1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
|
import numpy as np
import matplotlib.pyplot as plt
import math
# Python实现正态分布
# 绘制正态分布概率密度函数
u = 0 # 均值μ
u01 = - 2
sig = math.sqrt( 0.2 ) # 标准差δ
sig01 = math.sqrt( 1 )
sig02 = math.sqrt( 5 )
sig_u01 = math.sqrt( 0.5 )
x = np.linspace(u - 3 * sig, u + 3 * sig, 50 )
x_01 = np.linspace(u - 6 * sig, u + 6 * sig, 50 )
x_02 = np.linspace(u - 10 * sig, u + 10 * sig, 50 )
x_u01 = np.linspace(u - 10 * sig, u + 1 * sig, 50 )
y_sig = np.exp( - (x - u) * * 2 / ( 2 * sig * * 2 )) / (math.sqrt( 2 * math.pi) * sig)
y_sig01 = np.exp( - (x_01 - u) * * 2 / ( 2 * sig01 * * 2 )) / (math.sqrt( 2 * math.pi) * sig01)
y_sig02 = np.exp( - (x_02 - u) * * 2 / ( 2 * sig02 * * 2 )) / (math.sqrt( 2 * math.pi) * sig02)
y_sig_u01 = np.exp( - (x_u01 - u01) * * 2 / ( 2 * sig_u01 * * 2 )) / (math.sqrt( 2 * math.pi) * sig_u01)
plt.plot(x, y_sig, "r-" , linewidth = 2 )
plt.plot(x_01, y_sig01, "g-" , linewidth = 2 )
plt.plot(x_02, y_sig02, "b-" , linewidth = 2 )
plt.plot(x_u01, y_sig_u01, "m-" , linewidth = 2 )
# plt.plot(x, y, 'r-', x, y, 'go', linewidth=2,markersize=8)
plt.grid( True )
plt.show()
|
效果:
以上就是python 绘制正态曲线的示例的详细内容,更多关于python 绘制正态曲线的资料请关注服务器之家其它相关文章!
原文链接:https://www.cnblogs.com/marszhw/archive/2004/01/13/10962964.html