How to implement a good moving average in Python

Question

I'm doing some research and I'm implementing a moving average in Python based on this mathematical expression:

Where: n = sample and W1 = Window

I implemented like that:

def movingAverage(signal, window):

   sum = 0
   mAver = []
   k = int((window-1)/2)

   for i in np.arange(k, len(signal)-k):
       for ii in np.arange(i-k, i+k):
           sum = sum + signal[ii]
       #end-for
       mAver.append(sum / window)
       sum = 0
   #end-for

   zeros = [0]*k
   mAver = zeros + mAver + zeros

   return mAver

It work very well. But I'm trying to discover some way to implement the k variant to minimize the signal lost in beginning and in final (Now I'm using an list with zeros).

Can someone help me?

Alexander · Accepted Answer · 2018-09-13 00:30:36Z

2

You could just use Pandas and specify center=True for your moving average.

import numpy as np
import pandas as pd

np.random.seed(0)

s = pd.Series(np.random.randn(7)).round(1)
moving_avg = s.rolling(window=3).mean(center=True)
>>> pd.concat([s, moving_avg.round(2)], axis=1).rename(columns={0: 'signal', 1: 'MA'})
   signal    MA
0     1.8   NaN
1     0.4  1.07  # 1.07 = (1.8 + 0.4 + 1.0) / 3
2     1.0  1.20  # 1.20 = (0.4 + 1.0 + 2.2) / 3
3     2.2  1.70
4     1.9  1.03
5    -1.0  0.63
6     1.0   NaN

edited Sep 13, 2018 at 0:30

answered Sep 12, 2018 at 22:59

Alexander

111k32 gold badges212 silver badges208 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Cesar Abascal Over a year ago

Thanks for your answer, but I'm trying to implement it without using libraries. It can be good to implement in other languages.

fstop_22 · Accepted Answer · 2020-02-05 03:05:44Z

1

You could use a filter with taps of all 1

import scipy as sp
import scipy.signal as sig

h = sp.ones(10)/10
y = sig.lfilter(h, 1, x)

edited Feb 5, 2020 at 3:05

answered Sep 13, 2018 at 12:39

fstop_22

1,0427 silver badges11 bronze badges

4 Comments

Cesar Abascal Over a year ago

Thanks for your answer, but I'm trying to implement it without using libraries. It can be good to implement in other languages.

fstop_22 Over a year ago

It looks like you are already using numpy. Scipy includes numpy with additional packages thrown in for various signal and numeric prcessing functions.

mac13k Over a year ago

If you want to implement the moving average this way, you have to set a to the length of b, so the window sum is divided and the result is the average, otherwise you end up multiplying the input by the window length.

mac13k Over a year ago

I have just made the edit. You can plot the previous version vs. the corrected one to see the difference.

Gideon Kogan · Accepted Answer · 2018-09-13 17:38:27Z

0

At the time I have found the following code:

def moving_average(samples, wind_len=1000):
    wind_len = int(wind_len)
    cumsum_samples = np.cumsum(samples)
    cumsum_diff = cumsum_samples[wind_len:] - cumsum_samples[:-wind_len]
    samples_average = cumsum_diff / float(wind_len)
    return samples_average

def my_cumsum(samples):
    for ind in range(1, len(samples)):
       samples[ind] = samples[ind] + samples[ind - 1]

edited Sep 13, 2018 at 17:38

answered Sep 13, 2018 at 5:16

Gideon Kogan

7635 silver badges20 bronze badges

1 Comment

Cesar Abascal Over a year ago

Thanks for your answer, but I'm trying to implement it without using libraries. It can be good to implement in other languages.

Collectives™ on Stack Overflow

How to implement a good moving average in Python

3 Answers 3

1 Comment

4 Comments

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

1 Comment

4 Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related