设置神经网络结构(向前传播)
#coding:utf-8 import tensorflow as tf IMAGE_SIZE = 28 NUM_CHANNELS = 1 CONV1_SIZE = 5 CONV1_KERNEL_NUM = 32 CONV2_SIZE = 5 CONV2_KERNEL_NUM = 64 FC_SIZE = 512 OUTPUT_NODE = 10 # 初始化神经元 def get_weight(shape, regularizer): """ :param shape: 输入形状 :param regularizer: 正则化 :return: """ w = tf.Variable(tf.truncated_normal(shape,stddev=0.1)) if regularizer != None: tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w)) return w # 添加偏置 def get_bias(shape): b = tf.Variable(tf.zeros(shape)) return b # 卷积操作 def conv2d(x,w): # 全零填充,步长为1,第一个第四个参数为1,第二第三为步长 return tf.nn.conv2d(x, w, strides=[1, 1, 1, 1], padding='SAME') def max_pool_2x2(x): # 池化ksize表示池化尺寸2*2,步长为2 return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') # 搭建卷积神经网络体系 def forward(x, train, regularizer): # 创建第一层卷积层 conv1_w = get_weight([CONV1_SIZE, CONV1_SIZE, NUM_CHANNELS, CONV1_KERNEL_NUM], regularizer) # 创建第一次偏置层 conv1_b = get_bias([CONV1_KERNEL_NUM]) # 卷积操作 conv1 = conv2d(x, conv1_w) # 偏置操作,并去relu relu1 = tf.nn.relu(tf.nn.bias_add(conv1, conv1_b)) # 池化操作 pool1 = max_pool_2x2(relu1) conv2_w = get_weight([CONV2_SIZE, CONV2_SIZE, CONV1_KERNEL_NUM, CONV2_KERNEL_NUM],regularizer) conv2_b = get_bias([CONV2_KERNEL_NUM]) conv2 = conv2d(pool1, conv2_w) relu2 = tf.nn.relu(tf.nn.bias_add(conv2, conv2_b)) pool2 = max_pool_2x2(relu2) # 获取经过第二层池化后的张量形状 pool_shape = pool2.get_shape().as_list() nodes = pool_shape[1] * pool_shape[2] * pool_shape[3] # 改变张量形状,变为一维张量 reshaped = tf.reshape(pool2, [pool_shape[0], nodes]) # 经过第一层全连接层,正则化 fc1_w = get_weight([nodes, FC_SIZE], regularizer) fc1_b = get_bias([FC_SIZE]) fc1 = tf.nn.relu(tf.matmul(reshaped, fc1_w) + fc1_b) # 经过第二层全连接层使用0.5的dropout避免过拟合 if train: fc1 = tf.nn.dropout(fc1, 0.5) # 经过 fc2_w = get_weight([FC_SIZE, OUTPUT_NODE], regularizer) fc2_b = get_bias([OUTPUT_NODE]) y = tf.matmul(fc1, fc2_w) + fc2_b return y
训练(向后传播)
# coding:utf-8 import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data import mnist_lenet5_forward import os import numpy as np BATCH_SIZE = 100 # 100张图片为一批次 LEARNING_RATE_BASE = 0.005 # 学习率 LEARNING_RATE_DECAY = 0.99 # 学习衰减率 REGULARIZER = 0.0001 # 正则化参数 STEPS = 50000 # 训练步长 MOVING_AVERAGE_DECAY = 0.99 # 滑动平均衰减率 MODEL_SAVE_PATH = "./model/" # 模型保存路径 MODEL_NAME = "mnist_model" # 模型保存名字 def backward(mnist): # 创建输入数据占位符 x = tf.placeholder(tf.float32, [ BATCH_SIZE, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.NUM_CHANNELS]) # 输出数据占位符 y_ = tf.placeholder(tf.float32, [None, mnist_lenet5_forward.OUTPUT_NODE]) # 用向前传播,计算输出数据 y = mnist_lenet5_forward.forward(x, True, REGULARIZER) # 记录步长 global_step = tf.Variable(0, trainable=False) # 使用交叉熵代价函数 ce = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1)) #注意ce为向量 # 取平均得到数 cem = tf.reduce_mean(ce) # 加入正则化系数 loss = cem + tf.add_n(tf.get_collection('losses')) # 指数衰减学习率,计算公式为 learning_rate = learning_rate * learning_rate_decay^(global_step/decay_step) learning_rate = tf.train.exponential_decay( LEARNING_RATE_BASE, global_step, mnist.train.num_examples / BATCH_SIZE, LEARNING_RATE_DECAY, staircase=True) # 