强制数组中非零元素之间的最小间距

Enforce minimum spacing between non-zero elements in array

我正在尝试生成零和一的数组,其中零和一之间的间距是泊松分布的。

我相信这段代码可以解决问题:

import numpy as np

dt = 0.05
n_t = 500 / dt  # desired length of array
r = 100
spike_train = (np.random.rand(n_t) < r * dt).astype(int)

但是,我还想在 1 之间强制执行最小间距 - 例如任意两个 1 之间至少有两个零。

执行此操作的有效方法是什么?

我想在这个(不太优雅的)逻辑中保持分布:

def assure_2zeros(l):

    for idx in range(len(l)):
        # detect the problem when the ones are adjacent
        if l[idx] == 1 and l[idx+1] == 1:
            # check forward for a zero that could be realocated to split the ones
            for i in range(idx+2, len(l)-1):
                # check to not create other problems
                if l[i] == 0 and l[i-1]== 0 and l[i+1]== 0:
                    del l[i]
                    l.insert(idx+1, 0)
                    break
            # if doesnt found any zero forward, check backward
            else:
                for i in range(idx-1, 0, -1):
                    if l[i] == 0 and l[i-1]== 0 and l[i+1]== 0:
                        del l[i]
                        l.insert(idx+1, 0)
                        break

        # detects the problem when there are one zero between the ones
        if l[idx] == 1 and l[idx+2] == 1:
            for i in range(idx+3, len(l)-1):
                if l[i] == 0 and l[i-1]== 0 and l[i+1]== 0:
                    del l[i]
                    l.insert(idx+1, 0)
                    break
            else:
                for i in range(idx-1, 0, -1):
                    if l[i] == 0 and l[i-1]== 0 and l[i+1]== 0:
                        del l[i]
                        l.insert(idx+1, 0)
                        break

    return l

注意当有两个相邻的时,第一个if会在它们之间插入一个零,然后在第二个if中输入,并插入第二个零。但是,当列表的末尾或开头有两个或一个零时,它会失败。可以在 if 语句中添加更多条件,但对于您的情况,有很多零,我认为它可以工作。

这是一个相当有效的方法。它的工作原理是 (1) 首先忽略最短等待时间。 (2) 计算事件间时间 (3) 添加最小等待时间,(4) 返回到绝对时间,丢弃已从右端移出的事件。它可以在不到一秒的时间内创建 10**7 个样本。

import numpy as np

def train(T, dt, rate, min_wait):
    p = dt*rate
    # correct for minimum wait
    if p:
        p = 1 / (1 / p - min_wait) 
    if p > 0.1:
        print("warning: probability too high for approximation to be good")
    n = int(np.ceil(T/dt))
    raw_times, = np.where(np.random.random(n) < p)
    raw_times[1:] += min_wait - raw_times[:-1]
    good_times = raw_times.cumsum()
    cut = good_times.searchsorted(n)
    result = np.zeros(n, int)
    result[good_times[:cut]] = 1
    return result