TensorFlow实战--Chapter04单变量线性回归

约定不等于承诺〃 2022-10-23 08:02 244阅读 0赞

TensorFlow实战–Chapter04单变量线性回归

在这里插入图片描述

使用tensorflow实现单变量回归模型

文章目录

  • TensorFlow实战—Chapter04单变量线性回归
  • 监督式机器学习的基本术语
    • 标签和特征
    • 训练
    • 损失
    • 定义损失函数
    • 模型训练与降低损失
    • 样本和模型
  • 线性回归问题TensorFlow实战
    • 人工数据生成
    • 利用matplotlib绘图
    • 定义模型
    • 模型训练
    • 创建会话,变量初始化
    • 迭代训练
    • 打印结果
    • 可视化
    • 进行预测
    • 显示损失值
    • 图形现显示话损失值

使用TensorFlow进行算法设计与训练的核心步骤

  1. 准备数据
  2. 构建模型
  3. 训练模型
  4. 进行预测

上述步骤是我们使用TensorFlow进行算法设计与训练的核心步骤，贯穿于具体实践中。

监督式机器学习的基本术语

标签和特征

在这里插入图片描述

训练

在这里插入图片描述

损失

在这里插入图片描述
在这里插入图片描述

定义损失函数

在这里插入图片描述

模型训练与降低损失

在这里插入图片描述

样本和模型

在这里插入图片描述
在这里插入图片描述

线性回归问题TensorFlow实战

人工数据生成

  1. import warnings
  2. warnings.filterwarnings("ignore")
  3. import tensorflow as tf
  4. import numpy as np
  5. import matplotlib.pyplot as plt
  7. np.random.seed(5)

使用numpy生成等差数列的方法，生成100个点，每个点的取值在[-1,1]之间

y = 2x + 1 + 噪声,其中，噪声的维度与x_data一致

np.random.randn(x_data.shape)实参的前面加上和**，就意味着拆包，单个*号表示将元组拆成一个个单独的实参

  1. x_data = np.linspace(-1, 1, 100)
  2. y_data = 2 * x_data + 1.0 + np.random.randn(*x_data.shape) * 0.4

利用matplotlib绘图

  1. plt.scatter(x_data, y_data)
  2. plt.show()

在这里插入图片描述

  1. plt.plot(x_data, 2 * x_data + 1.0, color="red", linewidth=3)
  2. plt.show()

在这里插入图片描述

定义模型

定义训练数据的占位符，x是特征值，y是标签值

  1. x = tf.placeholder("float", name="x")
  2. y = tf.placeholder("float", name="y")

定义模型函数

  1. def model(x, w, b):
  2.     return tf.multiply(x, w) + b

创建变量

  • TensorFlow变量的声明函数是tf.Variable
  • tf,Variable的作用是保存和更新参数

  • 变量的初始值可以是随机数、常数，或是通过其他变量的初始值计算得到

    构建线性函数的斜率，变量2

    w = tf.Variable(1.0, name=”w0”)

    构建线性函数的截距，变量b

    b = tf.Variable(0.0, name=”b0”)

    pred是预测值，前向计算

    pred = model(x, w, b)

模型训练

  1. 设置训练参数

    train_epochs = 10 # 迭代次数
    learning_rate = 0.05 # 学习率

  2. 损失函数

    • 损失函数用于描述预测值和真实值之间的误差，从而指导模型的收敛方向
    • 常见的损失函数：
    1. 均方差(Mean Square Error)
    2. 交叉熵(cross-entropy)

    loss_function = tf.reduce_mean(tf.square(y - pred)) # 采用均方差

  3. 定义优化器
    优化器Optimizer，初始化一个GradientDescentOPtimizer
    设置学习率和优化目标：最小化损失

    梯度下降优化器

    optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(

    1. loss_function)

创建会话,变量初始化

  • 在真正进行计算之气，需将所有变量初始化

  • 通过==tf.global_variables_initializer()==函数可实现对所有变量的初始化

    sess = tf.Session()
    init = tf.global_variables_initializer()
    sess.run(init)

迭代训练

模型训练阶段，设置迭代轮次，通过将样本逐个输入模型，进行梯度优化操作。每轮迭代后，绘制出模型曲线

  1. for epoch in range(train_epochs):
  2.     for xs, ys in zip(x_data, y_data):
  3.         _, loss = sess.run([optimizer, loss_function],
  4.                            feed_dict={
  6.                                x: xs,
  7.                                y: ys
  8.                            })
  9.     b0temp = b.eval(session=sess)
  10.     w0temp = w.eval(session=sess)
  11.     plt.plot(x_data, w0temp * x_data + b0temp)

