[英]Is the numpy random generator biased?
The numpy.random.choice method can generate a random sample without replacement if different elements should have different probabilities. 如果不同的元素应该具有不同的概率,则numpy.random.choice方法可以生成一个随机样本而无需替换。 However, when I test it with 但是,当我用
import numpy
a = [0, 1, 2, 3, 4, 5]
p = [0.1, 0.3, 0.3, 0.1, 0.1, 0.1]
result = [0, 0, 0, 0, 0, 0]
N = 1000000
k = 3
for i in range(0, N):
temp = numpy.random.choice(a, k, False, p)
for j in temp:
result[j] += 1
for i in range(0, 6):
result[i] /= (N * k)
print(result)
the second and third elements only show up 25% of the time which is off by a lot. 第二个和第三个元素只显示25%的时间,这相差很大。 I tried different probability distributions (eg, [0.1, 0.2, 0.3, 0.1, 0.1, 0.2]) and every time the result didn't match the expectation. 我尝试了不同的概率分布(例如[0.1、0.2、0.3、0.1、0.1、0.2]),并且每次结果都不符合预期时。 Is there an issue with my code or is numpy really that inaccurate? 我的代码有问题吗?还是numpy确实不正确?
Your understanding of the np.random.choice
function is wrong. 您对np.random.choice
函数的理解是错误的。 Specifically the replace=
option. 特别是replace=
选项。 The documentation suggests that replace=False
means that once an item has been chosen, it can't be chosen again. 该文档建议replace=False
表示一旦选择了一项,就不能再次选择它。 This can be shown by running 这可以通过运行来显示
for _ in range(100):
assert set(np.random.choice(np.arange(5), 5, replace=False)) == set(range(5))
and seeing no error is ever raised. 并没有发现任何错误。 The order changes, but all 5 values must be returned. 顺序更改,但是必须返回所有5个值。
Your current method is giving strange results because of this property. 由于该属性,您当前的方法给出了奇怪的结果。 Even though 1 and 2 have a 0.3 chance of appearing as the first item, they have a less than 0.3 chance of appearing as the second or third item because if they were the first item, they can't be a later item. 即使1和2出现在第一项中的机率是0.3,但它们出现在第二或第三项中的机率却小于0.3,因为如果它们是第一项,那么它们就不能成为后一项。
The solution is obviously to use replace=True
(or ignore, True
is the default) like so: 解决方案显然是使用replace=True
(或忽略,默认为True
),如下所示:
import numpy as np
a = [0, 1, 2, 3, 4, 5]
p = [0.1, 0.3, 0.3, 0.1, 0.1, 0.1]
n = 100_000
choices = np.random.choice(a, n, p=p)
values, counts = np.unique(choices, return_counts=True)
result = dict(zip(values, counts / n))
# result == {0: 0.10063, 1: 0.30018, 2: 0.30003, 3: 0.09916, 4: 0.10109, 5: 0.09891}
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.