通过tensorflow实现神经网络

TensorFlow自然语言处理的例子：利用tensorflow实现神经网络进行手写数字识别。 

import struct
import gzip
import os
from six.moves.urllib.request import urlretrieve
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt

class ManualNN:
    def __init__(self):
        self.WEIGHTS_STRING = 'weights'
        self.BIAS_STRING = 'bias'
        self.batch_size = 100
        img_width = 28
        img_height = 28
        self.input_size = img_height * img_width
        self.num_labels = 10
        tf.reset_default_graph()

    #下载数据集
    def maybe_download(self,url,filename,expected_bytes,force=False):
        if force or not os.path.exists(filename):
            print('Attempting to download:', filename)
            filename,_ = urlretrieve(url+filename,filename)
            print('\nDownload Complete!')
        statInfo = os.stat(filename)

        if statInfo.st_size == expected_bytes:
            print('Found and verified', filename)
        else:
            raise Exception(
                'Failed to verify ' + filename + '. Can you get to it with a browser?')
        return filename
    #读取数据
    def read_mnist(self,fname_img,fname_lbl):
        print('\nReading files %s and %s' % (fname_img, fname_lbl))

        with gzip.open(fname_img) as fimg:
            #采用Big Endian的方式读取四个整数，magic为magic number用于表示文件格式
            magic,num,rows,cols = struct.unpack(">IIII", fimg.read(16))
            print(num,rows,cols)
            #frombuffer将data以流的形式读入转化成ndarray对象
            img = (np.frombuffer(fimg.read(num * rows * cols), dtype=np.uint8).reshape(num, rows * cols)).astype(
                np.float32)

            img = (img - np.mean(img)) / np.std(img)
            print(img.shape)
        
        with gzip.open(fname_lbl) as flbl:
            # flbl.read(8) reads upto 8 bytes
            magic, num = struct.unpack(">II", flbl.read(8))
            lbl = np.frombuffer(flbl.read(num), dtype=np.int8)
            print('(Labels) Returned a tensor of shape: %s' % lbl.shape)
            print('Sample labels: ', lbl[:10])

        return img, lbl

    def train(self,train_inputs,train_labels,test_inputs,test_labels):
        tf_inputs = tf.placeholder(shape=[self.batch_size,self.input_size],dtype=tf.float32,name='inputs')
        tf_labels = tf.placeholder(shape=[self.batch_size,self.num_labels],dtype=tf.float32,name='labels')
        #定义作用域重用变量
        self.define_net_param()
        tf_loss = tf.reduce_mean(
            tf.nn.softmax_cross_entropy_with_logits_v2(logits=self.inference(tf_inputs), labels=tf_labels))
        tf_loss_mini = tf.train.MomentumOptimizer(momentum=0.9, learning_rate=0.01).minimize(tf_loss)

        tf_prediction = tf.nn.softmax(self.inference(tf_inputs))

        NUM_EPOCHS = 50
        session = tf.InteractiveSession()
        tf.global_variables_initializer().run()

        test_accuracy_over_time = []
        train_loss_over_time = []
        for epoch in range(NUM_EPOCHS):
            train_loss = []
            #每次训练批量的数据，100条
            for step in range(train_inputs.shape[0]//self.batch_size):
                #[100,10]的标签矩阵
                labels_one_hot = np.zeros((self.batch_size,self.num_labels),dtype=np.float32)
                #设置每条数据的标签，即取出前100个标签，train_label数据格式[2,3,5,7,8]，则根据train_label索引设置每一行何处为1，从而设置标签                labels_one_hot[np.arange(self.batch_size),train_labels[step*self.batch_size:(step+1)*self.batch_size]] = 1.0
                if epoch == 0 and step == 0:
                    print('Sample labels (one-hot)')
                    print(labels_one_hot[:10])
                    print()

                loss,_ = session.run([tf_loss,tf_loss_mini],feed_dict={
                    tf_inputs: train_inputs[step * self.batch_size: (step + 1) * self.batch_size, :],
                    tf_labels: labels_one_hot
                })

                train_loss.append(loss)


            test_accuracy = []
            # Testing Phase
            for step in range(test_inputs.shape[0] // self.batch_size):
                test_predictions = session.run(tf_prediction, feed_dict={
                    tf_inputs: test_inputs[step * self.batch_size: (step + 1) * self.batch_size, :]})
                batch_test_accuracy = self.accuracy(test_predictions,
                                               test_labels[step * self.batch_size: (step + 1) * self.batch_size])
                test_accuracy.append(batch_test_accuracy)

