21、TensorFlow 实现单层 Softmax 分类

一、网络结构图(5大组件)

1、数据输入设计(Input)

# 准备训练/验证/测试数据集
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

# 使用 placeholder 将数据送入网络，None 表示张量的第一个维度可以是任意长度的
with tf.name_scope('Input'):
    X = tf.placeholder(dtype=tf.float32, shape=[None, 784], name='X_placeholder')
    Y = tf.placeholder(dtype=tf.int32, shape=[None, 10], name='Y_placeholder')

---

2、前向网络设计(Inference)

with tf.name_scope('Inference'):
    W = tf.Variable(initial_value=tf.random_normal(shape=[784, 10], stddev=0.01), name='Weights')
    b = tf.Variable(initial_value=tf.zeros(shape=[10]), name='bias')
    logits = tf.matmul(X, W) + b
    Y_pred = tf.nn.softmax(logits=logits)

---

3、损失函数设计(Loss)

with tf.name_scope('Loss'):
    # 求交叉熵损失
    cross_entropy = tf.nn.softmax_cross_entropy_with_logits(labels=Y, logits=logits, name='cross_entropy')
    # 求平均
    loss = tf.reduce_mean(cross_entropy, name='loss')

---

4、参数学习算法设计(Optimization)

with tf.name_scope('Optimization'):
    optimizer = tf.train.GradientDescentOptimizer(FLAGS.learning_rate).minimize(loss)

---

5、评估节点设计(Evaluate)

with tf.name_scope('Evaluate'):
    # 返回验证集/测试集预测正确或错误的布尔值
    correct_prediction = tf.equal(tf.argmax(Y_pred, 1), tf.argmax(Y, 1))
    # 将布尔值转换为浮点数后，求平均准确率
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

# eg：[True, True, False, True] 经过数据类型转换后变成 [1, 1, 0, 1]，取平均后得到0.75

---

二、完整代码及结果

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'


# 定义一个全局对象来获取参数的值，在程序中使用(eg：FLAGS.iteration)来引用参数
FLAGS = tf.app.flags.FLAGS

# 设置训练相关参数
tf.app.flags.DEFINE_integer("iteration", 10001, "Iterations to train [1e4]")
tf.app.flags.DEFINE_integer("disp_freq", 200, "Display the current results every display_freq iterations [1e2]")
tf.app.flags.DEFINE_integer("train_batch_size", 100, "The size of batch images [128]")
tf.app.flags.DEFINE_float("learning_rate", 0.1, "Learning rate of for adam [0.01]")
tf.app.flags.DEFINE_string("log_dir", "logs", "Directory of logs.")


def main(argv=None):
    # 0、准备训练/验证/测试数据集
    mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

    # 1、数据输入设计：使用 placeholder 将数据送入网络，None 表示张量的第一个维度可以是任意长度的
    with tf.name_scope('Input'):
        X = tf.placeholder(dtype=tf.float32, shape=[None, 784], name='X_placeholder')
        Y = tf.placeholder(dtype=tf.int32, shape=[None, 10], name='Y_placeholder')

    # 2、前向网络设计
    with tf.name_scope('Inference'):
        W = tf.Variable(initial_value=tf.random_normal(shape=[784, 10], stddev=0.01), name='Weights')
        b = tf.Variable(initial_value=tf.zeros(shape=[10]), name='bias')
        logits = tf.matmul(X, W) + b
        Y_pred = tf.nn.softmax(logits=logits)

    # 3、损失函数设计
    with tf.name_scope('Loss'):
        # 求交叉熵损失
        cross_entropy = tf.nn.softmax_cross_entropy_with_logits(labels=Y, logits=logits, name='cross_entropy')
        # 求平均
        loss = tf.reduce_mean(cross_entropy, name='loss')

    # 4、参数学习算法设计
    with tf.name_scope('Optimization'):
        optimizer = tf.train.GradientDescentOptimizer(FLAGS.learning_rate).minimize(loss)

    # 5、评估节点设计
    with tf.name_scope('Evaluate'):
        # 返回验证集/测试集预测正确或错误的布尔值
        correct_prediction = tf.equal(tf.argmax(Y_pred, 1), tf.argmax(Y, 1))
        # 将布尔值转换为浮点数后，求平均准确率
        accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

    print('~~~~~~~~~~~开始执行计算图~~~~~~~~~~~~~~')
    with tf.Session() as sess:
        summary_writer = tf.summary.FileWriter(logdir=FLAGS.log_dir, graph=sess.graph)
        # 初始化所有变量
        sess.run(tf.global_variables_initializer())
        total_loss = 0
        for i in range(0, FLAGS.iteration):
            X_batch, Y_batch = mnist.train.next_batch(FLAGS.train_batch_size)
            _, loss_batch = sess.run([optimizer, loss], feed_dict={X: X_batch, Y: Y_batch})
            total_loss += loss_batch
            if i % FLAGS.disp_freq == 0:
                val_acc = sess.run(accuracy, feed_dict={X: mnist.validation.images, Y: mnist.validation.labels})
                if i == 0:
                    print('step: {}, train_loss: {}, val_acc: {}'.format(i, total_loss, val_acc))
                else:
                    print('step: {}, train_loss: {}, val_acc: {}'.format(i, total_loss/FLAGS.disp_freq, val_acc))
                total_loss = 0

        test_acc = sess.run(accuracy, feed_dict={X: mnist.test.images, Y: mnist.test.labels})
        print('test accuracy: {}'.format(test_acc))
        summary_writer.close()

# 执行main函数
if __name__ == '__main__':
    tf.app.run()


# 输出结果如下：
Extracting MNIST_data\train-images-idx3-ubyte.gz
Extracting MNIST_data\train-labels-idx1-ubyte.gz
Extracting MNIST_data\t10k-images-idx3-ubyte.gz
Extracting MNIST_data\t10k-labels-idx1-ubyte.gz
~~~~~~~~~~~开始执行计算图~~~~~~~~~~~~~~
step: 0, train_loss: 2.3216300010681152, val_acc: 0.36899998784065247
step: 200, train_loss: 0.750925962626934, val_acc: 0.8835999965667725
......
......
......
......
......
......
step: 9800, train_loss: 0.26842106945812705, val_acc: 0.9269999861717224
step: 10000, train_loss: 0.27616902984678743, val_acc: 0.9254000186920166
test accuracy: 0.9226999878883362