5.1 MNIST数据集
MNIST数据集：提供6w张28*28像素点的0~9手写数字图片和标签，用于训练。
? ? ? ? ? ? ? ? ? ? ? ? ?提供1w张28*28像素点的0~9手写数字图片和标签，用于测试。
? ? ? ? ? ? ? ? ? ? ? ? ?黑底白字，黑底用0表示，白字用0~1之间浮点数表示，越接近1 颜色越白。
每张图片的784个像素点（28*28）组成长度为784的一维数组，作为输入特征。
eg: [0.,? 0. ,? ? ? .......? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 0.234,? 0.,??0. , 0.]?
图片的标签以一维数组形式给粗，每个元素表示对应分类出现的概率。
eg:[0., 0.,?0., 0.,?0., 0., 1.,0., 0.,?0., ]? ? ? ? ? ? ?表示 图片为数字6的概率为1，为其他数字的概率为0
导入数据集
form tensorflow.examples.tutorials.mnist import input_data
#在data目录下读入数据集合，如果没有，则自动导入数据集
mnist = input_data.read_data_sets('./data/', one_hot=True)
数据集常用操作
# 显示 训练集 校验集 测试集 大小
print "train data size :", mnist.train.num_examples
print "validation data size :", mnist.validation.num_examples
print "test data size :", mnist.test.num_examples
# 显示结果
train data size : 55000
validation data size : 5000
test data size : 10000
# 返回标签和数据
mnist.train.labels[0] # 返回第0张图的标签 是数字几的概率
mnist.train.images[0] # 返回784个像素点
#取一小撮数据，准备喂入神经网络训练
BATCH_SIZE = 200 # 定义一小撮是多少
xs, ys = mnist.train.next_batch(BATCH_SIZE)
print "xs shape:", xs.shape
print "ys shape:", ys.shape
# 显示结果
xs shape: (200， 784)
ys shape: (200,10)
常用函数
| tf.get_collection("") | 从集合中取出全部变量， 生成一个列表 | | tf.add_n([ ]) | 把列表内对应元素相加 | | tf.cast(x, dtype) | 把x转为dtype | | tf.argmax(x, axis) | 返回最大值所作的索引号，如：tf.argmax([1, 0, 0],1) 返回0 | | os.path.join("home", "name") | 将字符串按路径的命名规则拼接，返回home/name | | 字符串.split() | 按指定拆分符对字符串切片， 返回分割后的列表。 如：’./model/mnist_model-1000'.split('/')[-1].split('-')[-1]? ? (要是看不懂，自己用python试一下就明白了，[-1]是split()返回列表中最后一项的意思) | | with tf.Graph().as_default() as g: | 其定义的节点在计算图g中,一般用这种方法复现已经定义好的神经网络 |
保存模型
# 保存模型
saver = tf.train.Saver()
tf.Session() as sess:
for i in range(STEPS)：
if i%轮数 == 0：
# 拼接成 ./MODEL_SAVE_PATN/MODEL_NAME-global_step MODEL_SAVE_PATN 为字符串常量， MODEL_NAME同
# global体现在哪里？
saver.save(sess, os.path.jion(MODEL_SAVE_PATN, MODEL_NAME), global_step = global_step)
# 加载模型
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(存储路径）
if ckpt and ckpt.modal_checkpoint_path :
saver.restore(sess, ckpt.modal_checkpoint_path)
# 实例化可还原滑动平均值的saver
# ema 相关内容可以复习第4节的内容
ema = tf.train.ExponentialMovingAverage(滑动平均基数）
ema_restore = ema.variables_to_restore()
saver = tf.train.Saver(ema_restore)
# 准确率计算方法
correct_prediction = tf.equal(tf.argmax(y, 1),tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
5.2 模块化搭建神经网络八股
这部分是上一节内容的复习
# forward.py 伪代码
def forward(x, regularizer):
w =
b =
y =
return y
def get_weight(shape, regularizer):
def get_bias(shape):
----------------------------------------------------------------------------------
# backward.py 伪代码
def backward(mnist):
x = tf.placeholder
y_ =
y=
global_step =
loss =
< 正则化， 指数衰减学习率，滑动平均>
train_step =
实例化 saver
with tf.Session() as sess:
初始化op
for i in range(STEPS):
sess.run(train_step, feed_dict = {x: ,y_:})
if i%轮数 ==0：
print
saver.save()
----------------------------------------------------------------------------------
