stompi_update - History and streaming

[chris-english]

It is exciting to see the preparation and development of mechanism for streaming data in the Matrix Profile suite. I have no code to contribute at this time but some articles and references to worked solutions in R that might be applicable. Attached, one hopes, is a Keogh article, and the link is to a worked solution of the Keogh PLA in R.

Online_Amnesic_Approximation_of_Streaming_Time_Series_Keogh_etal_ICDM_2004.pdf

Keogh PLA in R, among many others. https://github.com/PetoLau/TSrepr

An application of Laurinec & Lucka to electrical forecasting:

Data Mining and Knowledge Discovery (2019) 33:413–445 https://doi.org/10.1007/s10618-018-0598-2 Laurinec's article: Interpretable multiple data streams clustering with clipped streams representation for the improvement of electricity consumption forecasting Peter Laurinec1 · Mária Lucká1 (Springer)

Multiple streams: https://github.com/PetoLau/petolau.github.io/blob/master/_Rscripts/10post_ClipStream.R, (to get beyond univariate stream of Streams pkg) (from above link) 350 # new data from new day 351 data_new <- data_cons[, (((i+(win-1))*period)+1):((i+win)*period)] 352 353 # sliding window of represention 354 clip_new <- repr_matrix(data_new, func = repr_feaclip) 355 data_clip <- cbind(data_clip[, -(1:ncol(clip_new))], clip_new)

So somewhat of a mess above, but if history is going to be kept, it will probably be an approximation. Hope some of this helps.