在这里插入图片描述

打印结果

  1. print("w:",sess.run(w))
  2. print("b:",sess.run(b))
  4. w: 1.9822965
  5. b: 1.0420128

可视化

  1. plt.scatter(x_data,y_data,label="original data")
  2. plt.plot(x_data,x_data*sess.run(w)+sess.run(b),label="fitter line",color="r",linewidth=3)
  3. plt.legend(loc=2)
  4. plt.show()

在这里插入图片描述

进行预测

  1. x_test = 3.21
  2. predict = sess.run(pred,feed_dict={
  3.     x:x_test})
  4. print(f"预测值:{
  5.       predict}")
  6. target = 3* x_test +1.0
  7. print(f"目标值:{
  8.       target}")
  10. 预测值:7.405184268951416
  11. 目标值:10.629999999999999

显示损失值

  1. step = 0
  2. loss_list = []
  3. display_step = 10
  4. for eopch in range(train_epochs):
  5.     for xs, ys in zip(x_data, y_data):
  6.         _, loss = sess.run([optimizer, loss_function],
  7.                            feed_dict={
  9.                                x: xs,
  10.                                y: ys
  11.                            })
  12.         loss_list.append(loss)
  13.         step += 1
  14.         if step % display_step == 0:
  15.             print(f"Train Epoch:{
  16.       eopch+1}", f"Step:{
  17.       step}", f"loss={
  18.       loss}")
  19.     b0temp = b.eval(session=sess)
  20.     w0temp = w.eval(session=sess)
  21.     plt.plot(x_data,w0temp*x_data+b0temp)
  23. Train Epoch:1 Step:10 loss=0.03621993958950043
  24. Train Epoch:1 Step:20 loss=0.08414854109287262
  25. Train Epoch:1 Step:30 loss=0.00047292912495322526
  26. Train Epoch:1 Step:40 loss=0.32646191120147705
  27. Train Epoch:1 Step:50 loss=0.027518080547451973
  28. Train Epoch:1 Step:60 loss=0.01023304183036089
  29. Train Epoch:1 Step:70 loss=0.12734712660312653
  30. Train Epoch:1 Step:80 loss=0.0010273485677316785
  31. Train Epoch:1 Step:90 loss=0.10288805514574051
  32. Train Epoch:1 Step:100 loss=0.048337195068597794
  33. Train Epoch:2 Step:110 loss=0.03621993958950043
  34. Train Epoch:2 Step:120 loss=0.08414854109287262
  35. Train Epoch:2 Step:130 loss=0.00047292912495322526
  36. Train Epoch:2 Step:140 loss=0.32646191120147705
  37. Train Epoch:2 Step:150 loss=0.027518080547451973
  38. Train Epoch:2 Step:160 loss=0.01023304183036089
  39. Train Epoch:2 Step:170 loss=0.12734712660312653
  40. Train Epoch:2 Step:180 loss=0.0010273485677316785
  41. Train Epoch:2 Step:190 loss=0.10288805514574051
  42. Train Epoch:2 Step:200 loss=0.048337195068597794
  43. Train Epoch:3 Step:210 loss=0.03621993958950043
  44. Train Epoch:3 Step:220 loss=0.08414854109287262
  45. Train Epoch:3 Step:230 loss=0.00047292912495322526
  46. Train Epoch:3 Step:240 loss=0.32646191120147705
  47. Train Epoch:3 Step:250 loss=0.027518080547451973
  48. Train Epoch:3 Step:260 loss=0.01023304183036089
  49. Train Epoch:3 Step:270 loss=0.12734712660312653
  50. Train Epoch:3 Step:280 loss=0.0010273485677316785
  51. Train Epoch:3 Step:290 loss=0.10288805514574051
  52. Train Epoch:3 Step:300 loss=0.048337195068597794
  53. Train Epoch:4 Step:310 loss=0.03621993958950043
  54. Train Epoch:4 Step:320 loss=0.08414854109287262
  55. Train Epoch:4 Step:330 loss=0.00047292912495322526
  56. Train Epoch:4 Step:340 loss=0.32646191120147705
  57. Train Epoch:4 Step:350 loss=0.027518080547451973
  58. Train Epoch:4 Step:360 loss=0.01023304183036089
  59. Train Epoch:4 Step:370 loss=0.12734712660312653
  60. Train Epoch:4 Step:380 loss=0.0010273485677316785
  61. Train Epoch:4 Step:390 loss=0.10288805514574051
  62. Train Epoch:4 Step:400 loss=0.048337195068597794
