convolutional_network.py 注释

教案

'''
A Convolutional Network implementation example using TensorFlow library.
This example is using the MNIST database of handwritten digits
(http://yann.lecun.com/exdb/mnist/)
Author: Aymeric Damien
Project: https://github.com/aymericdamien/TensorFlow-Examples/
'''
from __future__ import print_function
import tensorflow as tf
# Import MNIST data
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)
# Parameters
# 参数
learning_rate = 0.001       # 学习率
training_iters = 200000     # 训练次数
batch_size = 128            # 批处理大小，即每次输入多少个数据进入网络
display_step = 10           # 每隔多少步输出一次 log
# Network Parameters
# 卷积神经网络的参数
n_input = 784 # MNIST data input (img shape: 28*28) 
              # 输入特征维度，因为MNIST数据集是28x28的图像，所以维度是28x28=784
n_classes = 10 # MNIST total classes (0-9 digits)
               # 分类数，因为是区分0-9的数字，所以类别有10个
dropout = 0.75 # Dropout, probability to keep units
               # 优化网络的参数，代表每个神经元每次计算有 25% 的几率被忽略
# tf Graph input
# 输入数据 x: 特征， y: 人工标记类别
x = tf.placeholder(tf.float32, [None, n_input])
y = tf.placeholder(tf.float32, [None, n_classes])
keep_prob = tf.placeholder(tf.float32) #dropout (keep probability)
# Create some wrappers for simplicity
# 建立一个简化的建立卷积层的函数
# x: 输入特征值
# W：权重
# b: bias
# strides: 卷积核移动步长
def conv2d(x, W, b, strides=1):
    # Conv2D wrapper, with bias and relu activation
    x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding='SAME')
    x = tf.nn.bias_add(x, b)
    return tf.nn.relu(x)
# 定义一个简化的建立下采样层的函数
# x: 被下采样的数据，4-D tensor [块大小，长，宽，通道数(灰度就是1，rgb就是3)]
# k：下采样核的尺寸，2 代表 2x2
def maxpool2d(x, k=2):
    # MaxPool2D wrapper
    return tf.nn.max_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],
                          padding='SAME')
# Create model
# 建立卷积神经网络模型
def conv_net(x, weights, biases, dropout):
    # Reshape input picture
    # reshape 输入的图像，-1 代表让 tf 自动计算，这里是图像的数量
    x = tf.reshape(x, shape=[-1, 28, 28, 1])
    # ----------卷积网络----------
    # Convolution Layer
    # 加一个卷积层
    conv1 = conv2d(x, weights['wc1'], biases['bc1'])
    # Max Pooling (down-sampling)
    # 加一个下采样层
    conv1 = maxpool2d(conv1, k=2)
    # Convolution Layer
    # 再加一个卷积层
    conv2 = conv2d(conv1, weights['wc2'], biases['bc2'])
    # Max Pooling (down-sampling)
    # 再加一个下采样层
    conv2 = maxpool2d(conv2, k=2)
    # ----------全连接层----------
    # Fully connected layer
    # 建立全连接层
    # Reshape conv2 output to fit fully connected layer input
    # 把上一层输出的数据 reshape 到适应神经网络的输入
    fc1 = tf.reshape(conv2, [-1, weights['wd1'].get_shape().as_list()[0]])
    # wx + b
    fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1'])
    # 建立全连接层，激活函数为 relu
    fc1 = tf.nn.relu(fc1)
    # Apply Dropout
    # 丢弃参数
    fc1 = tf.nn.dropout(fc1, dropout)
    # ----------输出层----------
    # Output, class prediction
    # 输出层，用于分类。wx + b
    out = tf.add(tf.matmul(fc1, weights['out']), biases['out'])
    return out
# Store layers weight & bias
# 存储所有权重的 dict
weights = {
    # 5x5 conv, 1 input, 32 outputs
    # 第一个卷积层的权重，卷积核的尺寸是 5×5，1个输入，输出定为32
    'wc1': tf.Variable(tf.random_normal([5, 5, 1, 32])),
    # 5x5 conv, 32 inputs, 64 outputs
    # 上一层的32个输出变成这一层的输入，这层输出定为 64
    'wc2': tf.Variable(tf.random_normal([5, 5, 32, 64])),
    # fully connected, 7*7*64 inputs, 1024 outputs
    # 全连接层权重，因为28×28的图像经过了2次下采样，每次小一半，
    # 所以现在是 7×7，全连接层是普通的神经网络，所以数据要拉平
    # 成一维数组，即数量为 7×7×64，64为上一层的输出数量
    # 而这一层的输出定为 1024
    'wd1': tf.Variable(tf.random_normal([7*7*64, 1024])),
    # 1024 inputs, 10 outputs (class prediction)
    # 输出层，上一层的 1024 个输入
    # 输出类别为10个数字的概率，所以为 10
    'out': tf.Variable(tf.random_normal([1024, n_classes]))
}
# 存储所有 bias 的 dict
# 数量跟每层定好的输出数量一致
biases = {
    'bc1': tf.Variable(tf.random_normal([32])),
    'bc2': tf.Variable(tf.random_normal([64])),
    'bd1': tf.Variable(tf.random_normal([1024])),
    'out': tf.Variable(tf.random_normal([n_classes]))
}
# Construct model
# 调用上面的自定义函数来构建卷积神经网络
pred = conv_net(x, weights, biases, keep_prob)
# Define loss and optimizer
# 定义损失函数和训练函数
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
# Evaluate model
# 定义正确率评估方式
correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
# Initializing the variables
# 初始化变量，所有有变量的程序都要有这句
init = tf.global_variables_initializer()
# Launch the graph
# 启动图 （会话）
with tf.Session() as sess:
    sess.run(init)
    step = 1
    # Keep training until reach max iterations
    # 循环到达到最大训练次数
    while step * batch_size < training_iters:
        # 从训练集获取 batch_size 个数据
        batch_x, batch_y = mnist.train.next_batch(batch_size)
        # Run optimization op (backprop)
        # 训练
        sess.run(optimizer, feed_dict={x: batch_x, y: batch_y,
                                       keep_prob: dropout})
        # 每 10 步输出一次 log                               
        if step % display_step == 0:
            # Calculate batch loss and accuracy
            # 计算损失函数值和正确率
            loss, acc = sess.run([cost, accuracy], feed_dict={x: batch_x,
                                                              y: batch_y,
                                                              keep_prob: 1.})
            print("Iter " + str(step*batch_size) + ", Minibatch Loss= " + \
                  "{:.6f}".format(loss) + ", Training Accuracy= " + \
                  "{:.5f}".format(acc))
        step += 1
    print("Optimization Finished!")
    # Calculate accuracy for 256 mnist test images
    # 计算测试数据的正确率
    print("Testing Accuracy:", \
        sess.run(accuracy, feed_dict={x: mnist.test.images[:256],
                                      y: mnist.test.labels[:256],
                                      keep_prob: 1.}))