梯度下降优化 train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step) # 下面是一个滑动平均模型 ema = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step) ema_op = ema.apply(tf.trainable_variables()) with tf.control_dependencies([train_step, ema_op]): train_op = tf.no_op(name='train') # 保存模型 saver = tf.train.Saver() # 启动会话 with tf.Session() as sess: # 初始化 init_op = tf.global_variables_initializer() sess.run(init_op) # 通过checkpoint来定位模型 ckpt = tf.train.get_checkpoint_state(MODEL_SAVE_PATH) # 如果已经训练过模型,加载训练好的模型 if ckpt and ckpt.model_checkpoint_path: saver.restore(sess, ckpt.model_checkpoint_path) # 开始训练 for i in range(STEPS): # 每次取一批次的数据和标注 xs, ys = mnist.train.next_batch(BATCH_SIZE) reshaped_xs = np.reshape(xs, ( BATCH_SIZE, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.NUM_CHANNELS)) _, loss_value, step = sess.run([train_op, loss, global_step], feed_dict={x: reshaped_xs, y_: ys}) # 每100次输出一次,并保存模型 if i % 100 == 0: print("After %d training step(s), loss on training batch is %g." % (step, loss_value)) saver.save(sess, os.path.join(MODEL_SAVE_PATH, MODEL_NAME), global_step=global_step) def main(): # 读取模型,one_hot代表向量话 mnist = input_data.read_data_sets("./data/", one_hot=True) backward(mnist) if __name__ == '__main__': main()
预测
import time import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data import mnist_lenet5_forward import mnist_lenet5_backward import numpy as np TEST_INTERVAL_SECS = 5 # 测试函数 def test(mnist): # 创建一个图 with tf.Graph().as_default() as g: x = tf.placeholder(tf.float32, [mnist.test.num_examples, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.NUM_CHANNELS, ] ) y_ = tf.placeholder(tf.float32, [None, mnist_lenet5_forward.OUTPUT_NODE]) # 使用向前传播计算结果,第二参数代表训练过程,第三个参数代表不使用正则化 y = mnist_lenet5_forward.forward(x, False, None) # 使用滑动平均模型 ema = tf.train.ExponentialMovingAverage(mnist_lenet5_backward.MOVING_AVERAGE_DECAY) ema_restore = ema.variables_to_restore() # 保存模型 saver = tf.train.Saver(ema_restore) # 计算训练率 correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1)) # 比较预测值与实际值是否相等,为逻辑向量 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) # 求和取平均得到最终结果 while True: with tf.Session() as sess: # 查找check_point ckpt = tf.train.get_checkpoint_state(mnist_lenet5_backward.MODEL_SAVE_PATH) # 找到了,加载这个模型 if ckpt and ckpt.model_checkpoint_path: saver.restore(sess, ckpt.model_checkpoint_path) # 加载对应的迭代步长 global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1] # 将图片用四维张量的形式传入 reshaped_x = np.reshape(mnist.test.images, ( mnist.test.num_examples, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.IMAGE_SIZE, mnist_lenet5_forward.NUM_CHANNELS)) # 计算准确率 accuracy_score = sess.run(accuracy, feed_dict={x: reshaped_x, y_: mnist.test.labels}) print("After %s training step(s), test accuracy = %g" % (global_step, accuracy_score)) else: print('No checkpoint file found') return # 每个5秒搜索新的模型 time.sleep(TEST_INTERVAL_SECS) def main(): mnist = input_data.read_data_sets("./data/", one_hot=True) test(mnist) if __name__ == '__main__': main()