在b站上看视频学习tensorflow基本知识，同时记录一下~
由于课程第一节讲的是环境搭建，那部分自己完成了，没有听课，所以就从（二）开始吧
- 视频内容：06-1；06-2
- 主要内容：卷积神经网络CNN的讲解，以及用CNN解决MNIST分类问题。
- 代码链接参考：github，由于所有的代码都是在Jupyter Notebook中进行编写以及运行，所有代码编写过程中有很多次运行，查看输出结果。
- 代码格式是.ipynb格式，查看代码方法：已经安装好了Anaconda等环境，打开终端，输入
Jupyter Notebook，就可以按照路径打开代码了。
6-1：讲解CNN基本知识
可以参考：
博客一:CNN之自我理解
博客二：零基础入门深度学习
6-2：tensorflow中如何使用卷积
- 首先知道用到tf的函数：
tf.nn.conv2d以及tf.nn.max_pool - 卷积层自己定义一个卷积函数：
def conv2d(x,W):
return tf.nn.conv2d(x,W,strides=[1,1,1,1],padding='SAME')
#x是输入tensor [batch， in_height,in_width,in_cahnnels] 这里batch=100
#W:filter/kernel 也是一个tensor [filter_height , filter_width , in_channels , out_channels]
#stride: stride[0]=stride[3]=1 stride[1]：代表x方向的步长 stride[2]：代表y方向的步长
#padding ： SAME（补0）和VALID（不补0）
- 自己定义一个池化层函数：
#定义池化层
def max_pool_2x2(x):
return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')
#池化窗口2*2 横向纵向每次移动2个格子
- 非线性激活函数：
tf.nn.relu - 描述卷积层的时候，应当需要，先定义权重和偏置，再描述卷积操作。
完整代码如下：
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
batch_size = 100
n_batch = mnist.train.num_examples // batch_size
#初始化权值
def weight_variable(shape):
return tf.Variable(tf.truncated_normal(shape,stddev=0.1))
#初始偏置
def bias_vairable(shape):
return tf.Variable(tf.constant(0.1, shape=shape))
#定义卷及操作，调用了一个库
def conv2d(x,W):
return tf.nn.conv2d(x,W,strides=[1,1,1,1],padding='SAME')#至于函数内部是如何卷积的我们并不需要知道，只需要知道格式就可以。
#x是输入tensor [batch， in_height,in_width,in_cahnnels] 这里batch=100
#W:filter/kernel 也是一个tensor [filter_height , filter_width , in_channels , out_channels]
#stride: stride[0]=stride[3]=1 stride[1]：代表x方向的步长 stride[2]：代表y方向的步长
#padding ： SAME（补0）和VALID（不补0）
#定义池化层
def max_pool_2x2(x):
return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')
#池化窗口2*2 横向纵向每次移动2个格子
x = tf.placeholder(tf.float32,[None,784]) #None：代表行
y = tf.placeholder(tf.float32,[None,10])
keep_prob = tf.placeholder(tf.float32)
x_image = tf.reshape(x,[-1,28,28,1]) #转换形状 -1代表批次大小 28长 28宽 把图片复原为原来的样子 1灰白图片
W_conv1 = weight_variable([5,5,1,32]) # 5*5的采样窗口，1：输入通道数 32：输出的通道数 32个卷积核从1个平面抽取特征
b_conv1 = bias_vairable([32]) #每个卷积核一个偏置值
# 28*28*1 的图片卷积之后变为28*28*32
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
# 池化之后变为 14*14*32
h_pool1 = max_pool_2x2(h_conv1)
# 第二次卷积之后变为 14*14*64
W_conv2 = weight_variable([5,5,32,64]) #刚才的输出是32个通道，所以这里是32
b_conv2 = bias_vairable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1,W_conv2) + b_conv2)
# 第二次池化之后变为 7*7*64
h_pool2 = max_pool_2x2(h_conv2)
#28*28的图片 第一次卷积后还是28*28 第一次池化后变为14*14
#第二次卷积后为14*14， 第二次池化后变为7*7
#经过上面的操作后得到64张7*7的平面
# 第一个全连接层
W_fc1 = weight_variable([7*7*64,1024])#上一层神经元7*7*64个 全连接层1024个神经元
b_fc1 = bias_vairable([1024])
# 7*7*64的图像变成1维向量
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)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
# 第二个全连接层
W_fc2 = weight_variable([1024,10])
b_fc2 = bias_vairable([10])
logits = tf.matmul(h_fc1_drop,W_fc2) + b_fc2
prediction = tf.nn.sigmoid(logits)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=logits))
train_step = tf.train.AdamOptimizer(0.001).minimize(loss)
prediction_2 = tf.nn.softmax(prediction)
correct_prediction = (tf.equal(tf.argmax(prediction_2,1), tf.argmax(y,1)))
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for epoch in range(21):
for batch in range(n_batch):
batch_xs, batch_ys = mnist.train.next_batch(batch_size)
sess.run(train_step, feed_dict={x:batch_xs,y:batch_ys,keep_prob:0.7})
acc = sess.run(accuracy, feed_dict={x:mnist.test.images, y:mnist.test.labels, keep_prob:1.0})
print("Iter: " + str(epoch) + ", acc: " + str(acc))
加上tensorboard的代码如下：
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
batch_size = 128
n_batch = mnist.train.num_examples // batch_size
max_step = 1000
keep_ = 0.8
log_dir = "Logs/log-6.1"
# 生成权重
def weight_variable(shape):
return tf.Variable(tf.truncated_normal(shape,stddev=0.1),name='W')
# 生成偏差
def bias_vairable(shape):