  63. Train Epoch:5 Step:410 loss=0.03621993958950043
  64. Train Epoch:5 Step:420 loss=0.08414854109287262
  65. Train Epoch:5 Step:430 loss=0.00047292912495322526
  66. Train Epoch:5 Step:440 loss=0.32646191120147705
  67. Train Epoch:5 Step:450 loss=0.027518080547451973
  68. Train Epoch:5 Step:460 loss=0.01023304183036089
  69. Train Epoch:5 Step:470 loss=0.12734712660312653
  70. Train Epoch:5 Step:480 loss=0.0010273485677316785
  71. Train Epoch:5 Step:490 loss=0.10288805514574051
  72. Train Epoch:5 Step:500 loss=0.048337195068597794
  73. Train Epoch:6 Step:510 loss=0.03621993958950043
  74. Train Epoch:6 Step:520 loss=0.08414854109287262
  75. Train Epoch:6 Step:530 loss=0.00047292912495322526
  76. Train Epoch:6 Step:540 loss=0.32646191120147705
  77. Train Epoch:6 Step:550 loss=0.027518080547451973
  78. Train Epoch:6 Step:560 loss=0.01023304183036089
  79. Train Epoch:6 Step:570 loss=0.12734712660312653
  80. Train Epoch:6 Step:580 loss=0.0010273485677316785
  81. Train Epoch:6 Step:590 loss=0.10288805514574051
  82. Train Epoch:6 Step:600 loss=0.048337195068597794
  83. Train Epoch:7 Step:610 loss=0.03621993958950043
  84. Train Epoch:7 Step:620 loss=0.08414854109287262
  85. Train Epoch:7 Step:630 loss=0.00047292912495322526
  86. Train Epoch:7 Step:640 loss=0.32646191120147705
  87. Train Epoch:7 Step:650 loss=0.027518080547451973
  88. Train Epoch:7 Step:660 loss=0.01023304183036089
  89. Train Epoch:7 Step:670 loss=0.12734712660312653
  90. Train Epoch:7 Step:680 loss=0.0010273485677316785
  91. Train Epoch:7 Step:690 loss=0.10288805514574051
  92. Train Epoch:7 Step:700 loss=0.048337195068597794
  93. Train Epoch:8 Step:710 loss=0.03621993958950043
  94. Train Epoch:8 Step:720 loss=0.08414854109287262
  95. Train Epoch:8 Step:730 loss=0.00047292912495322526
  96. Train Epoch:8 Step:740 loss=0.32646191120147705
  97. Train Epoch:8 Step:750 loss=0.027518080547451973
  98. Train Epoch:8 Step:760 loss=0.01023304183036089
  99. Train Epoch:8 Step:770 loss=0.12734712660312653
  100. Train Epoch:8 Step:780 loss=0.0010273485677316785
  101. Train Epoch:8 Step:790 loss=0.10288805514574051
  102. Train Epoch:8 Step:800 loss=0.048337195068597794
  103. Train Epoch:9 Step:810 loss=0.03621993958950043
  104. Train Epoch:9 Step:820 loss=0.08414854109287262
  105. Train Epoch:9 Step:830 loss=0.00047292912495322526
  106. Train Epoch:9 Step:840 loss=0.32646191120147705
  107. Train Epoch:9 Step:850 loss=0.027518080547451973
  108. Train Epoch:9 Step:860 loss=0.01023304183036089
  109. Train Epoch:9 Step:870 loss=0.12734712660312653
  110. Train Epoch:9 Step:880 loss=0.0010273485677316785
  111. Train Epoch:9 Step:890 loss=0.10288805514574051
  112. Train Epoch:9 Step:900 loss=0.048337195068597794
  113. Train Epoch:10 Step:910 loss=0.03621993958950043
  114. Train Epoch:10 Step:920 loss=0.08414854109287262
  115. Train Epoch:10 Step:930 loss=0.00047292912495322526
  116. Train Epoch:10 Step:940 loss=0.32646191120147705
  117. Train Epoch:10 Step:950 loss=0.027518080547451973
  118. Train Epoch:10 Step:960 loss=0.01023304183036089
  119. Train Epoch:10 Step:970 loss=0.12734712660312653
  120. Train Epoch:10 Step:980 loss=0.0010273485677316785
  121. Train Epoch:10 Step:990 loss=0.10288805514574051
  122. Train Epoch:10 Step:1000 loss=0.048337195068597794

在这里插入图片描述

图形现显示话损失值

  1. plt.plot(loss_list)

在这里插入图片描述

  1. plt.plot(loss_list,"r+")

在这里插入图片描述

到这里就结束了，如果对你有帮助，欢迎点赞关注评论，你的点赞对我很重要

在这里插入图片描述

发表评论

表情：
评论列表 （有 0 条评论，244人围观）

还没有评论，来说两句吧...

相关阅读