TensorFlow写卷积神经网络_数据集MNist_详细说明

导入包和MNist from __future__ import division, print_function #__future__的作用是升级py2到py3的一些用法，division：精确除法 import tensorflow as tf import matplotlib.pyplot as plt import numpy as np from tensorflow.examples.tutorials.mnist import input_data #从当前文件夹中找到MNist数据，“”中为文件路径 #MNist中，每一幅图片都是由28*28像素矩阵构成的，（28*28=784即784个像素点） mnist = input_data.read_data_sets("datasets/MNIST_data/", one_hot = True) 观察数据 def train_size(num): print('Training = ' + str(mnist.train.images.shape)) print('-------------------------------------------') x_train = mnist.train.images[:num, :] print('x_Training = ' + str(x_train.shape)) y_train = mnist.train.labels[:num, :] print('y_Training = ' + str(y_train.shape)) print('') return x_train, y_train def test_size(num): print('Test = ' + str(mnist.train.images.shape)) print('-------------------------------------------') x_test = mnist.test.images[:num, :] print('x_Test = ' + str(x_test.shape)) y_test = mnist.test.labels[:num, :] print('y_Training = ' + str(y_test.shape)) print('') return x_test, y_test 显示图 def display_digit(num): print(y_train[num]) label = y_train[num].argmax(axis = 0) image = x_train[num].reshape([28, 28]) #将784位一组改一下形状写成28*28变成容易观察的图形 plt.title('Example: %d Label: %d' % (num, label)) plt.imshow(image, cmap = plt.get_cmap('gray_r')) #cmap：可视化色彩图。gray:【0：黑；255：白】，gray_r：【0:白，255：黑】 plt.show() 显示重新塑形的矩阵 def display_mult_flat(start, stop): images = x_train[start].reshape([1,784]) for i in range(start + 1, stop): #将多幅图形连接在一起显示，numpy.concatenate((a1, a2, ...), axis=0)（按轴axis连接array组成一个新的array，default=0：逐行） images = np.concatenate((images, x_train[i].reshape([1, 784]))) plt.imshow(images, cmap = plt.get_cmap('gray_r')) plt.show() x_train, y_train = train_size(55000) #随机显示一个数字。np.random.randint（low, high=None, size=None, dtype='l'）。函数的作用是，返回一个随机整型数，范围从低（包括）到高（不包括），即[low, high)。如果没有写参数high的值，则返回[0,low)的值。 display_digit(np.random.randint(0, x_train.shape[0])) #x_train.shape是55000*784 display_mult_flat(0, 400) Training = (55000, 784) ------------------------------------------- x_Training = (55000, 784) y_Training = (55000, 10) [0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]

设置学习参数 learning_rate = 0.001 training_iters = 500 batch_size = 128 display_step = 10 n_input = 784 n_classes = 10 dropout = 0.85 计算图输入的占位 x = tf.placeholder(tf.float32, [None,n_input]) #输入是none个，784列的矩阵 y = tf.placeholder(tf.float32, [None,n_classes]) keep_prob = tf.placeholder(tf.float32) #dropout参数，元素被保留下来的概率 建立网络 关键命令用法 x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding=‘SAME’)
其中strides=[B, H, W, C]分别代表在batch、矩阵的高度、矩阵的宽度、输入的通道四个维度移动的步长。
padding：SAME：补零；VAILID：原图 tf.nn.max_pool（x, ksize=[1, k, k, 1], strides=[1, k, k, 1], padding=‘SAME’)
其中ksize是池化窗口的大小，由于一般不在batch和channel维度上做池化，所以这里为1 x.get_shape()，只有tensor才可以使用这种方法，返回的是一个TensorShape对象。as_list()将元组转换成list. tf.nn.dropout()：tensorflow里面为了防止或减轻过拟合而使用的函数，它一般用在全连接层。Dropout就是在不同的训练过程中随机扔掉一部分神经元。也就是让某个神经元的激活值以一定的概率p，让其停止工作，这次训练过程中不更新权值，也不参加神经网络的计算。但是它的权重得保留下来（只是暂时不更新而已），因为下次样本输入时它可能又得工作了 网络结构定义 第一个卷积层：5 * 5卷积核，1个输入（[-1,28,28,1]中的1），32个输出 第二个卷积层：5 * 5卷积核，32个输入（上一层输出），64个输出 全连接层：7×7×64 个输入和 1024 个输出
（输出图像的个数为认为设定） 卷积层的结构理解：