            print('Average train loss for the %d epoch: %.3f\n' % (epoch + 1, np.mean(train_loss)))
            train_loss_over_time.append(np.mean(train_loss))
            print('\tAverage test accuracy for the %d epoch: %.2f\n' % (epoch + 1, np.mean(test_accuracy) * 100.0))
            test_accuracy_over_time.append(np.mean(test_accuracy) * 100)

        session.close()
        #可视化
        x_axis = np.arange(len(train_loss_over_time))

        fig, ax = plt.subplots(nrows=1, ncols=2)
        fig.set_size_inches(w=25, h=5)
        ax[0].plot(x_axis, train_loss_over_time)
        ax[0].set_xlabel('Epochs', fontsize=18)
        ax[0].set_ylabel('Average train loss', fontsize=18)
        ax[0].set_title('Training Loss over Time', fontsize=20)
        ax[1].plot(x_axis, test_accuracy_over_time)
        ax[1].set_xlabel('Epochs', fontsize=18)
        ax[1].set_ylabel('Test accuracy', fontsize=18)
        ax[1].set_title('Test Accuracy over Time', fontsize=20)
        fig.savefig('mnist_stats.jpg')


    def accuracy(self,predictions,labels):
        return np.sum(np.argmax(predictions,axis=1).flatten() == labels.flatten()) / self.batch_size

    #初始化作用域重用变量layer1/w、layer1/b
    def define_net_param(self):
        #输入到隐层1
        with tf.variable_scope('layer1'):
            # tf.get_variable(self.WEIGHT_STRING,shape=[self.input_size,500],initializer=tf.random_normal_initializer(0,0.02))
            # tf.get_variable(self.BIAS_STRING,shape=[500],initializer=tf.random_uniform_initializer(0,0.01))
            tf.get_variable(self.WEIGHTS_STRING, shape=[self.input_size, 500],
                            initializer=tf.random_normal_initializer(0, 0.02))
            tf.get_variable(self.BIAS_STRING, shape=[500],
                            initializer=tf.random_uniform_initializer(0, 0.01))
        #隐层1到隐层2
        with tf.variable_scope('layer2'):
            tf.get_variable(self.WEIGHTS_STRING,shape=[500,250],initializer=tf.random_normal_initializer(0,0.02))
            tf.get_variable(self.BIAS_STRING,shape=[250],initializer=tf.random_uniform_initializer(0,0.01))
        #隐层2到输出层
        with tf.variable_scope('output'):
            tf.get_variable(self.WEIGHTS_STRING,shape=[250,10],initializer=tf.random_normal_initializer(0,0.02))
            tf.get_variable(self.BIAS_STRING,shape=[10],initializer=tf.random_uniform_initializer(0,0.01))

    def inference(self,x):
        with tf.variable_scope('layer1',reuse=True):
            w,b = tf.get_variable(self.WEIGHTS_STRING),tf.get_variable(self.BIAS_STRING)
            tf_h1 = tf.nn.relu(tf.matmul(x,w)+b,name='hidden1')
        with tf.variable_scope('layer2',reuse=True):
            w,b = tf.get_variable(self.WEIGHTS_STRING),tf.get_variable(self.BIAS_STRING)
            tf_h2 = tf.nn.relu(tf.matmul(tf_h1,w)+b,name='hidden1')
        with tf.variable_scope('output',reuse=True):
            w,b = tf.get_variable(self.WEIGHTS_STRING),tf.get_variable(self.BIAS_STRING)
            tf_logits = tf.nn.bias_add(tf.matmul(tf_h2,w),b,name='logits')

        return tf_logits

if __name__ == '__main__':
    mn = ManualNN()

    # Download data if needed
    # url = 'http://yann.lecun.com/exdb/mnist/'
    # training data
    # mn.maybe_download(url, 'train-images-idx3-ubyte.gz', 9912422)
    # mn.maybe_download(url, 'train-labels-idx1-ubyte.gz', 28881)
    # testing data
    # mn.maybe_download(url, 't10k-images-idx3-ubyte.gz', 1648877)
    # mn.maybe_download(url, 't10k-labels-idx1-ubyte.gz', 4542)

    # Read the training and testing data
    train_inputs, train_labels = mn.read_mnist('train-images-idx3-ubyte.gz', 'train-labels-idx1-ubyte.gz')
    test_inputs, test_labels = mn.read_mnist('t10k-images-idx3-ubyte.gz', 't10k-labels-idx1-ubyte.gz')
    mn.train(train_inputs,train_labels,test_inputs,test_labels)