快捷導(dǎo)航

在Tensorflow中實(shí)現(xiàn)梯度下降法更新參數(shù)值

更新時(shí)間：2020年01月23日 18:07:20 作者：勿在浮沙筑高臺(tái)LS

今天小編就為大家分享一篇在Tensorflow中實(shí)現(xiàn)梯度下降法更新參數(shù)值，具有很好的參考價(jià)值，希望對(duì)大家有所幫助。一起跟隨小編過(guò)來(lái)看看吧

我就廢話不多說(shuō)了，直接上代碼吧！

tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

TensorFlow經(jīng)過(guò)使用梯度下降法對(duì)損失函數(shù)中的變量進(jìn)行修改值，默認(rèn)修改tf.Variable(tf.zeros([784,10]))

為Variable的參數(shù)。

train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy,var_list=[w,b])

也可以使用var_list參數(shù)來(lái)定義更新那些參數(shù)的值

#導(dǎo)入Minst數(shù)據(jù)集
import input_data
mnist = input_data.read_data_sets("data",one_hot=True)
 
#導(dǎo)入tensorflow庫(kù)
import tensorflow as tf
 
#輸入變量，把28*28的圖片變成一維數(shù)組（丟失結(jié)構(gòu)信息）
x = tf.placeholder("float",[None,784])
 
#權(quán)重矩陣，把28*28=784的一維輸入，變成0-9這10個(gè)數(shù)字的輸出
w = tf.Variable(tf.zeros([784,10]))
#偏置
b = tf.Variable(tf.zeros([10]))
 
#核心運(yùn)算，其實(shí)就是softmax（x*w+b）
y = tf.nn.softmax(tf.matmul(x,w) + b)
 
#這個(gè)是訓(xùn)練集的正確結(jié)果
y_ = tf.placeholder("float",[None,10])
 
#交叉熵，作為損失函數(shù)
cross_entropy = -tf.reduce_sum(y_ * tf.log(y))
 
#梯度下降算法，最小化交叉熵
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)
 
#初始化，在run之前必須進(jìn)行的
init = tf.initialize_all_variables()
#創(chuàng)建session以便運(yùn)算
sess = tf.Session()
sess.run(init)
 
#迭代1000次
for i in range(1000):
 #獲取訓(xùn)練數(shù)據(jù)集的圖片輸入和正確表示數(shù)字
 batch_xs, batch_ys = mnist.train.next_batch(100)
 #運(yùn)行剛才建立的梯度下降算法，x賦值為圖片輸入，y_賦值為正確的表示數(shù)字
 sess.run(train_step,feed_dict = {x:batch_xs, y_: batch_ys})
 
#tf.argmax獲取最大值的索引。比較運(yùn)算后的結(jié)果和本身結(jié)果是否相同。
#這步的結(jié)果應(yīng)該是[1,1,1,1,1,1,1,1,0,1...........1,1,0,1]這種形式。
#1代表正確，0代表錯(cuò)誤
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
 
#tf.cast先將數(shù)據(jù)轉(zhuǎn)換成float，防止求平均不準(zhǔn)確。
#tf.reduce_mean由于只有一個(gè)參數(shù)，就是上面那個(gè)數(shù)組的平均值。
accuracy = tf.reduce_mean(tf.cast(correct_prediction,"float"))
#輸出
print(sess.run(accuracy,feed_dict={x:mnist.test.images,y_: mnist.test.labels}))

計(jì)算結(jié)果如下

"C:\Program Files\Anaconda3\python.exe" D:/pycharmprogram/tensorflow_learn/softmax_learn/softmax_learn.py
Extracting data\train-images-idx3-ubyte.gz
Extracting data\train-labels-idx1-ubyte.gz
Extracting data\t10k-images-idx3-ubyte.gz
Extracting data\t10k-labels-idx1-ubyte.gz
WARNING:tensorflow:From C:\Program Files\Anaconda3\lib\site-packages\tensorflow\python\util\tf_should_use.py:175: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.
Instructions for updating:
Use `tf.global_variables_initializer` instead.
2018-05-14 15:49:45.866600: W C:\tf_jenkins\home\workspace\rel-win\M\windows\PY\35\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2018-05-14 15:49:45.866600: W C:\tf_jenkins\home\workspace\rel-win\M\windows\PY\35\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
0.9163
 
