Computing the Matrix Profile using iSAX
First of all, let us make something clear: we are going to present an approximate method. If you want to calculate the exact Matrix Profile, then you should use an implementation that uses the original algorithm.
The idea behind the used technique is the following: it is more likely that the nearest neighbor of a subsequence is going to be found in the subsequences stored in the same terminal node as the subsequence under examination. Therefore, we do not need to check all the subsequences of the time series, just a small subset of them.
The next subsection discusses and resolves an issue that might come up in our calculations, which is what are we going to do if we cannot find a proper match for a subsequence in a terminal node.
What happens if there is not a valid match?
In this subsection, we are going to clarify the problematic cases of the process. There exist two conditions that might end up in an undesired situation:
-
...