暑期 tensorflow 小练 mnist

暑期 tensorflow 小练 mnist

1. 下载

import numpy as npimport tensorflow as tfimport matplotlib.pyplot as plt# 下面方法二选一# 直接 API下载# from tensorflow.examples.tutorials.mnist import input_data# 文件路径下获取# import input_dataprint ("packs loaded.")

print ("download and extract mnist dataset...")# 独热编码（One-Hot Encoding）mnist = input_data.read_data_sets('data/', one_hot = True)printprint ("type of mnist is %s" % (type(mnist)))print ("number of train data is %d" % (mnist.train.num_examples))print ("number of test data is %d" % (mnist.test.num_examples))

• 独热编码：

• 独热码的作用：One-hot encoding是只存在一个1，其余全为0的n位序列。也可以称它为二元向量，二元就是里面只有0和1。通常被用来描述一个状态机的某个状态。

2. 构建网络拓扑

• 建立网络结构

# 网络拓扑 network topologyn_hidden1 = 256n_hidden2 = 128n_input = 784n_classes = 10# input outputx = tf.placeholder("float", [None, n_input])# batchsize的大小 range不限制可以为任意值，故为 Noney = tf.placeholder("float", [None, n_classes])# 网络参数stddev = 0.1weights = {    # [n_input, n_hidden1]为过渡矩阵w1的维度    'w1': tf.Variable(tf.random_normal([n_input, n_hidden1],stddev = stddev)),    'w2': tf.Variable(tf.random_normal([n_hidden1, n_hidden2],stddev = stddev)),    'out': tf.Variable(tf.random_normal([n_hidden2, n_classes],stddev = stddev))}biases = {    'b1': tf.Variable(tf.random_normal([n_hidden1])),    'b2': tf.Variable(tf.random_normal([n_hidden2])),    'out': tf.Variable(tf.random_normal([n_classes]))}print ("network ready!")

3. 网络训练相关的参数

# MLP定义def multilayer_perceptron(_X, _weights, _biases):    # x * w + b    layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(_X, weights['w1']), _biases['b1']))    layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['w2']), _biases['b2']))    return (tf.matmul(layer_2, weights['out']) + _biases['out'])# 预测pred = multilayer_perceptron(x, weights, biases)# loss & opt# 因为是分类任务 损失函数选择 softmax ；计算最后一层是softmax层的cross entropy交叉熵# tf.nn.softmax_cross_entropy_with_logits(logits, labels, name=None)# 第一个参数logits：神经网络最后一层的输出，如果有batch的话，它的大小就是[batchsize，num_classes]# 第二个参数labels：实际的标签，大小同上cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits = pred, labels = y))optm = tf.train.GradientDescentOptimizer(learning_rate = 0.001).minimize(cost)# 精度值 argmax(f(x))是使得 f(x)取得最大值 f(x0)所对应的变量 x0# tf.equal(A, B)是对比这两个矩阵的对应元素，如果相等就返回True，反之返回False，返回的值的矩阵维度和A一样# tf.cast(x, dtype, name=None) 此函数是类型转换函数corr = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))accr = tf.reduce_mean(tf.cast(corr, "float"))# initializerinit = tf.global_variables_initializer()print ("functions ready")

4. 网络训练

# 训练总共20轮training_epochs = 20# 每轮选取100个数据训练batch_size = 100# 训练结果展示比例1:4，结果共有epoch/display_step = 20/4 = 5display_step = 4# launch the graphsess = tf.Session()sess.run(init)# optimizefor epoch in range(training_epochs):    avg_cost = 0.    #batch总数为总数据量/一个batch的容量    total_batch = int(mnist.train.num_examples/batch_size)    #iteration循环    for i in range(total_batch):        # next_batch函数针对于内部提供数据集 如果是自己的数据集需要自己写函数        batch_xs, batch_ys = mnist.train.next_batch(batch_size)        feeds = {x: batch_xs, y: batch_ys}        sess.run(optm, feed_dict = feeds)        avg_cost += sess.run(cost, feed_dict = feeds)    avg_cost = avg_cost/total_batch    # display    if(epoch+1) % display_step == 0:        print ("epoch: %03d/%03d cost: %.9f" % (epoch, training_epochs, avg_cost))        feeds = {x: batch_xs, y: batch_ys}        train_acc = sess.run(accr, feed_dict = feeds)        print ("train accuracy: %.3f" % (train_acc))        feeds = {x: mnist.test.images, y: mnist.test.labels}        test_acc = sess.run(accr, feed_dict = feeds)        print ("test accuracy: %.3f" % (test_acc))print ("optimization finished!") 

---

结语

代码细节需要好好品味，其中训练的epoch、batch、display_step需要区分概念，并理解在训练过程中怎么安排的步骤。