# 损失函数loss 含正则化regularization
# backwoard.py 中加入 正则化
# 交叉熵和正则化一起使用
ce = tf.nn.sparse_soft_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1))
cem = tf.reduce_mean(ce)
loss = cem + tf.add_n(tf.get_collection("losses"))
# forward.py 中加入
if regularizer != None:tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))
----------------------------------------------------------------------------------
# 学习率 learning_rate
# backward.py中加入
learning_rate = tf.train.exponential_decay(
LERANING_RATE_BASE,
global_step,
LEARNING_RATE_STEP,
LEARNING_RATE_DECAY,
staircase=True) # staircase 楼梯
----------------------------------------------------------------------------------
# 滑动平均 ema
# backward.py 中加入
ema = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECY,global_step)
ema_op = ema.apple(tf.trainable_variables())
with tf.control_dependencies([train_step, ema_op]):
train_op = tf.no_op(name='train')
----------------------------------------------------------------------------------
# test.py 伪代码
def test(mnist):
with tf.Graph().as_default as g:
定义x, y_,y
实例化课还原滑动平均值的saver
计算争取率
while True:
with tf.Session() as sess:
加载ckpt模块 ckpt = tf.train.get_checkpoint_state(存储路径）
如果已有ckpt模型则恢复 if ckpt and ckpt.moddel_checkpoint_path:
恢复会话 saver.restore(sess, ckpt.moddel_checkpoint_path)
恢复轮数 global_step = ckpt.moddel_checkpoint_pathsplit('/')[-1].split('-')[-1]
计算准确率 accuracy_score = sess.run(accuary, feed_dict={x:mnist.test.images, y_::mnist.test.labels})
打印提示 print global_step,accuracy_score
如果没有模型 else:
给出提示 print 'No checkpiont file found'
return
def main():
mnist = input_data.read_data_sets('./data/', one_hot = True)
test(mnist)
if __name__ == '__main__':
main()
5.3 手写数字识别准确率输出
直接上课程中的代码，代码参考（https://github.com/upseaup/Tensorflow/tree/master/Tensorflow%205%20fc1）
import tensorflow as tf
INPUT_NODE = 784
OUTPUT_NODE = 10
LAYER1_NODE = 500
def get_weight(shape, regularizer):
w = tf.Variable(tf.truncated_normal(shape,stddev=0.1))
# 判断是否使用正则化
if regularizer != None :tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))
return w
def get_bias(shape):
b = tf.Variable(tf.zeros(shape))
return b
def forward(x, regularizer):
w1 = get_weight([INPUT_NODE, LAYER1_NODE], regularizer)
b1 = get_bias([LAYER1_NODE])
# 输出过激活函数
y1 = tf.nn.relu(tf.matmul(x, w1) + b1)
w2 = get_weight ([LAYER1_NODE, OUTPUT_NODE],regularizer)
b2 = get_bias([OUTPUT_NODE])
# 输出不过激活函数，并且不需要softmax
y = tf.matmul(y1,w2)+b2
return y
###not use softmax how to mut_classity???? p97
前向传播代码, 这段代码都比较熟悉了，需要注意的是，代码中forward函数y1过激活函数，y却没有过激活函数，而且数字识别是一个多分类问题，代码中竟然没有使用softmax函数。这是因为我们在求损失函数时，使用的是sparse_softmax_cross_entropy_with_logits（）函数。根据tf官网（https://tensorflow.google.cn/api_docs/python/tf/nn/sparse_softmax_cross_entropy_with_logits?hl=en）给出的解释：
Warning:?This op expects unscaled logits, since it performs a?softmax?on?logits?internally for efficiency. Do not call this op with the output of?softmax, as it will produce incorrect results.
