import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data #参数设置 input_nodes = 784 #输入节点数 output_nodes = 10 #输出节点数 layer1_nodes = 500 #隐层节点数 bitch_size = 100 #每次训练包含的数据个数 learning_rate = 0.8 #初始学习率 learning_rate_deacy = 0.99 #学习率衰减率 l2_regulation = 0.0001 #l2正则化系数 moving_rate_deacy = 0.99 #滑动模型那个衰减率 train_num = 10000
#前向传播,variable_average平均滑动模型参数 def inference(x,variable_average,w1,b1,w2,b2): if variable_average == None: layer1 = tf.nn.relu(tf.matmul(x,w1)+b1) return tf.matmul(layer1,w2)+b2 else: layer1 = tf.nn.relu(tf.matmul(x,variable_average.average(w1))+variable_average.average(b1)) return tf.matmul(layer1,variable_average.average(w2))+variable_average.average(b2)
def train(mnist): #features and labels x = tf.placeholder(tf.float32,[None,784]) y_ = tf.placeholder(tf.float32,[None,10]) #参数初始化 w1 = tf.Variable(tf.truncated_normal([input_nodes,layer1_nodes],stddev=0.1)) b1 = tf.Variable(tf.constant(0.1,shape=[layer1_nodes])) w2 = tf.Variable(tf.truncated_normal([layer1_nodes,output_nodes],stddev=0.1)) b2 = tf.Variable(tf.constant(0.1,shape=[output_nodes])) #不使用平均化滑动模型的前向传播结果 y = inference(x,None,w1,b1,w2,b2) #平均滑动模型 global_step = tf.Variable(0,trainable=False) #定义一个平均滑动模型的类 variable_average = tf.train.ExponentialMovingAverage(0.99,global_step) #定义一个平均华东模型操作,应用给所有可训练变量 variable_average_op = variable_average.apply(tf.trainable_variables()) #使用平均化滑动模型的前向传播结果 average_y = inference(x,variable_average,w1,b1,w2,b2) #交叉熵损失函数 cost_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y,labels=tf.arg_max(y_,1)) cost_entropy_mean = tf.reduce_mean(cost_entropy) #l2正则化 regulations = tf.contrib.layers.l2_regularizer(0.0001) l2_regulation = regulations(w1)+regulations(w2) #带有正则化的损失函数作为最终的损失函数 loss = cost_entropy_mean+l2_regulation #学习率衰减 learning_rate_deacy = tf.train.exponential_decay(learning_rate=0.8,global_step=global_step,decay_steps=100, decay_rate=0.99) #训练 train_step = tf.train.GradientDescentOptimizer(learning_rate_deacy).minimize(loss,global_step=global_step) #tf.group函数保证再一次迭代中,参数的train和参数的平均滑动都被执行 train_op = tf.group(train_step,variable_average_op) #准确率 correct_predict = tf.equal(tf.argmax(average_y,1),tf.argmax(y_,1)) accuracy = tf.reduce_mean(tf.cast(correct_predict,tf.float32)) #定义一个初始化的操作 init_op = tf.global_variables_initializer() with tf.Session() as sess: init_op.run() validation_feed = {x:mnist.validation.images,y_:mnist.validation.labels} #验证数据 test_feed = {x:mnist.test.images,y_:mnist.test.labels} #测试数据 for i in range(train_num): if i%1000 == 0: validation_acc = sess.run(accuracy,feed_dict=validation_feed) print('训练%d次验证集准确率是%g'%(i+1,validation_acc)) #训练数据 x_data,y_data = mnist.train.next_batch(bitch_size) train_feed = {x:x_data,y_:y_data} sess.run(train_op,feed_dict=train_feed) #测试精度 test_acc = sess.run(accuracy,feed_dict=test_feed) print('测试精度是%g'%test_acc)
mnist = input_data.read_data_sets('data',one_hot=True) train(mnist)