逆变换采样
逆变换采样(英语:inverse transform sampling),又称为逆采样(inversion sampling)、逆概率积分变换(inverse probability integral transform)、逆变换法(inverse transformation method)、斯米尔诺夫变换(Smirnov transform)、黄金法则(golden rule)等,是伪随机数采样的一种基本方法。在已知任意概率分布的累积分布函数时,可用于从该分布中生成随机样本。
import numpy as np import matplotlib.pyplot as plt def sampleExp(Lambda = 2,maxCnt = 50000): ys = [] standardXaxis = [] standardExp = [] for i in range(maxCnt): u = np.random.random() y = -1/Lambda*np.log(1-u) #F-1(X) ys.append(y) for i in range(1000): t = Lambda * np.exp(-Lambda*i/100) standardXaxis.append(i/100) standardExp.append(t) plt.plot(standardXaxis,standardExp,'r') #plt.hist(ys,1000,normed=True) # python 2 plt.hist(ys,1000,density=True) # python 3 plt.show() sampleExp()
————————————————
import numpy as np import matplotlib.pyplot as plt def f1(x): return (0.3*np.exp(-(x-0.3)**2) + 0.7* np.exp(-(x-2.)**2/0.3))/1.2113 x = np.arange(-4.,6.,0.01) plt.plot(x,f1(x),color = "red") size = int(1e+07) sigma = 1.2 z = np.random.normal(loc = 1.4,scale = sigma, size = size) qz = 1/(np.sqrt(2*np.pi)*sigma)*np.exp(-0.5*(z-1.4)**2/sigma**2) k = 2.5 #z = np.random.uniform(low = -4, high = 6, size = size) #qz = 0.1 #k = 10 u = np.random.uniform(low = 0, high = k*qz, size = size) pz = 0.3*np.exp(-(z-0.3)**2) + 0.7* np.exp(-(z-2.)**2/0.3) sample = z[pz >= u] #plt.hist(sample,bins=150, normed=True, edgecolor='black') # python 2 plt.hist(sample,bins=150, density=True, edgecolor='black') # python 3 plt.show()
————————————————
Inverse transform sampling反变换采样法
https://blog.csdn.net/eric2016_lv/article/details/81191430
(Good)逆采样(Inverse Sampling)和拒绝采样(Reject Sampling)原理详解
https://blog.csdn.net/anshuai_aw1/article/details/84840446
逆变换采样
https://baike.baidu.com/item/%E9%80%86%E5%8F%98%E6%8D%A2%E9%87%87%E6%A0%B7/22934385?fr=aladdin
参考文献
【2】采样方法(一)
【3】蒙特卡洛采样之拒绝采样(Reject Sampling)
【4】一文了解采样方法