_size=100):
  mean, stddev = tf.zeros([code_size]), tf.ones([code_size])
  return tfd.MultivariateNormalDiag(mean, stddev)

def make_encoder(images, code_size=100):
  images = tf.layers.flatten(images)
  hidden = tf.layers.dense(images, 128, tf.nn.relu)
  mean = tf.layers.dense(hidden, code_size)
  stddev = tf.layers.dense(hidden, code_size, tf.nn.softplus)
  return tfd.MultivariateNormalDiag(mean, stddev)

def make_decoder(code, data_shape=[28, 28]):
  hidden = tf.layers.dense(code, 128, tf.nn.relu)
  logit = tf.layers.dense(hidden, np.prod(data_shape))
  logit = tf.reshape(logit, [-1] + data_shape)
  return tfd.Independent(tfd.Bernoulli(logit), len(data_shape))

tfd = tf.contrib.distributions
#用minist数据集
#输入图片
images = tf.placeholder(tf.float32, [None, 28, 28])
#初始先验概率
prior = make_prior()
#编码
posterior = make_encoder(images)
#解码
dist = make_decoder(posterior.sample())
#取样
samples = make_decoder(prior.sample(10)).mean()
#损失函数
elbo =tf.reduce_mean(dist.log_prob(images)) - tf.reduce_mean(tfd.kl_divergence(posterior, prior))
optimize = tf.train.AdamOptimizer(learning_rate=0.01).minimize(-elbo)

mnist = tf.keras.datasets.mnist

#读取数据
(x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train=x_train/255.0

sess=tf.Session()
sess.run(tf.global_variables_initializer()) 
           
for i in range(10):
  _, loss=sess.run([optimize,-elbo], feed_dict={images:x_train})
  print(loss)
  
https://drive.google.com/open?id=1tDRo8ZGEqWNtXK516_8ZVB5wO_BrB8Ey