#给定步幅的卷积层，激活函数是 ReLU，padding 设定为 SAME 模式： def conv2d(x, W, b, strides=1): x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding='SAME') x = tf.nn.bias_add(x, b) #一个叫bias的向量加到一个叫value的矩阵上，是向量与矩阵的每一行进行相加，得到的结果和value矩阵大小相同。 return tf.nn.relu(x) #定义一个输入是 x 的 maxpool 层，卷积核为 ksize 并且 padding 为 SAME： def maxpool2d(x, k=2): return tf.nn.max_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1], padding='SAME') #定义convnet: 卷积-卷积-全连接-dropout def conv_net(x, weights, biases, dropout): x = tf.reshape(x, shape=[-1, 28, 28, 1]) conv1 = conv2d(x, weights['wc1'], biases['bc1']) #卷积 conv1 = maxpool2d(conv1, k=2) #池化 conv2 = conv2d(conv1, weights['wc2'], biases['bc2']) #conv conv2 = maxpool2d(conv2, k=2) #pooling #两层卷积之后的结果得到了：64个7*7大小矩阵（或者称为图片） #将卷积层的输出作为全连接层的输入，重新调整形状：7*7*64个元素为一列 fc1 = tf.reshape(conv2, [-1, weights['wd1'].get_shape().as_list()[0]]) fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1']) #fully connected fc1 = tf.nn.relu(fc1) fc1 = tf.nn.dropout(fc1, dropout) out = tf.add(tf.matmul(fc1, weights['out']), biases['out']) return out 定义网络的权重和偏置 weights = { 'wc1' : tf.Variable(tf.random_normal([5, 5, 1, 32])), 'wc2' : tf.Variable(tf.random_normal([5, 5, 32, 64])), 'wd1' : tf.Variable(tf.random_normal([7*7*64, 1024])), 'out' : tf.Variable(tf.random_normal([1024, n_classes])), } biases = { 'bc1' : tf.Variable(tf.random_normal([32])), 'bc2' : tf.Variable(tf.random_normal([64])), 'bd1' : tf.Variable(tf.random_normal([1024])), 'out' : tf.Variable(tf.random_normal([n_classes])), } 搭建模型 pred = conv_net(x, weights, biases, keep_prob) cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y)) optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost) correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) init = tf.global_variables_initializer() train_loss = [] train_acc = [] test_acc = [] with tf.Session() as sess: sess.run(init) step = 1 while step <= training_iters: batch_x, batch_y = mnist.train.next_batch(batch_size) #batch_x = np.reshape(batch_x, [-1, 28, 28, 1]) sess.run(optimizer, feed_dict={x: batch_x, y: batch_y, keep_prob: dropout}) #每隔display_step这些步显示一次 if step % display_step == 0: loss_train, acc_train = sess.run([cost, accuracy], feed_dict={x: batch_x, y: batch_y, keep_prob: dropout}) print("Iter" + str(step) + ",Minibatch Loss = " + \ "{:.2f}".format(loss_train) + ", Training Accuracy = " + \ "{:.2f}".format(acc_train)) acc_test = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels , keep_prob: 1}) print("Testing Accuracy = " + "{:.2f}".format(acc_train)) train_loss.append(loss_train) train_acc.append(acc_train) test_acc.append(acc_test) step += 1

