正如@Henry 所说,您所追求的并不是纯粹的均匀分布。
为了强制执行您的约束,我认为最好的解决方案是在您的随机数生成器中包含一个衰减因子。随着数字序列的增加,该数字接下来出现的概率会降低。
我在python 3中实现了一些原型代码,因为我的cpp技能目前有点生疏,但基本概念很容易翻译成cpp。这里是:
def my_random(range: int, iterations: int, decay_rate :float = 2) -> List[int]:
assert range > 0, "`range` must be a positive non-zero integer"
if range == 1:
return [0] * iterations
last_num: int = 0
last_prob: float = 1/range
rand_num_lst: List[int] = []
while iterations > 0:
rnd = random() # generates a random number: 0 <= rnd < 1
if rnd < last_prob:
num = last_num
last_prob /= decay_rate
else:
# The `int` function is converting the float into integer by
# flooring the number
num = int( (rnd - last_prob) / (1 - last_prob) * (range - 1) )
if num >= last_num:
num += 1
last_num = num
last_prob = 1/range/decay_rate
rand_num_lst.append(num)
iterations -= 1
return rand_num_lst
请注意,在 Python3 中默认除法是浮点除法,这意味着 1/2 = 0.5 而不是在 Python2 中发生的 1/2 = 0。
我进行了一些测试来检查最大序列长度以及由此生成的数字分布是否仍然均匀分布,并且它似乎继续保持这些属性:
以range = 2 和不同的衰减率运行:
decay_rate: 2.00000 max sequence length: 6 number count: {0: 499830, 1: 500170}
decay_rate: 1.50000 max sequence length: 6 number count: {0: 499455, 1: 500545}
decay_rate: 1.25000 max sequence length: 9 number count: {0: 500241, 1: 499759}
decay_rate: 1.12500 max sequence length: 11 number count: {0: 499799, 1: 500201}
decay_rate: 1.06250 max sequence length: 14 number count: {0: 500655, 1: 499345}
decay_rate: 1.03125 max sequence length: 16 number count: {0: 500495, 1: 499505}
decay_rate: 1.01562 max sequence length: 16 number count: {0: 500010, 1: 499990}
decay_rate: 1.00781 max sequence length: 18 number count: {0: 499748, 1: 500252}
decay_rate: 1.00391 max sequence length: 18 number count: {0: 499987, 1: 500013}
decay_rate: 1.00195 max sequence length: 21 number count: {0: 499503, 1: 500497}
decay_rate: 1.00098 max sequence length: 21 number count: {0: 500495, 1: 499505}
decay_rate: 1.00000 max sequence length: 19 number count: {0: 499451, 1: 500549}
以range = 5 和不同的衰减率运行:
decay_rate: 2.00000 max sequence length: 5 number count: {0: 200314, 1: 199245, 2: 200213, 3: 199962, 4: 200266}
decay_rate: 1.50000 max sequence length: 5 number count: {0: 199372, 1: 199829, 2: 199937, 3: 200527, 4: 200335}
decay_rate: 1.25000 max sequence length: 6 number count: {0: 199373, 1: 199784, 2: 200561, 3: 200062, 4: 200220}
decay_rate: 1.12500 max sequence length: 8 number count: {0: 199752, 1: 199931, 2: 200579, 3: 200287, 4: 199451}
decay_rate: 1.06250 max sequence length: 8 number count: {0: 199280, 1: 200286, 2: 199688, 3: 200446, 4: 200300}
decay_rate: 1.03125 max sequence length: 8 number count: {0: 199577, 1: 199582, 2: 200652, 3: 199870, 4: 200319}
decay_rate: 1.01562 max sequence length: 9 number count: {0: 200442, 1: 199916, 2: 200142, 3: 199729, 4: 199771}
decay_rate: 1.00781 max sequence length: 9 number count: {0: 199784, 1: 200544, 2: 199921, 3: 199557, 4: 200194}
decay_rate: 1.00391 max sequence length: 9 number count: {0: 199920, 1: 199054, 2: 200303, 3: 200833, 4: 199890}
decay_rate: 1.00195 max sequence length: 9 number count: {0: 200011, 1: 200530, 2: 199806, 3: 200321, 4: 199332}
decay_rate: 1.00098 max sequence length: 10 number count: {0: 199741, 1: 199861, 2: 199822, 3: 200081, 4: 200495}
decay_rate: 1.00000 max sequence length: 9 number count: {0: 199717, 1: 199184, 2: 200182, 3: 200891, 4: 200026}
当然,您可以明确地编写如下代码:如果运行序列的长度大于X,则忽略该数字并生成与最后一个随机数不同的数字。虽然我不确定这种方法是否会继续均匀分布。