tsmp is going big!


Since the beginning of the tsmp package, it was evident that a series of algorithms around the Matrix Profile would pop-up sooner or later.

After the creation of the Matrix Profile Foundation (MPF), the tsmp package had doubled the number of monthly downloads, and that is a good thing!

The current version of tsmp, as shown in the previous post had added the new Pan-Matrix Profile and introduced the Matrix Profile API that aims to standardize high-level tools across multiple programming languages.

Now my focus will be on speed and robustness of the package. Currently I can say that the speed improvement is 25% for STAMP and STOMP.

The tsmp package will keep the current structure, but the “core” algorithms will be moved to a new dependency package called matrixprofiler.

This package will be specialized on low-level (Rcpp) implementation of MASS, STAMP, STOMP, etc. This will let the tsmp more flexible implement algorithms that use the Matrix Profile as the base for the application tasks as “motif search”, “snippet search”, or other algorithms that may appear.

matrixprofiler will tend to be less updated and very stable. Those who just want the basics, will get to work with this package alone, but it is safer to install the tsmp and have the former as dependency.

Stay tuned for the first release.