I have a list:
sample_list = array([1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16])
I want to calculate the average of every, say 4 elements. But not 4 elements separately, rather the first 4:
1,2,3,4
followed by:
2,3,4,5
followed by:
3,4,5,6
and so on.
The result will be an array or list of average between every 4 elements in the first list.
Output:
array([2.5, 3.5, 4.5, ...])
My attempt:
sample_list = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]
splits = 4
def avgerage_splits(data):
datasum = 0
count = 0
for num in data:
datasum += num
count += 1
if count == splits:
yield datasum / splits
datasum = count = 0
if count:
yield datasum / count
print(list(average_splits(sample_list)))
[1.5, 3.5, 5.5, 7.5, 9.5, 11.0]
This is not the output I need as this calculates the average of every 4 elements before moving onto a new set of 4 elements. I want to only move one element up in the list and calculate the average of those 4 and so on.
If numpy is an option a simple way to achieve this is to use np.convolve, which can be used to compute a rolling mean when convolving with an array of np.ones:
import numpy as np
sample_list = np.array([1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16], dtype=float)
w = 4
np.convolve(sample_list, np.ones(w), 'valid') / w
Output
array([ 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5, 9.5, 10.5, 11.5, 12.5,
13.5, 14.5])
Details
np.convolve is performing a discrete convolution between the two input arrays. In this case np.ones(w), which will be an array of as many ones as the specified window length (4 in this case) array([1., 1., 1., 1.]) and sample_list.
The following list comprehension aims to replicate the way np.convolve is computing the output values:
w = 4
np.array([sum(ones*sample_list[m:m+w]) for m in range(len(sample_list)-(w-1))]) / w
array([ 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5, 9.5, 10.5, 11.5, 12.5,
13.5, 14.5])
So at each iteration it will take the inner product between the array of ones and the current window of sample_list .
Bellow is an example of how the first outputs are computed so that its a little clearer. Note that in this case the used mode specified for the convolution is valid, which means that, the overlap is specified to be always complete:
[1,1,1,1]
[1,2,3,4,5,6,7,8...]
= (1*1 + 1*2 + 1*3 + 1*4) / 4 = 2.5
And the following as:
[1,1,1,1]
[1,2,3,4,5,6,7,8...]
= (1*2 + 1*3 + 1*4 + 1*5) / 4 = 3.5
And so on, yielding as mentioned earlier a moving average of sample_list.