beta distribution

beta distribution

import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import beta
import d2l.torch as d2l
d2l.use_svg_display()

def plot(alist, blist):
    for a in alist:
        for b in blist:
            x = np.linspace(0, 1, 100) # 后面这个个数的选取只会影响ab小于1的时候的图形
            y = beta.pdf(x, a, b)
            plt.plot(x, y, label=f'a={a}, b={b}')
    plt.legend()

U 形是没有意义的…因为 $\alpha$ 和 $\beta$ 的选择都太小了, 一般选不到这么小

alist = [0.5, 1]
blist = [0.5, 1]
plot(alist, blist)

可以把 $\alpha$ 看作成功的概率, $\beta$ 看作失败的概率

alist = [36, 78]
blist = [13, 79]
plot(alist, blist)

alist = [10, 100]
blist = [10, 100]
plot(alist, blist)

参考文献:
推荐算法之Thompson(汤普森)采样
如何通俗理解 beta 分布？