Even if it was for multivariate time series, the model would first need to infer what machine are we talking about, then its working conditions, and only then make a reasonable forecast based on an hypothesis of its dynamics. I don’t know, seems pretty hard.
Comments
"Time series" is such an over-subscribed term. What sorts of time series is this actually useful for?
For instance, will it be able to predict dynamics for a machine with thousands of sensors?
Specifically, its referring to univariate, contiguous point forecasts. Honestly, I'm a little puzzled by the benchmarks.
Even if it was for multivariate time series, the model would first need to infer what machine are we talking about, then its working conditions, and only then make a reasonable forecast based on an hypothesis of its dynamics. I don’t know, seems pretty hard.
Indeed. An issue I ran into over and over while doing research for semiconductor manufacturing.
My complaint was more illustrative than earnest.