TensorFlow实现简单的CNN

这里，我们将采用Tensor Flow内建函数实现简单的CNN，并用MNIST数据集进行测试

第1步：加载相应的库并创建计算图会话

import numpy as np import tensorflow as tf from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets import matplotlib.pyplot as plt  #创建计算图会话 sess = tf.Session()

第2步：加载MNIST数据集，这里采用TensorFlow自带数据集，MNIST数据为28×28的图像，因此将其转化为相应二维矩阵

#数据集 data_dir = 'MNIST_data' mnist = read_data_sets(data_dir)  train_xdata = np.array([np.reshape(x,[28,28]) for x in mnist.train.images] ) test_xdata = np.array([np.reshape(x,[28,28]) for x in mnist.test.images] )  train_labels = mnist.train.labels test_labels = mnist.test.labels

第3步：设置模型参数

这里采用随机批量训练的方法，每训练10次对测试集进行测试，共迭代1500次，学习率采用指数下降的方式，初始学习率为0.1,每训练10次，学习率乘0.9，为了进行对比，后面会给出固定学习率为0.01的损失曲线图和准确率图

#设置模型参数  batch_size = 100 #批量训练图像张数 initial_learning_rate = 0.1 #学习率 global_step = tf.Variable(0, trainable=False)  ; learning_rate = tf.train.exponential_decay(initial_learning_rate,                                            global_step=global_step,                                            decay_steps=10,decay_rate=0.9)  evaluation_size = 500 #测试图像张数  image_width = 28  #图像的宽和高 image_height = 28  target_size = 10    #图像的目标为0~9共10个目标 num_channels = 1       #灰度图，颜色通道为1 generations  = 1500    #迭代500次 evaluation_step = 10  #每训练十次进行一次测试  conv1_features = 25   #卷积层的特征个数 conv2_features = 50  max_pool_size1 = 2   #池化层大小 max_pool_size2 = 2  fully_connected_size = 100 #全连接层的神经元个数

第4步：声明占位符，注意这里的目标y_target类型为int32整型

#声明占位符  x_input_shape = [batch_size,image_width,image_height,num_channels] x_input = tf.placeholder(tf.float32,shape=x_input_shape) y_target = tf.placeholder(tf.int32,shape=[batch_size])  evaluation_input_shape = [evaluation_size,image_width,image_height,num_channels] evaluation_input = tf.placeholder(tf.float32,shape=evaluation_input_shape) evaluation_target = tf.placeholder(tf.int32,shape=[evaluation_size])

第5步：声明卷积层和全连接层的权重和偏置，这里采用2层卷积层和1层隐含全连接层

#声明卷积层的权重和偏置 #卷积层1 #采用滤波器为4X4滤波器，输入通道为1，输出通道为25 conv1_weight = tf.Variable(tf.truncated_normal([4,4,num_channels,conv1_features],stddev=0.1,dtype=tf.float32)) conv1_bias = tf.Variable(tf.truncated_normal([conv1_features],stddev=0.1,dtype=tf.float32))  #卷积层2 #采用滤波器为4X4滤波器，输入通道为25，输出通道为50 conv2_weight = tf.Variable(tf.truncated_normal([4,4,conv1_features,conv2_features],stddev=0.1,dtype=tf.float32)) conv2_bias = tf.Variable(tf.truncated_normal([conv2_features],stddev=0.1,dtype=tf.float32))  #声明全连接层权重和偏置  #卷积层过后图像的宽和高 conv_output_width = image_width // (max_pool_size1 * max_pool_size2) #//表示整除 conv_output_height = image_height // (max_pool_size1 * max_pool_size2)  #全连接层的输入大小 full1_input_size = conv_output_width * conv_output_height *conv2_features  full1_weight = tf.Variable(tf.truncated_normal([full1_input_size,fully_connected_size],stddev=0.1,dtype=tf.float32)) full1_bias = tf.Variable(tf.truncated_normal([fully_connected_size],stddev=0.1,dtype=tf.float32))  full2_weight = tf.Variable(tf.truncated_normal([fully_connected_size,target_size],stddev=0.1,dtype=tf.float32)) full2_bias = tf.Variable(tf.truncated_normal([target_size],stddev=0.1,dtype=tf.float32))

