参考了ailearning
导入基础包:
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
多项式拟合
导入线多项式拟合工具:
from numpy import polyfit, poly1d
产生数据:
x = np.linspace(-3, 3, 100)
y = 4 * x + 1.5
noise_y = y + np.random.randn(y.shape[-1]) * 1.5
画出:
plt.xlabel("x")
plt.ylabel("y")
plt.title("y = 4x+3/2")
fitline1 = np.polyfit(x, noise_y, 1)
f = np.poly1d(fitline1)
p = plt.plot(x, f(x), label = "1fitline")
p = plt.scatter(x, noise_y,label = "noiseline")
p = plt.plot(x, y, label = "line")
plt.legend()
plt.figure()
进行线性拟合,polyfit
是多项式拟合函数,线性拟合即一阶多项式:
用 poly1d
生成一个以传入的 coeff
为参数的多项式函数:
多项式拟合正弦函数
正弦函数:
x = np.linspace(-np.pi,np.pi,100)
y = np.sin(x)
用一阶到九阶多项式拟合,类似泰勒展开:
y1 = poly1d(polyfit(x,y,1))
y3 = poly1d(polyfit(x,y,3))
y5 = poly1d(polyfit(x,y,5))
y7 = poly1d(polyfit(x,y,7))
y9 = poly1d(polyfit(x,y,9))
x = np.linspace(-3 * np.pi,3 * np.pi,100)
p = plt.plot(x, np.sin(x), 'k')
p = plt.plot(x, y1(x))
p = plt.plot(x, y3(x))
p = plt.plot(x, y5(x))
p = plt.plot(x, y7(x))
p = plt.plot(x, y9(x))
a = plt.axis([-3 * np.pi, 3 * np.pi, -1.25, 1.25])
可以看到,随着多项式拟合的阶数的增加,曲线与拟合数据的吻合程度在逐渐增大。
最小二乘拟合
导入相关的模块:
from scipy.linalg import lstsq
from scipy.stats import linregress
x = np.linspace(0,5,100)
y = 0.5 * x + np.random.randn(x.shape[-1]) * 0.35
plt.plot(x,y,'x')
Scipy.linalg.lstsq 最小二乘解
要得到 C
,可以使用 scipy.linalg.lstsq
求最小二乘解。
这里,我们使用 1 阶多项式即 N = 2
,先将 x
扩展成 X
:
x = np.linspace(-3, 3, 50)
y = np.sin(x)
noise_y = y + np.random.randn(y.shape[-1]) * 0.5
X = np.hstack((x[:, np.newaxis], np.ones((x.shape[-1], 1))))
C, resid, rank, s = sp.linalg.lstsq(X, noise_y)
A, B, X, X, X = sp.stats.linregress(x, noise_y)
f1 = np.poly1d(np.polyfit(x, noise_y, 1));
p = plt.plot(x, f1(x), label = "fit")
p = plt.plot(x, A * x + B, label = "linregress")
p = plt.plot(x, C[0] * x + C[1], label = "lstsq")
p = plt.scatter(x, noise_y,label = "noise")
plt.legend()