反向传播代码
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_forward
import os
BATCH_SIZE = 200
LEARNING_RATE_BASE = 0.1
LEARNING_RATE_DECAY = 0.99
REGULARIZER = 0.0001
STEPS = 50000
MOVING_AVERAGE_DECAY = 0.99
MODEL_SAVE_PATH = ".model/"
MODEL_NAME = './mnist_model'
def backward(mnist):
x = tf.placeholder(tf.float32, [None, mnist_forward.INPUT_NODE])
y_ = tf.placeholder(tf.float32, [None,mnist_forward.OUTPUT_NODE])
y = mnist_forward.forward(x, REGULARIZER)
global_step = tf.Variable(0, trainable=False)
# 以下两句可以使用 cem = tf.losses.sparse_softmax_cross_entropy(logits=y,
# onehot_labels=y_)
# 替代，根据tf官网， tf.losses.sparse_softmax_cross_entropy是用
# tf.nn.sparse_softmax_cross_entropy_with_logits 实现的
ce = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_,1))
cem= tf.reduce_mean(ce)
loss = cem + tf.add_n(tf.get_collection('losses'))
learning_rate = tf.train.exponential_decay(
LEARNING_RATE_BASE,
global_step,
mnist.train.num_examples/BATCH_SIZE,
LEARNING_RATE_DECAY,
staircase = True)
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step = global_step)
# 这里将前向传播网络的参数w b 进行滑动平均
ema = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step)
ema_op = ema.apply(tf.trainable_variables())
# 将两个操作合为一个操作， 具体可参考《TensorFlow实战Google学习框架》
# 下面两行 和 train_op = tf.group(train_step, ema_op)是等价的
with tf.control_dependencies([train_step, ema_op]):
train_op = tf.no_op(name = 'train')
saver = tf.train.Saver()
with tf.Session() as sess:
init_op = tf.global_variables_initializer()
sess.run(init_op)
for i in range(STEPS):
xs, ys = mnist.train.next_batch(BATCH_SIZE)
# sess.run arg ???
_, loss_value, step = sess.run([train_op, loss, global_step,], feed_dict = {x:xs, y_: ys})
if i % 1000 == 0 :
print 'After %d train steps, loss on train batch is %g. ' %(step, loss_value)
# lobal_step = global_step 将global_step 写入文件名中
saver.save(sess, os.path.join(MODEL_SAVE_PATH, MODEL_NAME), global_step = global_step)
def main():
mnist = input_data.read_data_sets('./data/', one_hot = True)
backward(mnist)
if __name__ == "__main__":
main()
根据《TensorFlow实战Google学习框架》p97? inference 函数 中关于滑动平均的例子说明，本文中应该是没有成功调用滑动平均。欢迎各位大佬讨论本例程是否成功配置滑动平均。
测试程序：
#coding:utf-8
import time
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_forward
import mnist_backward
TEST_INTERVAL_SECS = 5
def test(mnist):
with tf.Graph().as_default() as g:
x = tf.placeholder(tf.float32, [None, mnist_forward.INPUT_NODE])
y_ = tf.placeholder(tf.float32, [None, mnist_forward.OUTPUT_NODE])
y = mnist_forward.forward(x, None)
ema = tf.train.ExponentialMovingAverage(mnist_backward.MOVING_AVERAGE_DECAY)
# variables_to_restore()
ema_restore = ema.variables_to_restore()
# tf.train.Saver类可以参考 《TensorFlow实战Google学习框架》p111部分内容
# 以及tf官网部分内容
saver = tf.train.Saver(ema_restore)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
while True :
with tf.Session() as sess:
ckpt =tf.train.get_checkpoint_state(mnist_backward.MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess,ckpt.model_checkpoint_path)
# 通过存储文件名恢复 global_step
global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
accuracy_score = sess.run(accuracy, feed_dict={x:mnist.test.images, y_: mnist.test.labels})
print 'after %s training step(s), test accuracy = %g' %(global_step, accuracy_score)
else:
print 'No checkpiont flie found'
time.sleep(TEST_INTERVAL_SECS)
def main():
mnist = input_data.read_data_sets('./data', one_hot =True)
test(mnist)
if __name__ == '__main__':
main()
? tf.train.Saver类可以参考 《TensorFlow实战Google学习框架》p111部分内容?,以及tf官网部分内容
(https://tensorflow.google.cn/versions/r1.15/api_docs/python/tf/train/Saver?hl=en)
|