『TensorFlow』讀書筆記_降噪自編碼器

『TensorFlow』降噪自編碼器設計

 之前學習過的程式碼，又敲了一遍，新的收穫也還是有的，因為這次註釋寫的比較詳盡，所以再次記錄一下，具體的相關知識查閱之前寫的文章即可（見上面連結）。

# Author : Hellcat
# Time   : 2017/12/6

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

def xavier_init(fan_in,fan_out, constant = 1):
    '''
    xavier 權重初始化方式
    :param fan_in: 行數
    :param fan_out: 列數
    :param constant: 常數權重，調節初始化範圍的倍數
    :return: 初始化後的權重tensor
    '''
    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)

class AdditiveGaussianNoiseAutoencoder():

    def __init__(self, n_input, n_hidden,
                 transfer_function=tf.nn.softplus,
                 optimizer=tf.train.AdamOptimizer(),scale=0.1):
        '''
        初始化自編碼器
        :param n_input: 輸入層結點數
        :param n_hidden: 隱藏層節點數
        :param transfer_function: 隱藏層啟用函式
        :param optimizer: 優化器，是例項化的物件
        :param scale: 高斯噪聲係數
        '''
        self.n_input = n_input
        self.n_hidden = n_hidden
        self.transfer = transfer_function
        self.scale = tf.placeholder(tf.float32) # 實際網路中呼叫的
        self.training_scale = scale # 訓練用噪聲係數
        network_weights = self._initialize_weights()
        self.weights = network_weights

        self.x = tf.placeholder(tf.float32, [None, self.n_input])
        self.hidden = \
            self.transfer(
                tf.add(
                    tf.matmul(
                        self.x + self.scale * tf.random_normal((n_input,)),
                        self.weights['w1']),
                    self.weights['b1']))

        # 重建部分沒有使用啟用函式
        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))
        # 可以將類的例項過程作為實參傳入函式
        self.optimizer = optimizer.minimize(self.cost)

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

    def _initialize_weights(self):
        '''
        初始化全部變數
        :return: 裝有變數的字典
        '''
        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

    def partial_fit(self, X):
        '''
        進行單次訓練並返回loss
        :param X: 訓練資料
        :return: 本次損失函式值
        '''
        cost, opt = self.sess.run((self.cost, self.optimizer),
                                  feed_dict={self.x:X, self.scale:self.training_scale})
        return cost

    def calc_totul_cost(self, X):
        '''
        計算損失函式，不觸發訓練
        :param X: 訓練資料
        :return: 損失函式
        '''
        return self.sess.run(self.cost, feed_dict={self.x:X, self.scale:self.training_scale})

    def transform(self, X):
        '''
        返回隱藏層輸出結果，目的是獲取抽象後的特徵
        :param X: 訓練資料
        :return: 隱藏層輸出
        '''
        return self.sess.run(self.hidden, feed_dict={self.x:X, self.scale:self.training_scale})

    def generate(self, hidden=None):
        '''
        通過隱藏層特徵重建
        :param hidden: 隱藏層特徵
        :return: 重建資料
        '''
        if hidden is None:
            hidden = np.random.normal(size=[self.n_input])
        return self.sess.run(self.reconstruction, feed_dict={self.hidden:hidden})

    def reconstruct(self,X):
        '''
        從原始資料重建
        :param X: 訓練資料
        :return: 重建資料
        '''
        return self.sess.run(self.reconstruction,
                             feed_dict={self.x:X, self.scale:self.training_scale})

    def getWeights(self):
        '''
        獲取引數值
        :return: 隱藏層權重
        '''
        return self.sess.run(self.weights['w1'])

    def getBaises(self):
        '''
        獲取引數值
        :return: 隱藏層偏置
        '''
        return self.sess.run(self.weights['b1'])

def standard_scale(X_train, X_test):
    '''
    標準化資料
    :param X_train: 訓練資料
    :param X_test: 測試資料
    :return: 標準化之後的訓練、測試資料
    '''
    preprocessor = prep.StandardScaler().fit(X_train)
    X_train = preprocessor.transform(X_train)
    X_test = preprocessor.transform(X_test)
    return X_train, X_test

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)]

if __name__ == '__main__':
    mnist = input_data.read_data_sets('../../../Mnist_data/',one_hot=True)
    X_train, X_test = standard_scale(mnist.train.images, mnist.test.images)

    n_samples = int(mnist.train.num_examples)
    train_epochs = 20
    batch_size = 20
    display_step = 1

    autoencoder = AdditiveGaussianNoiseAutoencoder(
        n_input=784,
        n_hidden=200,
        transfer_function=tf.nn.softplus,
        optimizer=tf.train.AdamOptimizer(learning_rate=0.001),
        scale=0.01)

    for epoch in range(train_epochs):
        avg_cost = 0.
        totu_batch = int(n_samples / batch_size)
        for i in range(totu_batch):
            batch_xs = get_random_block_from_data(X_train, batch_size)

            # 單資料塊訓練並計算損失函式
            cost = autoencoder.partial_fit(batch_xs)
            avg_cost += cost / n_samples * batch_size

            if epoch % display_step == 0:
                print('epoch : %04d, cost = %.9f' % (epoch + 1,avg_cost))

            # 計算測試集上的cost
    print('Total coat:',str(autoencoder.calc_totul_cost(X_test)))

部分輸出如下：

……

epoch : 0020, cost = 1509.876800515
epoch : 0020, cost = 1510.107261985
epoch : 0020, cost = 1510.332509055
epoch : 0020, cost = 1510.551538707
Total coat: 768927.0

1.xavier初始化權重方法

2.函式實參可以是class()，即例項化的類