Top K Motif about matrixprofile-ts

[vanbenschoten]

We should add a routine to do this, but you can zip the matrix profile and matrix profile index arrays together, order by matrix profile value, remove redundant indices together (for instance, if index m and q both have the same motif and hence have a minimum with index d, then d will be represented twice), then select the k values with the lowest matrix profile values.

[vanbenschoten]

Two things worth calling out: since most datasets have a degree of randomness, you'll find that the same motif in two places might actually be slightly different, and hence the above method will treat them as different motifs even though visual inspection will reveal that they're the same. Also, the matrix profile is intentionally agnostic to why a given motif is present in the data, so it won't automatically remove motifs caused by artifacts in your data collection process. Thus, there will always need to be a degree of manual inspection necessary to get the most relevant information.

[whitewalker007]

Thanks for the precise explanation.

[vanbenschoten]

@whitewalker007 If this works well for you, feel free to add a Jupyter notebook tutorial to the repo! We're going to start gathering examples of the code in order to enable others to use it better (with synthetic data, of course).