运行结果：

Iter10,Minibatch Loss = 43936.12, Training Accuracy = 0.14 Testing Accuracy = 0.14 Iter20,Minibatch Loss = 27234.01, Training Accuracy = 0.30 Testing Accuracy = 0.30 Iter30,Minibatch Loss = 15954.67, Training Accuracy = 0.50 Testing Accuracy = 0.50 Iter40,Minibatch Loss = 15314.48, Training Accuracy = 0.50 Testing Accuracy = 0.50 Iter50,Minibatch Loss = 11097.48, Training Accuracy = 0.59 Testing Accuracy = 0.59 Iter60,Minibatch Loss = 10651.96, Training Accuracy = 0.62 Testing Accuracy = 0.62 Iter70,Minibatch Loss = 8680.00, Training Accuracy = 0.71 Testing Accuracy = 0.71 Iter80,Minibatch Loss = 7176.53, Training Accuracy = 0.69 Testing Accuracy = 0.69 Iter90,Minibatch Loss = 5430.07, Training Accuracy = 0.71 Testing Accuracy = 0.71 Iter100,Minibatch Loss = 6548.58, Training Accuracy = 0.72 Testing Accuracy = 0.72 Iter110,Minibatch Loss = 3333.26, Training Accuracy = 0.81 Testing Accuracy = 0.81 Iter120,Minibatch Loss = 4729.37, Training Accuracy = 0.79 Testing Accuracy = 0.79 Iter130,Minibatch Loss = 2650.15, Training Accuracy = 0.84 Testing Accuracy = 0.84 Iter140,Minibatch Loss = 4700.38, Training Accuracy = 0.83 Testing Accuracy = 0.83 Iter150,Minibatch Loss = 1974.79, Training Accuracy = 0.85 Testing Accuracy = 0.85 Iter160,Minibatch Loss = 2988.53, Training Accuracy = 0.81 Testing Accuracy = 0.81 Iter170,Minibatch Loss = 3457.32, Training Accuracy = 0.85 Testing Accuracy = 0.85 Iter180,Minibatch Loss = 4411.01, Training Accuracy = 0.84 Testing Accuracy = 0.84 Iter190,Minibatch Loss = 3413.62, Training Accuracy = 0.84 Testing Accuracy = 0.84 Iter200,Minibatch Loss = 2411.36, Training Accuracy = 0.87 Testing Accuracy = 0.87 Iter210,Minibatch Loss = 2040.41, Training Accuracy = 0.85 Testing Accuracy = 0.85 Iter220,Minibatch Loss = 2574.14, Training Accuracy = 0.86 Testing Accuracy = 0.86 Iter230,Minibatch Loss = 3515.91, Training Accuracy = 0.84 Testing Accuracy = 0.84 Iter240,Minibatch Loss = 1911.09, Training Accuracy = 0.90 Testing Accuracy = 0.90 Iter250,Minibatch Loss = 2576.21, Training Accuracy = 0.85 Testing Accuracy = 0.85 Iter260,Minibatch Loss = 2118.93, Training Accuracy = 0.87 Testing Accuracy = 0.87 Iter270,Minibatch Loss = 1865.62, Training Accuracy = 0.84 Testing Accuracy = 0.84 Iter280,Minibatch Loss = 2111.70, Training Accuracy = 0.88 Testing Accuracy = 0.88 Iter290,Minibatch Loss = 2465.34, Training Accuracy = 0.84 Testing Accuracy = 0.84 Iter300,Minibatch Loss = 1801.69, Training Accuracy = 0.86 Testing Accuracy = 0.86 Iter310,Minibatch Loss = 1379.81, Training Accuracy = 0.92 Testing Accuracy = 0.92 Iter320,Minibatch Loss = 1124.99, Training Accuracy = 0.92 Testing Accuracy = 0.92 Iter330,Minibatch Loss = 1956.47, Training Accuracy = 0.88 Testing Accuracy = 0.88 Iter340,Minibatch Loss = 1452.55, Training Accuracy = 0.86 Testing Accuracy = 0.86 Iter350,Minibatch Loss = 1092.12, Training Accuracy = 0.89 Testing Accuracy = 0.89 Iter360,Minibatch Loss = 928.34, Training Accuracy = 0.90 Testing Accuracy = 0.90 Iter370,Minibatch Loss = 1805.11, Training Accuracy = 0.88 Testing Accuracy = 0.88 Iter380,Minibatch Loss = 1248.89, Training Accuracy = 0.92 Testing Accuracy = 0.92 Iter390,Minibatch Loss = 1489.37, Training Accuracy = 0.88 Testing Accuracy = 0.88 Iter400,Minibatch Loss = 1462.85, Training Accuracy = 0.94 Testing Accuracy = 0.94 Iter410,Minibatch Loss = 926.76, Training Accuracy = 0.91 Testing Accuracy = 0.91 Iter420,Minibatch Loss = 1127.02, Training Accuracy = 0.95 Testing Accuracy = 0.95 Iter430,Minibatch Loss = 786.22, Training Accuracy = 0.95 Testing Accuracy = 0.95 Iter440,Minibatch Loss = 1727.24, Training Accuracy = 0.92 Testing Accuracy = 0.92 Iter450,Minibatch Loss = 920.39, Training Accuracy = 0.89 Testing Accuracy = 0.89 Iter460,Minibatch Loss = 1150.05, Training Accuracy = 0.92 Testing Accuracy = 0.92 Iter470,Minibatch Loss = 1181.85, Training Accuracy = 0.92 Testing Accuracy = 0.92 Iter480,Minibatch Loss = 1799.36, Training Accuracy = 0.89 Testing Accuracy = 0.89 Iter490,Minibatch Loss = 1111.24, Training Accuracy = 0.92 Testing Accuracy = 0.92 Iter500,Minibatch Loss = 602.50, Training Accuracy = 0.94 Testing Accuracy = 0.94 eval_indices = range(0, training_iters, display_step) #画Loss图像 plt.plot(eval_indices, train_loss, 'k-') plt.title('Softmax Loss per iteration') plt.xlabel('Itertion') plt.ylabel('Softmax Loss') plt.show() #画acc图像 plt.plot(eval_indices, train_acc, 'k-', label='Train Set Accuracy') plt.plot(eval_indices, test_acc, 'r-', label='Test Set Accuracy') plt.title('Train and Test Accuracy') plt.xlabel('Generation') plt.ylabel('Accuracy') plt.legend(loc='lower right') plt.show()

参考文章

作者：pycolar

数据 mnist 卷积神经网络 tensorflow 神经网络 卷积