mvgavg 0.0.2

Moving averages

Why??

Numpy does not include a built-in moving average function as of yet. Most solutions are tedious and complicated. This solution is very nice. This operates similar to the Wolfram Language’s MovingAverage[] function, but has the advantage that it can specify axis (Wolfram gets away with f[#]&/@, but that’s a story for another day).

Usage

See docs for examples. But essentially, the usage is just:

from mvgavg import mvgavg

mvgavg(array, n, axis=0, weights = [list of weights])
mvgavg(array, n, axis=0, binning = bool)

What This Code Does

Consider the array [a,b,c,d, … x, y, z]. Taking the moving average of length n=3 results in a new array wherein the first element is (a+b+c)/3, the second (b+c+d)/3 and so on. This can be done for arrays of arbitrary depth, so if you put in a time series [[t1,x1],[t2,x2],…], a moving average of along axis=0 (the default) will give you back the moving average of all t’s along side all x’s.

The output is a vector that is the same size and shape but has been shortened on the axis axis by a length of n-1.

Axis and Weight Options

Axis: The axis lets you operate at an increased depth, so using the axis=1 parameter, you can operate horizontally across columns with your moving average. You can do this as deep as the array itself.

Weights allow you to customize the function. The default weight (None) is described above (all values are given the same weight). One other weight you can use is ‘pascal’. This gives a nth-line pascal’s triangle weight, so for instance, for n = 5, the ith item in the array is (i + 4j + 6k + 4l + m) / 16, since the 5th line in pascal’s triangle is [1,4,6,4,1]. The other default weight options are triangle (i.e. triangle(5)=[1,2,3,2,1], and quadratic is triangle squared. Or make your own list and put whatever you want in here (make sure len(weights) == n)

Binning

Binning greatly shortens the array and loses some precision. This is desirable if you have an enormous amount of data and don’t need to preserve every point. The difference between a default moving average and a binned moving average is that, for an input array [a b c d e f…] and an output [A B C D E F…] over distance 3, the default moving average looks like this:

A = [a b c]/3
B =   [b c d]/3
C =     [c d e]/3
D =       [d e f]/3
E =         [e f g]/3
....

but if binning = True:

A = [a b c]/3
B =       [d e f]/3
C =             [g h i]/3
.....

As you can see the output array is greatly shortened. As arrays get very large, binning can become orders of magnitude faster to compute, and if you don’t need the resolution of a moving average, are a much more efficient way of handling data, because you may end up throwing most of your data away later when you go to plot it up anyways.

Why Is This Function Important?

Moving averages smooth data and illuminate trends that otherwise may not be as apparent. They also help with reverse interpolation when different x’s yield the same y. The reasons for using moving averages are myriad, so a decent arbitrary-depth moving average function with numpy-speed and arbitrary weighting needed to be written.

Acknowledgements

Credit to @fnjn on github for the sliding window function.

Misc

If you have issues or questions, open an issue on Github at https://github.com/NGeorgescu/python-moving-average or if you think there’s some functionality that you would like to see, or if you have a faster algorithm

Thanks and Enjoy!