今天分享同样数据集的CNN处理方式，同时加上tensorboard，可以看到清晰的结构图，迭代1000次acc收敛到0.992 先放代码，注释比较详细，变量名字看单词就能知道啥意思
import tensorflow as tf
import tensorflow.examples.tutorials.mnist.input_data as input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
batch_size = 100
n_batch = mnist.train.num_examples // batch_size
def weight_variable(shape, name):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial, name=name)
def bias_variable(shape, name):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial, name=name)
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding="SAME") # 1, 3位是1/ 2， 4位是步长
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME") # 2， 3是步长
with tf.name_scope('input'):
x = tf.placeholder(tf.float32, [None, 784], name='x_input')
y = tf.placeholder(tf.float32, [None, 10], name='y_input')
with tf.name_scope('x_image'):
# 转变x格式为4D向量[batch, in_height, in_width, in_channels] 通道为1表示黑白
x_image = tf.reshape(x, [-1, 28, 28, 1], name='x_image')
with tf.name_scope('Converlution_1'):
# 初始化第一个卷积层的权重和偏置
with tf.name_scope('Weight_Converlution_1'):
Weight_Converlution_1 = weight_variable([5, 5, 1, 32],
name='Weight_Converlution_1') # 5 * 5的卷积窗口，32个卷积核从1个平面抽取特征
# 生成32个特征图
with tf.name_scope('Biase_Converlution_1'):
Biase_Converlution_1 = bias_variable([32], name='Biase_Converlution_1') # 每个卷积核一个偏置值
# 把x_image和权值向量进行卷积，再加上偏置值，应用于relu激活函数
with tf.name_scope('Converlution2d_1'):
Converlution2d_1 = conv2d(x_image, Weight_Converlution_1) + Biase_Converlution_1
with tf.name_scope('ReLu_1'):
ReLu_Converlution_l = tf.nn.relu(Converlution2d_1)
with tf.name_scope('Pool_1'):
Pooling_1 = max_pool_2x2(ReLu_Converlution_l) # max-pooling
with tf.name_scope('Converlution_2'):
# 初始化第二个卷积层的权重和偏置
with tf.name_scope('Weight_Converlution_2'):
Weight_Converlution_2 = weight_variable([5, 5, 32, 64],
name='Weight_Converlution_2') # 5 * 5的卷积窗口，64个卷积核从32个平面抽取特征
# 生成64个特征图
with tf.name_scope('Biase_Converlution_2'):
Biase_Converlution_2 = bias_variable([64], name='Biase_Converlution_2')
with tf.name_scope('Cov2d_2'):
Converlution2d_2 = conv2d(Pooling_1, Weight_Converlution_2) + Biase_Converlution_2
with tf.name_scope('ReLu_2'):
ReLu_Converlution_2 = tf.nn.relu(Converlution2d_2)
with tf.name_scope('Pool_2'):
Pooling_2 = max_pool_2x2(ReLu_Converlution_2)
# 初始化第一个全连接层
with tf.name_scope('Fully_Connected_L1'):
with tf.name_scope('Weight_Fully_Connected_L1'):
Weight_Fully_Connected_L1 = weight_variable([7 * 7 * 64, 1024],
name='Weight_Fully_Connected_L1') # 上一层有7*7*64个神经元，全连接层有1024个神经元
with tf.name_scope('Biase_Fully_Connected_L1'):
Biase_Fully_Connected_L1 = bias_variable([1024], name='Biase_Fully_Connected_L1')
# 把池化层2的输出扁平化为1维
with tf.name_scope('Pooling_2_to_Flat'):
Pooling_2_to_Flat = tf.reshape(Pooling_2, [-1, 7 * 7 * 64], name='Pooling_2_to_Flat') # -1表示任意值
with tf.name_scope('Wx_Plus_B1'):
Wx_Plus_B1 = tf.matmul(Pooling_2_to_Flat, Weight_Fully_Connected_L1) + Biase_Fully_Connected_L1
with tf.name_scope('ReLu_Fully_Connected_L1'):
ReLu_Fully_Connected_L1 = tf.nn.relu(Wx_Plus_B1)
# Keep——Prob
with tf.name_scope('keep_prob'):
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
with tf.name_scope('Fully_Connected_L1_Drop'):
Fully_Connected_L1_Drop = tf.nn.dropout(ReLu_Fully_Connected_L1, keep_prob, name='Fully_Connected_L1_Drop')
# 初始化第二个全连接层
with tf.name_scope('Fully_Connected_L2'):
with tf.name_scope('Weight_Fully_Connected_L2'):
Weight_Fully_Connected_L2 = weight_variable([1024, 10], name='Weight_Fully_Connected_L2')
with tf.name_scope('Biase_Fully_Connected_L2'):
Biase_Fully_Connected_L2 = bias_variable([10], name='Biase_Fully_Connected_L1')
with tf.name_scope('Wx_Plus_B2'):
Wx_Plus_B2 = tf.matmul(Fully_Connected_L1_Drop, Weight_Fully_Connected_L2) + Biase_Fully_Connected_L2
with tf.name_scope('SoftMax'):
prediction = tf.nn.softmax(Wx_Plus_B2)
# 交叉熵函数
with tf.name_scope('cross_entropy'):
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y,
logits=prediction), name='cross_entropy')
tf.summary.scalar('cross_entropy', cross_entropy) # 显示标量信息 tf.summary.scalar(tags, values, collections=None, name=None)
# AdamOptimizer优化器
with tf.name_scope('train'):
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
with tf.name_scope('accuracy'):
with tf.name_scope('correct_prediction'):
correct_prediction = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1))
with tf.name_scope('accuracy'):
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
tf.summary.scalar('accuracy', accuracy)
# 合并所有summary
merged = tf.summary.merge_all()
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
train_writer = tf.summary.FileWriter('logs/train', sess.graph)
test_writer = tf.summary.FileWriter('logs/test', sess.graph)
for i in range(101):
batch_xs, batch_ys = mnist.train.next_batch(batch_size)
sess.run(train_step, feed_dict={x: batch_xs,
y: batch_ys,
keep_prob: 0.5})
# 记录训练集计算的参数
summary = sess.run(merged, feed_dict={x: batch_xs,
y: batch_ys,
keep_prob: 1})
train_writer.add_summary(summary, i)
# 记录训练集计算的参数
batch_xs, batch_ys = mnist.test.next_batch(batch_size)
summary = sess.run(merged, feed_dict={x: batch_xs,
y: batch_ys,
keep_prob: 1.0})
test_writer.add_summary(summary, i)
test_acc = sess.run(accuracy, feed_dict={x: mnist.test.images,
y: mnist.test.labels,
keep_prob: 1.0})
train_acc = sess.run(accuracy, feed_dict={x: mnist.train.images[:10000],
y: mnist.train.labels[:10000],
keep_prob: 1.0})
print("Iter " + str(i) +
", Testing Accuracy " + str(test_acc) +
", Training Accuracy " + str(train_acc))