第6步：声明CNN模型，这里的两层卷积层均采用Conv-ReLU-MaxPool的结构，步长为[1,1,1,1],padding为SAME

全连接层隐层神经元为100个，输出层为目标个数10

def my_conv_net(input_data):      #第一层：Conv-ReLU-MaxPool     conv1 = tf.nn.conv2d(input_data,conv1_weight,strides=[1,1,1,1],padding='SAME')     relu1 = tf.nn.relu(tf.nn.bias_add(conv1,conv1_bias))     max_pool1 = tf.nn.max_pool(relu1,ksize=[1,max_pool_size1,max_pool_size1,1],strides=[1,max_pool_size1,max_pool_size1,1],padding='SAME')      #第二层:Conv-ReLU-MaxPool     conv2 = tf.nn.conv2d(max_pool1, conv2_weight, strides=[1, 1, 1, 1], padding='SAME')     relu2 = tf.nn.relu(tf.nn.bias_add(conv2, conv2_bias))     max_pool2 = tf.nn.max_pool(relu2, ksize=[1, max_pool_size2, max_pool_size2, 1],                                strides=[1, max_pool_size2, max_pool_size2, 1], padding='SAME')      #全连接层     #先将数据转化为1*N的形式     #获取数据大小     conv_output_shape = max_pool2.get_shape().as_list()     #全连接层输入数据大小     fully_input_size = conv_output_shape[1]*conv_output_shape[2]*conv_output_shape[3]  #这三个shape就是图像的宽高和通道数     full1_input_data = tf.reshape(max_pool2,[conv_output_shape[0],fully_input_size])   #转化为batch_size*fully_input_size二维矩阵     #第一层全连接     fully_connected1 = tf.nn.relu(tf.add(tf.matmul(full1_input_data,full1_weight),full1_bias))     #第二层全连接输出     model_output = tf.nn.relu(tf.add(tf.matmul(fully_connected1,full2_weight),full2_bias))#shape = [batch_size,target_size]      return model_output  model_output = my_conv_net(x_input) test_model_output = my_conv_net(evaluation_input)

第7步：定义损失函数，这里采用softmax函数作为损失函数

#损失函数  loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=model_output,labels=y_target))

第8步：建立测评与评估函数，这里对输出层进行softmax，再通过np.argmax找出每行最大的数所在位置，再与目标值进行比对，统计准确率

#预测与评估 prediction = tf.nn.softmax(model_output) test_prediction = tf.nn.softmax(test_model_output)  def get_accuracy(logits,targets):     batch_predictions = np.argmax(logits,axis=1)#返回每行最大的数所在位置     num_correct = np.sum(np.equal(batch_predictions,targets))     return 100*num_correct/batch_predictions.shape[0]

第9步：初始化模型变量并创建优化器

#创建优化器 opt = tf.train.GradientDescentOptimizer(learning_rate=learning_rate) train_step = opt.minimize(loss)  #初始化变量 init = tf.initialize_all_variables() sess.run(init)

第10步：随机批量训练并进行绘图

#开始训练  train_loss = [] train_acc = [] test_acc = [] Learning_rate_vec = [] for i in range(generations):     rand_index = np.random.choice(len(train_xdata),size=batch_size)     rand_x = train_xdata[rand_index]     rand_x = np.expand_dims(rand_x,3)     rand_y = train_labels[rand_index]     Learning_rate_vec.append(sess.run(learning_rate, feed_dict={global_step: i}))     train_dict = {x_input:rand_x,y_target:rand_y}      sess.run(train_step,feed_dict={x_input:rand_x,y_target:rand_y,global_step:i})     temp_train_loss = sess.run(loss,feed_dict=train_dict)     temp_train_prediction = sess.run(prediction,feed_dict=train_dict)     temp_train_acc = get_accuracy(temp_train_prediction,rand_y)      #测试集     if (i+1)%evaluation_step ==0:         eval_index = np.random.choice(len(test_xdata),size=evaluation_size)         eval_x = test_xdata[eval_index]         eval_x = np.expand_dims(eval_x,3)         eval_y = test_labels[eval_index]           test_dict = {evaluation_input:eval_x,evaluation_target:eval_y}         temp_test_preds = sess.run(test_prediction,feed_dict=test_dict)         temp_test_acc = get_accuracy(temp_test_preds,eval_y)          test_acc.append(temp_test_acc)     train_acc.append(temp_train_acc)     train_loss.append(temp_train_loss)     #画损失曲线 fig = plt.figure() ax = fig.add_subplot(111) ax.plot(train_loss,'k-') ax.set_xlabel('Generation') ax.set_ylabel('Softmax Loss') fig.suptitle('Softmax Loss per Generation')  #画准确度曲线 index = np.arange(start=1,stop=generations+1,step=evaluation_step) fig2 = plt.figure() ax2 = fig2.add_subplot(111) ax2.plot(train_acc,'k-',label='Train Set Accuracy') ax2.plot(index,test_acc,'r--',label='Test Set Accuracy') ax2.set_xlabel('Generation') ax2.set_ylabel('Accuracy') fig2.suptitle('Train and Test Set Accuracy')   #画图 fig3 = plt.figure() actuals = rand_y[0:6] train_predictions = np.argmax(temp_train_prediction,axis=1)[0:6] images = np.squeeze(rand_x[0:6]) Nrows = 2 Ncols =3  for i in range(6):     ax3 = fig3.add_subplot(Nrows,Ncols,i+1)     ax3.imshow(np.reshape(images[i],[28,28]),cmap='Greys_r')     ax3.set_title('Actual: '+str(actuals[i]) +' pred: '+str(train_predictions[i]))   #画学习率 fig4 = plt.figure() ax4 = fig4.add_subplot(111) ax4.plot(Learning_rate_vec,'k-') ax4.set_xlabel('step') ax4.set_ylabel('Learning_rate') fig4.suptitle('Learning_rate')    plt.show()

下面给出固定学习率图像和学习率随迭代次数下降的图像：

首先给出固定学习率图像：

下面是损失曲线