Understanding the Matrix Profile
Time series are everywhere, and there are many tasks that we might need to perform on large time series including similarity search, outlier detection, classification, and clustering. Dealing directly with a large time series is very time-consuming and is going to slow down the process. Most of the aforementioned tasks are based on the computation of the nearest neighbor of subsequences using a given sliding window size. This is where the Matrix Profile comes into play because it helps you perform the previous tasks once you have computed them.
We already saw the Matrix Profile in Chapter 1, but in this section, we are going to discuss it in more detail in order to understand better the reason that it is so slow to compute.
Various research papers exist that present and extend the Matrix Profile, including the following:
- Matrix Profile I: All Pairs Similarity Joins for Time Series: A Unifying View That Includes Motifs, Discords and Shapelets...