TensorFlow_MNIST

# Author: Lee
import tensorflow as tf
import numpy as np

# 下载并载入mnist手写数据库
from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets('mnist', one_hot=True)

#None表示张量Tensor可以是任何维度, /255.是对灰色图像做归一化
input_x = tf.placeholder(tf.float32, [None, 28 * 28]) / 255.
output_y = tf.placeholder(tf.float32, [None, 10])
#对输入数据进行改变形状28 * 28 * 1, -1是维度设置为auto
input_x_image = tf.reshape(input_x, [-1, 28, 28, 1])

#从测试Test数据集中选取3000个手写数字的图片和对应的标签
test_x = mnist.test.images[:3000]
test_y = mnist.test.labels[:3000]

#构建神经网络
#第一层卷积 filters,kernals size,strides
conv1 = tf.layers.conv2d(
    inputs = input_x_image, # shape = [28, 28, 1]
    filters = 32,           # 32个过滤器,输出深度为32
    kernel_size = [5, 5],   # 过滤器在二维的大小为5 * 5,相当于过滤器大小
    strides = 1,            # 步长为1
    padding = 'SAME',       # padding补零方案,same表示输出大小不变(same和valid的算法需要参考官方文档),需要在外围补零两圈
    activation = tf.nn.relu
    )
#经过第一层卷积之后的输出数据shape为28 * 28 * 32

#第一层池化(亚采样)pooling
pool1 = tf.layers.max_pooling2d(
    inputs = conv1,
    pool_size = [2, 2],     # 过滤器在二维的大小,类比kernel_size
    strides = 2,            # 步长2
    )
#经过第一层池化之后的输出数据shape为14 * 14 * 32

#第二层卷积 filters,kernals size,strides
conv2 = tf.layers.conv2d(
    inputs = pool1,         # shape = [14, 14, 32]
    filters = 64,           # 64个过滤器,输出深度为64
    kernel_size = [5, 5],   # 过滤器在二维的大小为5 * 5,相当于过滤器大小
    strides = 1,            # 步长为1
    padding = 'SAME',       # padding补零方案,same表示输出大小不变,需要在外围补零两圈
    activation = tf.nn.relu
    )
#经过第二层卷积之后的输出数据shape为14 * 14 * 64

#第二层池化(亚采样)pooling
pool2 = tf.layers.max_pooling2d(
    inputs = conv2,
    pool_size = [2, 2],     # 过滤器在二维的大小,类比kernel_size
    strides = 2,            # 步长2
    )
#经过第一层池化之后的输出数据shape为7 * 7 * 64

#平坦化(flat)
flat = tf.reshape(pool2, [-1, 1, 1, 1024])

#1024个神经元的全连接层
dense = tf.layers.dense(
    inputs = flat,
    units = 1024,
    activation = tf.nn.relu
    )

# Dropout 丢弃率为50%, Dropout的rate在[0,1]
dropout = tf.layers.dropout(
    inputs = dense,
    rate = 0.5
    )

# 10个神经元的全连接层, 这里不用激活函数做非线性化
logits = tf.layers.dense(inputs = dropout, units = 10)
#输出形状1 * 1 * 10

#计算误差<计算Cross_entropy(交叉熵),再用Softmax进行计算百分比>
loss = tf.losses.softmax_cross_entropy(onehot_labels = output_y, logits = logits)

#使用Adam优化器,Adam为默认优化器,learning_rate = 0.001
train_op = tf.train.AdamOptimizer(learning_rate = 0.001).minimize(loss)

#预测值和匹配度
#返回(accuracy, update_op),能够创建两个局部变量
accuracy = tf.metrics.accuracy(
    labels = tf.argmax(output_y, axis = 1),
    predictions = tf.argmax(logits, axis = 1))[1]


#创建会话Session
sess = tf.Session()

#初始化变量:全局变量和局部变量
init = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
sess.run(init)

for i in range(30000):
    batch = mnist.train.next_batch(50)
    train_loss, train_op_ = sess.run([loss, train_op], {input_x: batch[0], output_y: batch[1]})
    if i % 100 == 0:
        test_accuracy = sess.run(accuracy, {input_x: test_x, output_y: test_y})
        print("accuracy:", test_accuracy,'loss:', train_loss)


#测试, 预测值与真实值对比
test_output = sess.run(logits, {input_x: test_x[:20]})
inferenced_y = np.argmax(test_output, 1)
#推测的数字
print('inferenced_data:',inferenced_y)
#实际数字
print('test_output_data:', np.argmax(test_y[:20],1))

按照如下CNN过程写的,但是RUN的时候出现了一些问题

为什么不兼容?????

由于时间问题,暂且先将代码贴上去, 放在这周末解决!主要就conv2d pool2d内的参数, reshape内的参数注重研究一下!