Process finished with exit code 0

如果限制，只更新參數(shù)W查看效果

"C:\Program Files\Anaconda3\python.exe" D:/pycharmprogram/tensorflow_learn/softmax_learn/softmax_learn.py
Extracting data\train-images-idx3-ubyte.gz
Extracting data\train-labels-idx1-ubyte.gz
Extracting data\t10k-images-idx3-ubyte.gz
Extracting data\t10k-labels-idx1-ubyte.gz
WARNING:tensorflow:From C:\Program Files\Anaconda3\lib\site-packages\tensorflow\python\util\tf_should_use.py:175: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.
Instructions for updating:
Use `tf.global_variables_initializer` instead.
2018-05-14 15:51:08.543600: W C:\tf_jenkins\home\workspace\rel-win\M\windows\PY\35\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2018-05-14 15:51:08.544600: W C:\tf_jenkins\home\workspace\rel-win\M\windows\PY\35\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
0.9187
 
Process finished with exit code 0

可以看出只修改W對(duì)結(jié)果影響不大，如果設(shè)置只修改b

#導(dǎo)入Minst數(shù)據(jù)集
import input_data
mnist = input_data.read_data_sets("data",one_hot=True)
 
#導(dǎo)入tensorflow庫(kù)
import tensorflow as tf
 
#輸入變量，把28*28的圖片變成一維數(shù)組（丟失結(jié)構(gòu)信息）
x = tf.placeholder("float",[None,784])
 
#權(quán)重矩陣，把28*28=784的一維輸入，變成0-9這10個(gè)數(shù)字的輸出
w = tf.Variable(tf.zeros([784,10]))
#偏置
b = tf.Variable(tf.zeros([10]))
 
#核心運(yùn)算，其實(shí)就是softmax（x*w+b）
y = tf.nn.softmax(tf.matmul(x,w) + b)
 
#這個(gè)是訓(xùn)練集的正確結(jié)果
y_ = tf.placeholder("float",[None,10])
 
#交叉熵，作為損失函數(shù)
cross_entropy = -tf.reduce_sum(y_ * tf.log(y))
 
#梯度下降算法，最小化交叉熵
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy,var_list=[b])
 
#初始化，在run之前必須進(jìn)行的
init = tf.initialize_all_variables()
#創(chuàng)建session以便運(yùn)算
sess = tf.Session()
sess.run(init)
 
#迭代1000次
for i in range(1000):
 #獲取訓(xùn)練數(shù)據(jù)集的圖片輸入和正確表示數(shù)字
 batch_xs, batch_ys = mnist.train.next_batch(100)
 #運(yùn)行剛才建立的梯度下降算法，x賦值為圖片輸入，y_賦值為正確的表示數(shù)字
 sess.run(train_step,feed_dict = {x:batch_xs, y_: batch_ys})
 
#tf.argmax獲取最大值的索引。比較運(yùn)算后的結(jié)果和本身結(jié)果是否相同。
#這步的結(jié)果應(yīng)該是[1,1,1,1,1,1,1,1,0,1...........1,1,0,1]這種形式。
#1代表正確，0代表錯(cuò)誤
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
 
#tf.cast先將數(shù)據(jù)轉(zhuǎn)換成float，防止求平均不準(zhǔn)確。
#tf.reduce_mean由于只有一個(gè)參數(shù)，就是上面那個(gè)數(shù)組的平均值。
accuracy = tf.reduce_mean(tf.cast(correct_prediction,"float"))
#輸出
print(sess.run(accuracy,feed_dict={x:mnist.test.images,y_: mnist.test.labels}))

計(jì)算結(jié)果：

"C:\Program Files\Anaconda3\python.exe" D:/pycharmprogram/tensorflow_learn/softmax_learn/softmax_learn.py
Extracting data\train-images-idx3-ubyte.gz
Extracting data\train-labels-idx1-ubyte.gz
Extracting data\t10k-images-idx3-ubyte.gz
Extracting data\t10k-labels-idx1-ubyte.gz
WARNING:tensorflow:From C:\Program Files\Anaconda3\lib\site-packages\tensorflow\python\util\tf_should_use.py:175: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.
Instructions for updating:
Use `tf.global_variables_initializer` instead.
2018-05-14 15:52:04.483600: W C:\tf_jenkins\home\workspace\rel-win\M\windows\PY\35\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2018-05-14 15:52:04.483600: W C:\tf_jenkins\home\workspace\rel-win\M\windows\PY\35\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
0.1135
 
Process finished with exit code 0

如果只更新b那么對(duì)效果影響很大。

以上這篇在Tensorflow中實(shí)現(xiàn)梯度下降法更新參數(shù)值就是小編分享給大家的全部?jī)?nèi)容了，希望能給大家一個(gè)參考，也希望大家多多支持腳本之家。

您可能感興趣的文章: