我觉得使用瓶颈可以很好地解决这个问题

参见下面的基本示例:

import numpy as np
import bottleneck as bn

a = np.random.randint(4, 1000, size=100)
mm = bn.move_mean(a, window=5, min_count=1)

“mm”是“a”的移动平均值。 “窗口”是考虑移动均值的最大条目数。 "min_count"是考虑移动平均值的最小条目数(例如，对于前几个元素或如果数组有nan值)。

好在瓶颈有助于处理nan值，而且非常高效。

你可以用以下方法计算运行平均值:

import numpy as np

def runningMean(x, N):
    y = np.zeros((len(x),))
    for ctr in range(len(x)):
         y[ctr] = np.sum(x[ctr:(ctr+N)])
    return y/N

但是速度很慢。

幸运的是，numpy包含一个卷积函数，我们可以用它来加快速度。运行均值相当于将x与一个长度为N的向量进行卷积，其中所有元素都等于1/N。卷积的numpy实现包括起始瞬态，所以你必须删除前N-1点:

def runningMeanFast(x, N):
    return np.convolve(x, np.ones((N,))/N)[(N-1):]

在我的机器上，快速版本要快20-30倍，这取决于输入向量的长度和平均窗口的大小。

请注意，卷积确实包括一个“相同”模式，它似乎应该解决开始的瞬态问题，但它在开始和结束之间分割。

我的解决方案是基于维基百科上的“简单移动平均”。

from numba import jit
@jit
def sma(x, N):
    s = np.zeros_like(x)
    k = 1 / N
    s[0] = x[0] * k
    for i in range(1, N + 1):
        s[i] = s[i - 1] + x[i] * k
    for i in range(N, x.shape[0]):
        s[i] = s[i - 1] + (x[i] - x[i - N]) * k
    s = s[N - 1:]
    return s

与之前建议的解决方案相比，它比scipy最快的解决方案“uniform_filter1d”快两倍，并且具有相同的错误顺序。 速度测试:

import numpy as np    
x = np.random.random(10000000)
N = 1000

from scipy.ndimage.filters import uniform_filter1d
%timeit uniform_filter1d(x, size=N)
95.7 ms ± 9.34 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
%timeit sma(x, N)
47.3 ms ± 3.42 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

错误的比较:

np.max(np.abs(np.convolve(x, np.ones((N,))/N, mode='valid') - uniform_filter1d(x, size=N, mode='constant', origin=-(N//2))[:-(N-1)]))
8.604228440844963e-14
np.max(np.abs(np.convolve(x, np.ones((N,))/N, mode='valid') - sma(x, N)))
1.41886502547095e-13

对于一个简短、快速的解决方案，在一个循环中完成所有事情，没有依赖关系，下面的代码工作得很好。

mylist = [1, 2, 3, 4, 5, 6, 7]
N = 3
cumsum, moving_aves = [0], []

for i, x in enumerate(mylist, 1):
    cumsum.append(cumsum[i-1] + x)
    if i>=N:
        moving_ave = (cumsum[i] - cumsum[i-N])/N
        #can do stuff with moving_ave here
        moving_aves.append(moving_ave)

高效的解决方案

卷积比直接的方法好得多，但(我猜)它使用FFT，因此相当慢。但是，下面的方法特别适用于计算运行平均值

def running_mean(x, N):
    cumsum = numpy.cumsum(numpy.insert(x, 0, 0)) 
    return (cumsum[N:] - cumsum[:-N]) / float(N)

要检查的代码

In[3]: x = numpy.random.random(100000)
In[4]: N = 1000
In[5]: %timeit result1 = numpy.convolve(x, numpy.ones((N,))/N, mode='valid')
10 loops, best of 3: 41.4 ms per loop
In[6]: %timeit result2 = running_mean(x, N)
1000 loops, best of 3: 1.04 ms per loop

注意numpy。allclose(result1, result2)为True，两个方法等价。 N越大，时间差异越大。

警告:虽然cumsum更快，但会增加浮点错误，这可能导致您的结果无效/不正确/不可接受

这里的评论指出了这个浮点错误问题，但我在回答中让它更明显。

# demonstrate loss of precision with only 100,000 points
np.random.seed(42)
x = np.random.randn(100000)+1e6
y1 = running_mean_convolve(x, 10)
y2 = running_mean_cumsum(x, 10)
assert np.allclose(y1, y2, rtol=1e-12, atol=0)

the more points you accumulate over the greater the floating point error (so 1e5 points is noticable, 1e6 points is more significant, more than 1e6 and you may want to resetting the accumulators) you can cheat by using np.longdouble but your floating point error still will get significant for relatively large number of points (around >1e5 but depends on your data) you can plot the error and see it increasing relatively fast the convolve solution is slower but does not have this floating point loss of precision the uniform_filter1d solution is faster than this cumsum solution AND does not have this floating point loss of precision

有点晚了，但我已经做了我自己的小函数，它不环绕端点或垫与零，然后用于查找平均值。进一步的处理是，它还在线性间隔点上对信号进行重新采样。随意定制代码以获得其他特性。

该方法是一个简单的矩阵乘法与规范化高斯核。

def running_mean(y_in, x_in, N_out=101, sigma=1):
    '''
    Returns running mean as a Bell-curve weighted average at evenly spaced
    points. Does NOT wrap signal around, or pad with zeros.
    
    Arguments:
    y_in -- y values, the values to be smoothed and re-sampled
    x_in -- x values for array
    
    Keyword arguments:
    N_out -- NoOf elements in resampled array.
    sigma -- 'Width' of Bell-curve in units of param x .
    '''
    import numpy as np
    N_in = len(y_in)

    # Gaussian kernel
    x_out = np.linspace(np.min(x_in), np.max(x_in), N_out)
    x_in_mesh, x_out_mesh = np.meshgrid(x_in, x_out)
    gauss_kernel = np.exp(-np.square(x_in_mesh - x_out_mesh) / (2 * sigma**2))
    # Normalize kernel, such that the sum is one along axis 1
    normalization = np.tile(np.reshape(np.sum(gauss_kernel, axis=1), (N_out, 1)), (1, N_in))
    gauss_kernel_normalized = gauss_kernel / normalization
    # Perform running average as a linear operation
    y_out = gauss_kernel_normalized @ y_in

    return y_out, x_out

正弦信号加正态分布噪声的一个简单用法: