autoencoder

import numpy as np
import sklearn.preprocessing as prep
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

#initialize parameter with xavier
#Yoshua Bengio indecate that xavier, which is 0-average and 2/(num_in+num_out)-variance, is a good initialal pamameter method
def xavier_init(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)

#autoencoder Class
class AdditvieGussianNoiseAutoencoder(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_function = transfer_function
        self.scale = tf.placeholder(tf.float32)
        self.training_scale = scale
        network_weights = self._initialize_weights()
        self.weights = network_weights
        self.x = tf.placeholder(tf.float32, [None,self.n_input])
        self.hidden = self.transfer_function(tf.add(tf.matmul(self.x+scale*tf.random_normal((n_input,)),self.weights['w1']),self.weights['b1']))
        self.reconstruction = tf.add(tf.matmul(self.hidden, self.weights['w2']),self.weights['b2'])
        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)

#initialize parameter
#since w1 has an activation function, it should be initialized by avixer. otherwise, zeros are ok
    def _initialize_weights(self):
        all_weights = dict()
        all_weights['w1'] = tf.Variable(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

#batch training
#calculate the batch cost and then optimize the network
    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

#calculate the total cost
    def calc_total_cost(self, X):
        return self.sess.run(self.cost, feed_dict= {self.x: X, self.scale: self.training_scale})

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

#transform the hidden_output to output
    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})

#from input to output
    def reconstruct(self,X):
        return self.sess.run(self.reconstruction, feed_dict={self.x:X, self.scale: self.training_scale})

#get hidden-layer weights w1
    def getweights(self):
        return self.sess.run(self.weights['w1'])

#get hidden-layer biases b1
    def getBiases(self):
        return self.sess.run(self.weights['b1'])

#load the MNIST_data_set
mnist = input_data.read_data_sets('/home/star/MNIST_data/',one_hot= True)

#standard the input with StandardScaler
def standard_scale(X_train, X_test):
    preprocessor = prep.StandardScaler().fit(X_train)
    X_train = preprocessor.transform(X_train)
    X_test = preprocessor.transform(X_test)
    return X_train, X_test

#choose a start_index of batch randomly
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)]

#Basic parameter
X_train, X_test = standard_scale(mnist.train.images, mnist.test.images)
n_samples = int(mnist.train.num_examples)
training_epochs = 20
batch_size = 128
display_step = 1

#create an instance
autoencoder = AdditvieGussianNoiseAutoencoder(n_input= 784, n_hidden= 200, transfer_function= tf.nn.softplus, optimizer= tf.train.AdamOptimizer(learning_rate= 0.001), scale= 0.01)

#training
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 = autoencoder.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))
#testing
print('Total cost:' + str(autoencoder.calc_total_cost(X_test)))