梯度下降法python实现

川长思鸟来 2021-09-26 19:58 463阅读 0赞

参考博客：[https://blog.csdn.net/huahuazhu/article/details/73385362][https_blog.csdn.net_huahuazhu_article_details_73385362]

参考的博客中的示例代码是直线，学习后将其改为曲线

#!/usr/bin/python
    #coding=utf-8
    import numpy as np
    import matplotlib.pyplot as plt
    
    # 构造训练数据
    x = np.arange(0., 10., 0.2)
    # 训练数据点数目
    m = len(x)
    #打印训练数据点数目
    print(m)
    #创建矩阵1*m,每个元素值为1.0
    x0 = np.full(m, 1.0)
    x1 = []
    for i in x:
        x1.append(i**2)
    input_data = np.vstack([x0, x,x1]).T  # 将偏置b作为权向量的第一个分量，权重w1作为第二个分量,权重w2作为第二个分量
    target_data = x**2+2 * x + 5 + np.random.randn(m)
    
    # 两种终止条件
    loop_max = 10000  # 最大迭代次数(防止死循环)
    epsilon = 1e-3
    
    # 初始化权值
    np.random.seed(0)
    theta = np.random.randn(3)#初始化theta
    
    alpha = 0.00001  # 步长(注意取值过大会导致振荡即不收敛,过小收敛速度变慢)
    diff = 0.
    #创建0矩阵
    error = np.zeros(3)
    count = 0  # 循环次数
    finish = 0  # 终止标志
    
    while count < loop_max:
        count += 1
    
        # 标准梯度下降是在权值更新前对所有样例汇总误差，而随机梯度下降的权值是通过考查某个训练样例来更新的
        # 在标准梯度下降中，权值更新的每一步对多个样例求和，需要更多的计算
        sum_m = np.zeros(3)
        for i in range(m):
            dif = (np.dot(theta, input_data[i]) - target_data[i]) * input_data[i]
            sum_m = sum_m + dif  # 当alpha取值过大时,sum_m会在迭代过程中会溢出
    
        theta = theta - alpha * sum_m  # 注意步长alpha的取值,过大会导致振荡
        # 判断是否已收敛
        if np.linalg.norm(theta - error) < epsilon:
            finish = 1
            break
        else:
            error = theta
        print ('loop count = %d' % count, '\tw:',theta)
    print ('loop count = %d' % count, '\tw:',theta)
    
    #绘图
    plt.plot(x, target_data, 'g.')
    plt.plot(x, theta[2]*x**2+theta[1] * x + theta[0], 'r')
    plt.show()

实验结果：  
![这里写图片描述][70]

补充一些函数说明：  
1.scipy.stats.linregress  
scipy.stats.linregress(x, y=None)\[source\]

> Calculate a linear least-squares regression for two sets of measurements.  
> 计算两组测量的线性最小二乘回归。  
> **Parameters**:  
> x, y : array\_like  
> Two sets of measurements. Both arrays should have the same length. If only x is given (and y=None), then it must be a two-dimensional array where one dimension has length 2. The two sets of measurements are then found by splitting the array along the length-2 dimension.
> 
> **Returns:**  
> *slope* : float  
> slope of the regression line
> 
> *intercept* : float  
> intercept of the regression line
> 
> *rvalue* : float  
> correlation coefficient
> 
> *pvalue* : float  
> two-sided p-value for a hypothesis test whose null hypothesis is that the slope is zero, using Wald Test with t-distribution of the test statistic.
> 
> *stderr* : float  
> Standard error of the estimated gradient.  
> 例如：

>>> import matplotlib.pyplot as plt
    >>> from scipy import stats
    >>>> np.random.seed(100)
    >>> x = np.random.random(10)
    >>> print(x)
    [0.54340494 0.27836939 0.42451759 0.84477613 0.00471886 0.12156912
     0.67074908 0.82585276 0.13670659 0.57509333]
    >>> y = np.random.random(10)
    >>> print(y)
    [0.89132195 0.20920212 0.18532822 0.10837689 0.21969749 0.97862378
     0.81168315 0.17194101 0.81622475 0.27407375]
    >>> slope, intercept, r_value, p_value, std_err = stats.linregress(x, y)
    >>> print('slope=%s, intercept=%s'%(slope, intercept))
    slope=-0.3442654203664301, intercept=0.6190108485958797
    >>> print('r_value=%s, p_value=%s,std_err=%s'%(p_value, p_value,std_err))
    r_value=0.41832390506284134, p_value=0.41832390506284134,std_err=0.4034667235997838
    >>> plt.plot(x, y, 'o', label='original data')
    [<matplotlib.lines.Line2D object at 0x000001AC5B23B198>]
    >>> plt.plot(x, intercept + slope*x, 'r', label='fitted line')
    [<matplotlib.lines.Line2D object at 0x000001AC534240B8>]
    >>> plt.legend()
    <matplotlib.legend.Legend object at 0x000001AC5B23B8D0>
    >>> plt.show()

![这里写图片描述][70 1]

2.np.random.seed(0)  
作用：使得随机数据可预测。  
即不设置随机种子，每次生成的随机数都一样

>>> import numpy as np
    >>> np.random.random(5)
    array([0.75967512, 0.981992  , 0.15152915, 0.49788026, 0.21685212])
    >>> np.random.random(5)
    array([0.14585632, 0.6247151 , 0.7466104 , 0.44686791, 0.19680322])

若设置随机种子，则每次生成的数据都一样

>>> np.random.seed(5)
    >>> np.random.random(5)
    array([0.22199317, 0.87073231, 0.20671916, 0.91861091, 0.48841119])
    >>> np.random.seed(5)
    >>> np.random.random(5)
    array([0.22199317, 0.87073231, 0.20671916, 0.91861091, 0.48841119])

3.x = np.full((1,2), 1)  
创建1\*2的矩阵，每个元素为1

>>> x = np.full((1,2),1)
    >>> print(x)
    [[1 1]]

3.np.vstack(\[x,y\]) np.stack(\[x,y\]) np.hstack(\[x,y\])

>>> x=(1,2,3)
    >>> y=(4,5,6)
    >>>> z = np.vstack([x,y])
    >>> print(z)
    [[1 2 3] [4 5 6]]

>>> z = np.hstack([x,y])
    >>> print(z)
    [1 2 3 4 5 6]
    >>>

>>> z = np.stack([x,y])
    >>> print(z)
    [[1 2 3] [4 5 6]]
    >>> z = np.stack([x,y],1)
    >>> print(z)
    [[1 4] [2 5] [3 6]]

[https_blog.csdn.net_huahuazhu_article_details_73385362]: https://blog.csdn.net/huahuazhu/article/details/73385362
[70]: /images/20210923/3a472d3948594fb08def3ad301b5ca5d.png
[70 1]: /images/20210923/f9ef3cf83afe40478af9d55480bb18ec.png