分享一下我老师大神的人工智能教程!零基础,通俗易懂!http://blog.csdn.net/jiangjunshow
也欢迎大家转载本篇文章。分享知识,造福人民,实现我们中华民族伟大复兴!
本文不手动实现最小二乘,调用scipy库中实现好的相关优化函数。
考虑如下的含有4个参数的函数式:
f(x)=A−D1+(x/C)B+D
构造数据
import numpy as npfrom scipy import optimizeimport matplotlib.pyplot as pltdef logistic4(x, A, B, C, D): return (A-D)/(1+(x/C)**B)+Ddef residuals(p, y, x): A, B, C, D = p return y - logisctic4(x, A, B, C, D)def peval(x, p): A, B, C, D = p return logistic4(x, A, B, C, D) A, B, C, D = .5, 2.5, 8, 7.3x = np.linspace(0, 20, 20)y_true = logistic4(x, A, B, C, D)y_meas = y_true + 0.2 * np.random.randn(len(y_true))
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
调用工具箱函数,进行优化
p0 = [1/2]*4plesq = optimize.leastsq(residuals, p0, args=(y_meas, x)) # leastsq函数的功能其实是根据误差(y_meas-y_true) # 估计模型(也即函数)的参数
- 1
- 2
- 3
- 4
- 5
绘图
plt.figure(figsize=(6, 4.5))plt.plot(x, peval(x, plesq[0]), x, y_meas, 'o', x, y_true)plt.legend(['Fit', 'Noisy', 'True'], loc='upper left')plt.title('least square for the noisy data (measurements)')for i, (param, true, est) in enumerate(zip('ABCD', [A, B, C, D], plesq[0])): plt.text(11, 2-i*.5, '{} = {:.2f}, est({:.2f}) = {:.2f}'.format(param, true, param, est))plt.savefig('./logisitic.png')plt.show()
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8