TensotFlow 應用實例：11-使用CNN預測手寫數字MNIST

本文是我在學習TensotFlow 的時候所記錄的筆記，共享出來希望能夠幫助一些需要的人。

什么是卷積神經網絡 CNN (深度學習)?
What is Convolutional Neural Networks (deep learning)?

卷積神經網絡 最常應用于 圖片識別
卷積是說神經網絡不在對每一個點的數據進行處理，而是對一個區域進行處理

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data


# 什么是卷積神經網絡 CNN (深度學習)?
# What is Convolutional Neural Networks (deep learning)?

# 卷積神經網絡 最常應用于 圖片識別
# 卷積 神經網絡
# 卷積是說神經網絡不在對每一個點的數據進行處理，
# 而是對一個區域進行處理
# Google 自己的 CNN 教程
# https://classroom.udacity.com/courses/ud730/lessons/6377263405/concepts/63796332430923


# number 1 to 10 image data
# 如果本地沒有相應的數據包，會先下載，然后解壓數據包
# MNIST_data 是下載數據要保存的位置
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

# 添加神經層
def add_layer(inputs, in_size, out_size, activation_function=None):
    # Weights define
    # 權重，盡量要是一個隨機變量
    # 隨機變量在生成初始變量的時候比全部為零效果要好的很多
    Weights = tf.Variable(tf.random_normal([in_size, out_size]))
    # biases define
    # 偏值項，是一個列表，不是矩陣，默認設置為0 + 0.1
    biases = tf.Variable(tf.zeros([1, out_size]) + 0.1)
    # W * x + b
    Wx_plus_b = tf.matmul(inputs, Weights) + biases
    # 如果activation_function是空的時候就表示是一個線性關系直接放回即可
    if activation_function is None:
        outputs = Wx_plus_b
    else:
        outputs = activation_function(Wx_plus_b)
    return outputs


# 計算精確度
# compute_accuracy 要使用
def compute_accuracy(v_xs, v_ys):
    global prediction
    y_pre = sess.run(prediction, feed_dict={xs: v_xs, keep_prob: 1})
    correct_prediction = tf.equal(tf.argmax(y_pre, 1), tf.argmax(v_ys, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    # result 是一個百分比，百分比越高證明越準確
    result = sess.run(accuracy, feed_dict={xs: v_xs, ys: v_ys, keep_prob: 1})
    return result


def weight_variable(shape):
    # normal 產生隨機變量
    # stddev: A 0-D Tensor or Python value of type `dtype`. The standard deviation
    # of the truncated normal distribution.
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)


def bias_variable(shape):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)


# x input W is weight
def conv2d(x, W):
    # strides [1, x_movement, y_movement, 1]
    # 前后都要為1
    # VALID SAME padding方式
    # VALID 較小， SAME 和原圖一樣
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')


def max_pool_2x2(x):
    result = tf.nn.max_pool(x,
                            ksize=[1, 2, 2, 1],
                            strides=[1, 2, 2, 1],
                            padding='SAME')
    return result


# 定義 placeholder
xs = tf.placeholder(tf.float32, [None, 784])/255.
ys = tf.placeholder(tf.float32, [None, 10])
keep_prob = tf.placeholder(tf.float32)
# 將輸入的xs轉換為圖片的形式
# -1 不管維度
# 28*28 像素點
# 1 channel 是黑白
x_image = tf.reshape(xs, [-1, 28, 28, 1])
# print(x_image.shape) # [n_samples, 28, 28, 1]


# conv1 layer
# 5 * 5 patch ，長*寬
# in size is 1， image的厚度，輸入的厚度
# out is 32， 輸出的深度，厚度
W_conv1 = weight_variable([5, 5, 1, 32])
# 32個輸出，所有b為32
b_conv1 = bias_variable([32])
# conv2d output size 28x28X32
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
# max_pool_2x2 output 14x14x32
h_pool1 = max_pool_2x2(h_conv1)

# conv2 layer
# out is 64， 輸出的深度，厚度
W_conv2 = weight_variable([5, 5, 32, 64])
# 32個輸出，所有b為64
b_conv2 = bias_variable([64])
# conv2d output size 14x14x64
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
# max_pool_2x2 output 7x7x64
h_pool2 = max_pool_2x2(h_conv2)

# func1 layer
W_fc1 = weight_variable([7*7*64, 1024])
b_fc1 = bias_variable([1024])
# [n_samples, 7, 7, 64] >> [n_samples, 7*7*64]
h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
# drop out
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)


# func2 layer
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])
prediction = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)

# prediction = add_layer(xs, 784, 10,  activation_function=tf.nn.softmax)

# cross_entropy 分類的時候經常使用softmax + cross_entropy來計算的
cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys * tf.log(prediction),
                                              reduction_indices=[1]))
# train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
# AdamOptimizer 需要的學習速率應該更小
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)


sess = tf.Session()

# important step
# tf.initialize_all_variables() no long valid from
# "2017-03-02", "Use `tf.global_variables_initializer` instead."
init = tf.global_variables_initializer()
sess.run(init)


for i in range(1000):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={xs: batch_xs, ys: batch_ys, keep_prob: 0.5})
    if i % 50 == 0:
        print(compute_accuracy(mnist.test.images, mnist.test.labels))

本文代碼GitHub地址 tensorflow_learning_notes