Windowing your data

Typically, before applying operations that extract features, we window our time series. For example, if our time series is daily sales information and we have a year of data, we may window our data into months before calculating the mean so we can understand how it changes over time. Likewise, the Fourier transform extracts frequency information that we can apply to windows of our time series to generate cross-sectional data to be used for classification or anomaly detection tasks. When using the Fourier transform for forecasting or curve fitting, which we don’t cover in this chapter, we apply it to the entire series.

In the following figure, boxes represent windows of the signal. We’ll apply our operation separately to these windows and each one will eventually become one cross-sectional data point composed of the extracted features.

Figure 8.2 – Example window on raw audio data

In the preceding figure, you can see...