TensorFlow梯度下降法中var_list参数对权重W与偏置b更新的影响分析

Luna

发布时间：2026-10-13| 4人浏览过

原创

本文通过TensorFlow代码示例，深入解析梯度下降法中var_list参数的作用。默认情况下，优化器会同时更新权重矩阵W（[784,10]）和偏置b。实验对比显示，若仅更新权重W，模型准确率约为0.9187，影响有限；而若仅更新偏置b，准确率骤降至0.1135，效果显著变差。这证明了在Softmax回归中，合理分配W和b的更新策略对模型性能至关重要。

TensorFlow梯度下降法参数更新示例

直接展示相关代码实现。


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

在TensorFlow中，通过梯度下降法对损失函数涉及的变量进行数值更新，默认情况下会修改由tf.Variable(tf.zeros([784,10]))定义的变量参数。

作为Variable的初始化参数。


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

此外，还可以利用var_list参数来指定具体需要更新哪些参数的值。


#导入Minst数据集
import input_data
mnist = input_data.read_data_sets("data",one_hot=True)
 
#导入tensorflow库
import tensorflow as tf
 
#输入变量，把28*28的图片变成一维数组（丢失结构信息）
x = tf.placeholder("float",[None,784])
 
#权重矩阵，把28*28=784的一维输入，变成0-9这10个数字的输出
w = tf.Variable(tf.zeros([784,10]))
#偏置
b = tf.Variable(tf.zeros([10]))
 
#核心运算，其实就是softmax（x*w+b）
y = tf.nn.softmax(tf.matmul(x,w) + b)
 
#这个是训练集的正确结果
y_ = tf.placeholder("float",[None,10])
 
#交叉熵，作为损失函数
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之前必须进行的
init = tf.initialize_all_variables()
#创建session以便运算
sess = tf.Session()
sess.run(init)
 
#迭代1000次
for i in range(1000):
 #获取训练数据集的图片输入和正确表示数字
 batch_xs, batch_ys = mnist.train.next_batch(100)
 #运行刚才建立的梯度下降算法，x赋值为图片输入，y_赋值为正确的表示数字
 sess.run(train_step,feed_dict = {x:batch_xs, y_: batch_ys})
 
#tf.argmax获取最大值的索引。比较运算后的结果和本身结果是否相同。
#这步的结果应该是[1,1,1,1,1,1,1,1,0,1...........1,1,0,1]这种形式。
#1代表正确，0代表错误
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
 
#tf.cast先将数据转换成float，防止求平均不准确。
#tf.reduce_mean由于只有一个参数，就是上面那个数组的平均值。
accuracy = tf.reduce_mean(tf.cast(correct_prediction,"float"))
#输出
print(sess.run(accuracy,feed_dict={x:mnist.test.images,y_: mnist.test.labels}))

运行上述代码得到的计算结果如下所示。

默认更新W与b的准确率结果


"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 15:49:45.866600: 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

为了观察限制更新范围的效果，这里尝试仅更新参数W。

仅更新权重W的效果分析


"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

从结果可以看出，仅修改W对整体模型性能的影响相对较小；接着测试如果设置仅修改偏置b的情况。


#导入Minst数据集
import input_data
mnist = input_data.read_data_sets("data",one_hot=True)
 
#导入tensorflow库
import tensorflow as tf
 
#输入变量，把28*28的图片变成一维数组（丢失结构信息）
x = tf.placeholder("float",[None,784])
 
#权重矩阵，把28*28=784的一维输入，变成0-9这10个数字的输出
w = tf.Variable(tf.zeros([784,10]))
#偏置
b = tf.Variable(tf.zeros([10]))
 
#核心运算，其实就是softmax（x*w+b）
y = tf.nn.softmax(tf.matmul(x,w) + b)
 
#这个是训练集的正确结果
y_ = tf.placeholder("float",[None,10])
 
#交叉熵，作为损失函数
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之前必须进行的
init = tf.initialize_all_variables()
#创建session以便运算
sess = tf.Session()
sess.run(init)
 
#迭代1000次
for i in range(1000):
 #获取训练数据集的图片输入和正确表示数字
 batch_xs, batch_ys = mnist.train.next_batch(100)
 #运行刚才建立的梯度下降算法，x赋值为图片输入，y_赋值为正确的表示数字
 sess.run(train_step,feed_dict = {x:batch_xs, y_: batch_ys})
 
#tf.argmax获取最大值的索引。比较运算后的结果和本身结果是否相同。
#这步的结果应该是[1,1,1,1,1,1,1,1,0,1...........1,1,0,1]这种形式。
#1代表正确，0代表错误
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
 
#tf.cast先将数据转换成float，防止求平均不准确。
#tf.reduce_mean由于只有一个参数，就是上面那个数组的平均值。
accuracy = tf.reduce_mean(tf.cast(correct_prediction,"float"))
#输出
print(sess.run(accuracy,feed_dict={x:mnist.test.images,y_: mnist.test.labels}))

此时的计算结果如下。

仅更新偏置b的效果分析


"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

实验表明，如果只更新b，会对最终效果产生巨大的负面影响。

以上关于在Tensorflow中利用梯度下降法更新参数值的实现方法就是本文分享的全部内容，希望能为大家提供参考，也希望大家多多支持本站。