return tf.Variable(tf.constant(0.1, shape=shape),name='b')
# 记录变量
def variable_summaries(var):
with tf.name_scope('summaries'):
mean = tf.reduce_mean(var)
tf.summary.scalar('mean', mean)
with tf.name_scope('stddev'):
stddev = tf.sqrt(tf.reduce_mean(tf.square(var-mean)))
tf.summary.scalar('stddev', stddev)
tf.summary.scalar('max', tf.reduce_max(var))
tf.summary.scalar('min', tf.reduce_min(var))
tf.summary.histogram('histogram', var)
def conv2d(x,W):
return tf.nn.conv2d(x,W,strides=[1,1,1,1],padding='SAME',name='conv2d')
def conv_layer(input_tensor, weight_shape, layer_name, act=tf.nn.relu):
with tf.name_scope(layer_name):
with tf.name_scope('weights'):
weights = weight_variable(weight_shape)
variable_summaries(weights)
with tf.name_scope('biases'):
biases = bias_vairable([weight_shape[-1]])
variable_summaries(biases)
with tf.name_scope('conv_comput'):
preactivate = conv2d(input_tensor,weights) + biases
with tf.name_scope('activate'):
activations = act(preactivate)
return activations
def linear_layer(input_tensor, input_dim, output_dim, layer_name, act=tf.nn.relu):
with tf.name_scope(layer_name):
with tf.name_scope('weights'):
weights = weight_variable([input_dim, output_dim])
variable_summaries(weights)
with tf.name_scope('biases'):
biases = bias_vairable([output_dim])
variable_summaries(biases)
with tf.name_scope('linear_comput'):
preactivate = tf.matmul(input_tensor,weights) + biases
with tf.name_scope('activate'):
activations = act(preactivate)
return activations
def max_pool_2x2(x):
return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',name='Max_pool')
with tf.name_scope('Input'):
x = tf.placeholder(tf.float32,[None,784],name='input_x')
with tf.name_scope('Input_reshape'):
x_image = tf.reshape(x,[-1,28,28,1],name='x-image')
tf.summary.image('input',x_image,10)
y = tf.placeholder(tf.float32,[None,10],name='input_y')
keep_prob = tf.placeholder(tf.float32,name='keep_prob')
# 第一次卷积 28*28*1->28*28*32
conv_layer1 = conv_layer(x_image,[5,5,1,32],'conv_layer1')
# 池化之后变为 14*14*32
with tf.name_scope('Max_pool1'):
h_pool1 = max_pool_2x2(conv_layer1)
# 第二次卷积 14*14*32->14*14*64
conv_layer2 = conv_layer(h_pool1,[5,5,32,64],'conv_layer2')
# 第二次池化之后变为 7*7*64
with tf.name_scope('Max_pool2'):
h_pool2 = max_pool_2x2(conv_layer2)
with tf.name_scope('Flatten'):
flatten_ = tf.reshape(h_pool2,[-1,7*7*64])
# 第一个全连接层 7*7*64 - 1024
fc1 = linear_layer(flatten_, 7*7*64, 1024, 'FC1')
with tf.name_scope('Dropput'):
fc1_drop = tf.nn.dropout(fc1, keep_prob)
# 第二个全连接层 1024 - 10
logits = linear_layer(fc1_drop, 1024, 10, 'FC2',act=tf.nn.sigmoid)
with tf.name_scope('loss'):
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=logits))
tf.summary.scalar('loss',loss)
with tf.name_scope('train'):
train_step = tf.train.AdamOptimizer(0.001).minimize(loss)
with tf.name_scope('accuracy'):
prediction = tf.nn.softmax(logits)
correct_prediction = tf.equal(tf.argmax(prediction,1), tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
tf.summary.scalar('accuracy', accuracy)
merged = tf.summary.merge_all()
def get_dict(train):
if train:
xs, ys = mnist.train.next_batch(batch_size)
k = keep_
else:
xs, ys = mnist.test.images, mnist.test.labels
k = 1.0
return {x:xs, y:ys, keep_prob: k}
with tf.Session() as sess:
train_writer = tf.summary.FileWriter(log_dir + '/train', sess.graph)
test_writer = tf.summary.FileWriter(log_dir + '/test')
sess.run(tf.global_variables_initializer())
for i in range(max_step):
if i%10 == 0:
summary,acc = sess.run([merged,accuracy], feed_dict=get_dict(False))
test_writer.add_summary(summary, i)
print("Step: " + str(i) + ", acc: " + str(acc))
else:
summary,_ = sess.run([merged,train_step], feed_dict=get_dict(True))
train_writer.add_summary(summary,i)
train_writer.close()
test_writer.close()
1276
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
