...
/Window Operations—Weighted, Expanding, and Exponential Moving
Window Operations—Weighted, Expanding, and Exponential Moving
Discover how to work with windowing operations in pandas, with a focus on weighted windows, expanding windows, and exponential moving windows.
We'll cover the following...
Introduction
Having covered the rolling window operation, we’ll now take a look at the other three operations, namely weighted windows, expanding windows, and exponential moving windows. As before, we’ll work with the Amazon 2017 stock price dataset in this lesson.
Preview of Amazon 2017 Stock Price Dataset
Date | Open | High | Low | Close | Volume | Stock |
3/1/2017 | 757.92 | 758.7595 | 747.7 | 753.67 | 3521066 | AMZN |
4/1/2017 | 758.39 | 759.68 | 754.2 | 757.18 | 2510526 | AMZN |
5/1/2017 | 761.55 | 782.3999 | 760.2557 | 780.45 | 5830068 | AMZN |
6/1/2017 | 782.36 | 799.44 | 778.48 | 795.99 | 5986234 | AMZN |
9/1/2017 | 798 | 801.7742 | 791.77 | 796.92 | 3446109 | AMZN |
Weighted windows
We learned earlier about rolling windows, where rolling() can be combined with aggregation functions to compute statistical measures over a number of periods. However, this implementation of rolling windows assigns gives equal weight to all periods.
There are cases where we want to assign different weights to different periods. The good thing is that rolling windows can be easily customized to set weights to different elements in the window. When we say a window is weighted, it means that some values contribute more to the output than the rest.
We can generate weighted windows by using the win_type parameter in rolling(), which supports various types of windows, such as Gaussian, Hamming, Blackman, and triangular.
Note: The types of windows that we can specify must be a string that corresponds to the window functions in the SciPy signal module, which contains a suite of window functions for filtering and spectral estimation. It also means that the SciPy library must be installed in the virtual environment.
For example, we can apply triangular weights to the elements in the window by setting win_type='triang', as shown in the code example below. A triangular window is one with the highest weight in the middle, decreasing linearly toward the edges.
# Apply rolling window calculation to close prices with a triangular windowrolling_triang_mean = df['close'].rolling(window=3, win_type='triang').mean()# View output (first 10 rows)print(rolling_triang_mean.head(10))
In the example above, rolling() applies a moving average with a triangular window of size 3 to the close price of the dataset.
We can also pass supplementary parameters of the SciPy window methods inside the aggregation function. For instance, the example below demonstrates the implementation of a Gaussian (aka normal distribution) weighting scheme. This means that the center of the window will have the highest weight and the weights decrease symmetrically towards the edges. ...